The Atari Compendium/The Atari Compendium (Toad Computers) (1994).iso/files/prgtools/gnustuff/tos/gnulib/libsrc98.zoo/Changelo

Text File | 1994-03-29 | 159.0 KB | 4,949 lines

previous change history muddled at best (i can track things in my
RCS files, but i did'nt keep a log).

changes since first  release of library with TOS gcc v1.35
in cronologically increasing order:

ChangeLog:: ++jrb
    New file.

all makefiles:: ++jrb
    adjusted, CMakefile.* now add -fstrength-reduce to CFLAGS

bcopy.s:: ++jrb
    replaces bcopy.c -- a better mouse trap :-)
    bcopy.s has lbcopy() entry point too.
    bcopy.c moved to notused/

bcmp.c, bzero.c:: ++jrb
    added lbXX() entry points
    this makes lbcmp.c & lbzero.c redundant moved to notused/

strchr,strrchr:: ++jrb
    now have entry points for index()/rindex()
    index.c & rindex.c moved to notused/

unlink.c:: ++jrb
    now has entry point for remove()
    remove.c moved to notused/

atof.c:: ++jrb
    redefined HUGE (to be ieee inf)
    HUGE defined in this way in <pmluser.h> (aka <math.h>) too now.

ctime.c:: ++jrb
    cleaned up some ^M's in there

dflonum.c:: ++jrb
    bug fix: defn of XXADDL was wrong (when -O, there were not
    enough constraints to let gcc know that the `zero' pointer
    was being decremented, so that it would re-load (XXADDL is used
    in a loop, this made it worse, since it just assumed it was
    constant and moved the init out of the loop).
    lot of constraints cleanup here and in flonum.h
    redundant defined commented out.


dirent.c:: ++jrb
    moved mydat, olddta into opendir()
    cleanups (i am trying to put explicit casts for mallocs etc
    for the sake of compilers that dont understand protos).

__DEFAULT_BUFSIZ__ :: ++jrb
    added an extern size_t  __DEFAULT_BUFSIZ__ that is normally set
    to BUFSIZ, but you can override it at any point, and get default
    buffers of that size for buffered i/o (a cheap way to have large
    buffers, instead of putting setvbuf()'s all over the place,
    but its ansi non-compatible).
    changes in fxxx.c files refplect this addition.

fputc.c:: ++jrb
    put in support for line buffered streams
    in main.c, stdout is now initialized line buffered
    in stdio.h _IOLBF was not unique, changed that.

getlogin.c:: ++jrb
    logname is now dynamically allocated
    the length is no longer static (used alloca)

getpw.c:: ++jrb
    file always open in text mode should _binmode(1) be in effect

lseek.c:: ++jrb
    now incorporates BSD semantics too, where if you seek past
    eof you get a zero filled hole.

ltoa.c:: ++jrb
    added strrev proto ifdef __STRICT_ANSI__
    (string.h has the ifdef to make it kosher)

main:: ++jrb
    dynamisized _at_exit []
    this gets rid of ATEXIT_MAX (ansi says should be at least 32,
    hopefully we'll meet that and more)
    the return type of atexit() was opposite -- adjusted
    this also required adjustment in tmpfile.c
    stdout is now truely line buffered

malloc.c:: ++jrb
    added
    _malloczero() that set/unsets flag to tell malloc to zero out blocks
        useful with TOS 1.4 where TPA clearing above BSS may be
        turned off. (see new util toglclr.ttp). Some badly
        written programs (gcc-ar for instance) assume malloc
        clears.
    _mallocChunkSize() lets you tune min chunk size.
       CAUTION: use _mallocChunkSize() to tailor to your environment,
            do not make the default too large, as the compiler
            gets screwed on a 1M machine otherwise (stack/heap clash)

    added entry points for malloc/realloc etc

printf.c:: ++jrb
    added support for `#' flag
    added `h' flag

prtfld.c:: ++jrb
    needed <string.h>

qsort.c:: ++jrb
    totally replaced with one from dlib
    the one from dlibs needed quite a few fixes.

read.c, write.c:: ++jrb
    added entry points for read()/write() when not __MSHORT__

sbrk.c:: ++jrb
    added entry point for sbrk()

setjmp.s:: ++jrb
    was bug ridden (thanks pekka!)

strcat.c, strncat.c:: ++jrb
    moved check for null src to outer level

system.c:: ++jrb
    cleanups

tmpfile.c:
    everything dynamacized
    the atexit fn delete_tmpfile() now closes before removing
      this needed new data str.

tmpnam.c:: ++jrb
    replaced sprintf with strcpy/strcat

unx2dos.c:: ++jrb
    made the file name mapping functions more flex (thanks dale for
    suggesting the idea). the mapping functions now go through
    settable (via new function fnmapfunc()) function pointers to
    call the mapping routines. these pointers are by default
    initialized to _unx2dos/_dos2unx that were the old functions.

scanf.c:: ++jrb
    float conversions (%ef) were severly broken. fixed.


vfork.c:: ers
New functions vfork() and wait() added.

mktemp.c:: ers
Fixed ranges and handling of multiple "mktemp"'s; because of a typo, the
old version tried only one number, and if the file existed it gave up.

localtime.c:: ers
Added global variable timezone to hold timezone offset; also function
tzset() to initialize it. (These seem common in Unixes).
Fixed bug in calculation of timezone offset (EST5EDT was being translated
as EST5:05EDT).

system.c:: ers
Added I/O redirection, using vfork().

read.c:: ers
Fixed some bugs due to evaluation of arguments in macro expansions.

stat.c:: ers
Fixed up values returned for st_mode to be consistent with chmod(); also
added a check for executable files, so most regular files get permissions
rw-rw-rw- instead of rwxrwxrwx.
Also added a work around to TOS's refusal to find "." or ".." in an
Fsfirst without wildcards.

rename.c:: ers
Changed so that rename(old, new) fails if "new" already exists. H & S
claims this is implementation defined, and this way seems more intuitive
to me. (If a later draft of the pANS changed this, let me know).

access.c:: ers
Removed the call to _unx2dos (it wasn't necessary, since stat calls _unx2dos).

write.c:: ers
Put def'n of __col_pos here instead of in read.c (to avoid pulling in
read() unless necessary -- write always gets used).

sgtty.c:: ers
New functions stty() and gtty() added for compatibility with old code
(e.g. curses).

varargs.h, stdargs.h:: ers
Added some checks to guard against both of these being included.

osbind.h:: ers
Fixed defn for Mshrink.
Fixed defn of PE_CBASEPAGE.
Moved defn of struct _dta to here from stat.h.

types.h:: ers
Removed #include of <time.h>, added a guarded definition for
time_t.

time.h:: ers
Added a guard to typedef of time_t.

stat.h:: ers
Moved defn of struct _dta to osbind.h, where it belongs. Made a "real"
definition for S_IFREG.


NEW LIBRARY curses.olb:: ers
. Combines BSD curses with Fred Fish's PD termcap
library. Everything seems pretty standard. I made the following
enhancements:

If no termcap file can be found, the Atari ST's built in VT52 is assumed.

A replacement for console_read_byte() is provided that allows binding of
macros to keys with the call:
    console_set_key(int scancode, char *r, char *s, char *a)
where "scancode" is the key scan code, "r" is the string to be sent when
the key is pressed without any shift keys, "s" the string to be sent when
SHIFT+key is pressed, and "a" the string to be sent when ALT+key is
pressed. A NULL string leaves the previous key binding unchanged; an
empty string "" erases any previous key binding.
The new console_read_byte() is slower and takes more memory, so it only
gets linked if -lcurses (or -lcurses16) is used and there is a call to
console_set_key().

New header files <curses.h> and <keycodes.h> are provided. <keycodes.h>
has the prototype for console_set_key(), and has various scan codes
provided as constants (e.g. F_1 is the scan code for the f1 key).

fpprint.c: ++jrb
    replaced decfrexp with a hopefully  better one
    added rounding at prec+1 (output certainly looks nicer now)

scanf.c, printf.c, mktemp.c: ++jrb
    replaced expressions of the form
        n * 10    [n a integer]
    with    TEN_MUL(n) == (((n) << 2) + n) << 1
            10*n == 2(4n + n)

atof.c:: ++jrb
    added ansi func strtod, atof now calls this.
    added d|D as valid chars for 'E' notation as per ansi.
    added proto for strtod to stdlib.h
    corrected bugs in counting total exp.
    now passes its builtin test (which is biased!)
    pulled out old dummy strtod #define from stdlib.h

close.c:: ers
Fixed bug with isatty status not being properly reset unless an error
occured.

unx2dos.c:: ers
Added a check for two slashes in a row (a common error which can really
make GEMDOS confused) and for a directory name with a trailing slash
(Unix apparently accepts access("DIR/",0) as well as access("DIR",0)).

curses.h:: ers
Fixed definition of strlen to agree with definition in <string.h>
(namely size_t strlen(const char *)).

osbind.h:: ers
Changed definitions of Fcreate, Fopen, and Fdup to cast result to long
instead of short. This is because file handles for CON:, AUX:, and PRN:
are word negative but not long negative, whereas errors are long negative;
casting to short makes Fopen("CON:",2) appear to give an error.

getcwd.c:: ers
The GCC is too clever with this; I had to add the keyword 'volatile' to
an array to stop it from optimizing away the test in:
path[2] = '\0'; Dgetpath(path, 0); if (!path[2]) ...

fflush.c:: ers
Changed test for buffer full from (fp->_cnt) to (fp->_cnt > 0), since
when reading fp->_cnt becomes < 0 at EOF, and the subsequent
lseek(file,-fp->_cnt,1) call will (now) cause the file to grow.

main.c:: ++jrb
    if stdout !isatty then make it a full buffered stream rather
    than a line buffered stream;

fclose.c:: ++jrb
    cosmetic changes.

fopen.c:: ++jrb
    freopen() was foobared (see comments there). new coding.

setbuf.c :: ++jrb
    added bezerkly setlinebuf() and a few asserts.

include/unixlib.h:: ++jrb
    added proto for setlinebuf()

include/assert.h:: ++jrb
    changed assert #def'n for !__STDC__ compilers that don't
    do token pasting.

fpprint.c:: ++jrb
    brought back the old version(sigh!)
    added rounding at prec
    This has to be redone. Due to prec. loss for numbers near 1 or near 10
    it prints stuff like "0.9:00e0" etc.
    now that modf() etc work almost exactly, we can re-code using those.

Special thanks to peter housel who posted his x86 based floating
lib. many of the alg used in the now improved floating support
are fashioned after peters alg.

norm.c:: ++jrb
    new file, normalize a ieee double

atof.c:: ++jrb
    totally re-worked, now its much more like i wanted it to be.
    better test

ldexp.c, modf.c, frexp.c:: ++jrb
    totally reworked. ldexp now always takes an int as it should have,
    so sldexp is no longer need for shortlib. (see #ifdefs in ldexp.c)
    better tests (see also pml/envtest)
    the combined test for ldexp/frexp is in frexp.c

sldexp.c:: ++jrb
    moved to notused/sdlexp.c

*akefile.16:: ++jrb
    took out sldexp.o target from all of them

*akefile*:: ++jrb
    added target norm.o

doprnt.c, sprintf.c, fprintf.c:: ++jrb
    Finally, adopted berkley's PD doprnt(). It does things right
    and much better than kludges i was trying to get especially
    the fpprint right (i still cannot come up with a alg for
    decrexp that does not loose badly on prec). appropriate
    adjustments made to sprintf.c and fprintf.c.

fpprint.c, printf.c, prtfld.c:: ++jrb
    moved to notused/ because of above.

*akefile:: ++jrb
    adjusted to above two changes.

varargs.h, stdarg.h:: ++jrb
    fixed bugs when va_arg is expanded with type 'char'
    thanks to dale schumacher for detecting this.

*makefile.*:: ers
    added abs.o (how did we miss this one???)
    fixed typo in makefile.16

qsort.c:: ers
    added cast to an argument (needed for -mshort)

vfork.c:: ers
    changed so that child process uses the same stack as the parent; this
    is what Un*x does, and saves memory.
    fixed wait() to be Un*x compatible.

spawnve.c:: ers
    made storage for environment+args dynamic; gnu make was losing on
    big files (e.g. nethack) because 2K wasn't enough.

system.c:: ers
    changed call to wait to agree with changes above.

findfile.c:: ers
    fixed search for extensions (.ttp, .prg, etc) so that it works with
    _unixmode > 2.

doprnt.c,scanf.c:: ers
    changed __NO_FLOATS__ to __NO_FLOAT__ to agree with makefiles.

qsort.c:: ++jrb
    merged in minix changes (i have to support both!)
    as per ers change above, adjust arg for _wqsort()and _lqsort() too.
    additional assumption: if SHORT alignment is desired then
    it is ok to align a long at a short boundary too. this seems
    to be generally true (of course someone will come up with a
    weird machine where it is untrue - maybe intel!).
    To do : define a threshold and do insertion sort below threshold
        remove recursion

qsort.c :: ++jrb
    Finally a qsort that kicks ass! (and works :-).
    the old one moved to notused/dl_qsort.c
    qsort.c: i started with doug smidth's version,
    first hacked it to handle SIZE sized items (his was ints only)
    the performance sucked, so i totally rewamped the algorithm in
    qsort(), following Sedgewick and suggestions in K V3.

dflonum.c:: ++jrb
    House cleaning.

fixnum.s::  ++jrb
    Merged in kai-uwe's code (thanks kai).
    Fixed a few bugs. Cleanup.

popen.c:: ++jrb
    Added kau-uwe's popen/pclose with appropriate mods for this library.
    The basic difference was that this libraries system() already did
    re-direction, so we take advantage of that fact to avoid doing
    re-direction in popen/pclose. Thanks again kai-uwe.

stdio.h:: ++jrb
    added protos for popen()/pclose() (in non strict ansi part).

*akefile:: ++jrb
    reflect the addition of popen.c

dirent.c:: ers
    fixed opendir() to work correctly with non-existent directories.
    Also made seekdir() actually do some work.

sys/dir.h:: ers
    took advantage of undocumented feature of dirent.c, namely that
    d->reclen contains the length of the name.

fixnum.s:: ++jrb
    kai-uwe's fixnum has problem. pulled it out and re-instated the
    old one. will look into whats going on.

fixnum.s:: ++jrb
    fixed bugs in kai's version and put it back (it is much more
    efficient).

fcntl.h, close.c, open.c, isatty.c, fhandle.c:: ++jrb
    replaced char __handle_stat[] with a open files status structure
    __open_stat[], that keeps more info around about open files,
    enabling impl of fstat(). later more fields can be added as required.
    Also Note: while making these changes, took care of the fact that
    the smallest valid gemdos handle is -3 and not 0. See
    __SMALLEST_VALID_HANDLE and __OPEN_INDEX macros in <fcntl.h>

stat.c, stat.h :: ++jrb
    added fstat(int, struct stat *)

scanf.c :: ++jrb
    floating conversion uses our super atof now after collecting the
    input number as a string.

popen.c:: ers
    fixed up popen(cmd, "r") calls to redirect properly; also they
    return NULL if cmd fails (return code < 0).

vfork.c:: ers
    changed to reflect the fact that exit codes are really short on
    the ST.

gnulib2.c:: ++jrb
    new file. contains long long support for gcc, and a few funcs
    that moved over from dflonum.

*akefile*:: ++jrb
    added gnulib2 targets

close.c:: ers
    checked that __open_stat.filename was non-null before freeing

malloc.c:: ers
    allowed free(0) to do nothing (I think ANSI wants this)

open.c:: ers
    added _canonical() function to get full pathname of a file, so
    fstat() can work after a chdir().

textio.c:: ers
    new functions to read/write text files doing nl conversion
    (this seems to crop up a lot, especially in GNU stuff)

localtime.c:: ers
    fixed so that days have 24 hours, not 23 :-)

strdup.c:: ers
    new function strdup() added (GNU and system V both use it)

unlink.c:: ers
    changed so unlink of an open file works (well, usually)

dup.c:: ers
    changed dup2 so that it closes the original file, and returns
    the file descriptor on success

unlink.c:: ers
    deleted debugging print.

traps.c:: ++jrb
    made changes corresponding to the osbind.h changes (stricter
    constraints so that ea is evaluated into reggies before we
    change the stack pointer from under gcc's nose).

*akefile.32:: ++jrb
    removed spurious trap-X.o targets (they should only be in *akefile.16)

crt0.s:: ers
    removed unneeded declaration of _main (crt0 calls __main)

ftw.c:: ers
    new function (from System V) to do file tree walks

minimal.h:: ers
    header file for programs that don't use stdio or any other "high
    level" stuff (basically programs that do all OS stuff directly
    through osbind.h). This results in a substantially smaller
    executable.

linea.h:: ers
    a couple of cleanups of type mismatches. TODO: make linea0()
    a statement expression a la osbind.h (after all, it returns a
    useful value).

dflonum.c:: ++jrb
    fix as per P.Deutch net article.

spawnve.c:: ers
    made the handling of extended arguments compatible with the
    new GEMDOS standard.

linea.h:: ++jrb
    replaced it with a new more complete version. supports inline
    linea bindings (inline is default unless __NO_INLINE__ if #def'ed)

linea.c:: ++jrb
    support file for functional form of linea bindings.

crt0.s, crtinit.c:: ers
    replaces crt0.s. NOTE: compatible with old MWC style startup and
    the new atari standard, except it does not check for
    command tail length == 127 for compatibility with older stuff
    and until shells are revised.

crt0.c:: ++jrb
    combined functionality of crt0.s/crtinit.c into one file.

*akefile:: ++jrb
    adjustments for crt0.c and line.c

div.cpp:: ++jrb
    new file, implement ldiv() and div()
    note: these may not be used if -fpcc-struct-return

div.c:: ++jrb
    a more portable ldiv()/div() compat with -fpcc-struct-return too.
    not used!

*akefile:: ++jrb
    added targets for div.o

fgetc.c:: ++jrb
    flush stdout before reading from stdin if stdin isatty and stdout is
    line buffered (more berklix cloning).

fseek.c:: ++jrb
    bux fix: <unixlib.h> was not #include'ed, as a result all the
    lseek() calls were messing up on large files. Thanks to Bill Shroka
    for finding this. (and thanks to him for giving the library the benfit
    of doubt!)

findfile.c:: ers
    added sanity check

vfork.c:: ers
    added Mfree of environment. Also changed Makefiles to reflect
    necessity of a frame pointer

system.c:: ers
    replaced calls to open() and close() with calls to Fopen() and
    Fclose(), to avoid problems with child process malloc'ing and
    free'ing

*makefile.*:: ers
    used GNU Make's include feature to seperate common 16 and 32
    bit stuff

unx2dos, support.h:: ++jrb
    adjusted types of first arg to mapping functions. should be
    `const char *'

strchr, strpbrk, strrchr, strstr:: ++jrb
    cast return type to (char *)

memchr:: ++jrb
    cast return type to (void *)

getopt,strtol:: ++jrb
    cast assignments

getpw.c:: ++jrb
    forward decl of nextent()

strtol.c:: ++jrb
    replacement function, adapted from norbert schelenkar's stdio
    this detects overflow as req by ansi

strtoul.c:: ++jrb
    replacement function, adapted from norbert schelenkar's stdio
    this detects overflow as req by ansi. new file.

doprnt.c:: ++jrb
    eff hacks. (as far as i can tell i did'nt break it!)
    calls fputc() directly
    keith bostics type cleanup inc.

fprintf.c, sprintf.c:: ++jrb
    doprint() does'nt need put fn arg any more.
    sprintf    - fake a file pointer so that doprint can call fputc()
    new flag _IOSTRING in stdio.h to support this.

filbuf.c:: ++jrb
    support function for getc(). getc() is now half way efficient
    (the old dlibs fgetc sucked). getc now a macro.
    new file

fgetc.c:: ++jrb
    functional form of getc.

fungetc.c:: ++jrb
    new file, broken out of old fgetc.c

getbuf.c:: ++jrb
    new file, broken out of old fgetc.c

fopen.c, fdopen.c:: ++jrb
    now call _getbuf() when file is opened rather than at first
    read/wr.

main.c:: ++jrb
    call _getbuf() for std streams.

fputc.c::
    dont call _getbuf() any more

getw, gets.c, fgets.c:: ++jrb
    use getc() instead of fgetc()

stdio.h: ++jrb
    getc() now a macros
    new flag _IOLOOKLF set when looking for lf across calls to
    _filbuf(). doing all of the above does seems to help
    performance.
    new flag _IOSTRING to support string FILE's for sprintf and
    friends.

fread.c:: ++jrb
    dont call fflush(), instead use any buffered data.

*makefiles:: ++jrb
    adjusted

filbuf.c:: ++jrb
    improved translated processing. the copydown idea did'nt turn
    out to be so hot.

crt0.c:: ++jrb
    added __DEFAULT_BUFSIZ__ processing here (ala _stacksize). Now
    __DEFAULT_BUFSIZ__ gets added to _stksize to try and make
    library initialed alloca's a little bit safer. Note useage
    of __DEFAULT_BUFSIZ__ has changed, now its more like
    _stacksize where you declare it as a initialized global
    ie: size_t __DEFAULT_BUFSIZ__ = <XXXX>L; (outside the scope of
    any function). Also note: if BUFSIZ is changed in stdio.h then
    that change must be reflected in crt0.c too. for obvious
    reasons we dont #include <stdio.h> in crt0.c.

getbuf.c:: ++jrb
    deleted initial __DEFAULT_BUFSIZ__ processing from here, and
    made it extern.

random.c:: ++jrb
    added for bsd compat. will try and work in the state stuff. for
    now it is just a copy of rand() with types adjusted.

gnulib2.c:: ++jrb
    fixed __floatdidf as per net article.

filbuf.c:: ++jrb
    type adjustment for var `f'

fread.c:: ++jrb
    - added missing _IORW at top
    - if after consuming present buffer, the amount left to read
    is < fp->_bsiz, then read fp->_bsiz into buffer and give it
    bytes from there, instead of reading into user buffer. this
    helps a *lot* with large buffers, and is not too shabby with
    ordinary ones.

fseek.c:: ++jrb
    rewrite: fseek() ,ftell() : give them some semblence of
    efficiency, rather than flush on every instance. some of
    the old dLibs stuff is really lacking in efficient use of
    the buffered data.

fopen.c:: ++jrb
    moved _getbuf() out of _fopen() into fopen(). This simplifies
    freopen() and does'nt cause _fopen() called from freopen()
    to allocate a buffer.

setbuf.c, setvbuf.c:: ++jrb
    if _IOMYBUF then free buffer first, since we _getbuf() on open now.
    if new discipline is _IONBF or new buffer NULL then use tiny
    buffer fp->ch (this was a bug in the old version)

getbuf.c:: ++jrb
    if line buffered, make _bsiz min(__DEFAULT__BUFSIZ__, BUFSIZ).

main.c:: ++jrb
    move test for stdio re-directed  to before _getbuf(), so that
    it does the right thing.

strtol.c:: ers
    strtol() (and hence atoi()) weren't working at all, because
    (short)LONG_MAX is a negative number.

doprnt.c:: ++jrb
    in doprnt, ICONV compare number against 65535L instead of
    USHRT_MAX. The problem was that when !mshort, this generated
    a cmpl #-1,NUMBER as (unsigned short)~0 was promoted before
    the compare. TODO: fix limit.h so that such things dont
    kill you.

fclose.c,fflush.c,fopen.c,fread.c,fwrite.c::
fdopen.c,filbuf.c,fputc.c,fseek.c, main.c :: ++jrb
    the inplementation of update mode (w+, r+, r+w) was totally
    screwed up. it worked till now by sheer luck, because the
    buffers were being flushed and synced at evey fseek, ftell,
    fwrite etc. Chnaged the menaing of _IORW to mean that a file
    is open for update, and keep track of whether it was last
    read or written by setting/resetting _IOREAD and _IOWRT as
    appro (or clearing them both on a sync). This similar to 4.3
    but not exactly the same.

fwrite.c:: ++jrb
    make better use of buffer, as per other changes.

crt0.c:: ers
    changed all "ints" into longs so that it compiles the same with
    or without -mshort. moved definition of errno to main.c for the
    same reason

stdio.h, fgetc.c, filbuf.c:: ers
    ftell/fseek on text files were broken; when we stripped all the
    carriage returns out of the input buffer, ftell had no way to
    tell the real binary position. changed getc() to an inline
    function which, like fgetc(), strips out the CR's as we go. this
    simplifies filbuf a lot. also: no need to check for _IODEV, since
    read() and write() already translate '\r' to '\n' where appropriate.

utime.c:: ers
    added sanity checks

crt0.c:: ++jrb
    fixed so that including the insufficient memory test does'nt
    cause the initial gemdos stack to blow up on a 1040. initially set
    the stack at hitpa, and then later set it to its final value.
    a dummy routine introduced to set up sp, which in turn calls
    the real startup routine (that was pushing too many registers
    on the gemdos stack causing it to blow up).

crt0.c:: ++jrb
    The (new) meaning of _stksize: (thanks to allan pratt for the feedback)

    _stksize            meaning
      -1L        keep all of memory (except 8k at top) and do
            mallocs from own heap, with heap grown upwards towards
            stack, and the stack growing down towards heap,
            with a minimum slush between them so that they
            dont meet (only checked while malloc'ing). With
            this model, further spawning is not possible, but it is
            well suited for programs such as gcc-cc1 etc.
            Thanks to Piet van Oostrum & Atze Dijkstra for this idea

    0L        keep minimum amount of memory. this is also the
            case when _stksize is undefined by the user.
    1L        keep 1/4 of memory, free 3/4
    2L        keep 2/4 (1/2), free rest
    3L        keep 3/4, free 1/4
    other        keep that many bytes
    -other        keep all BUT that many bytes

malloc.c:: ++jrb
    added code for allocating from own heap when _stksize == -1L

main.c:: ++jrb
    as proposed by allan pratt, make gemdos handle 2 the file for
    stderr. then if isatty(2) Fforce(2, -1). this allows for
    re-direction of stderr (by shells that can) to a file. if
    stderr is not re-directed  it is Fforce'd to gemdos handle -1
    (console). if it is re-directed, stderr becomes a full
    buffered stream.

sflonum.s:: ++jrb
    replacement __cmpsf2 courtesy of Piet van Oostrum.

unlink.c:: ers
    was missing proto for malloc.

crt0.c:: ++jrb
    sp was being set incorrectly to (bp->p_tbase + m) instead
    of (bp + m). (m already includes sizeof(BASEPAGE)

gnulib2.c:: ++jrb
    added missing proto for __fixunsdfdi()

stdio.h:: ers
    added proto for _filbuf(), needed for getc().

string.h:: ers
    provide inline definitions for strcat, strcpy, and strlen
    (these can be undone by #undef strcat, etc.)

malloc.c, sbrk.c:: ers
    move HeapAlloc from malloc.c to sbrk.c, and merge its functionality
    into lsbrk()
    provide dynamic readjustment of MINHUNK, so that a program that makes
    lots of little requests is less likely to run into the 20 malloc bug

strchr.c, strcmp.c, strcpy.c, strlen.c, strncat.c, strncmp.c, strrchr.c:: ers
    adjustments to reduce code size and improve execution speed

fopen.c:: ers
    allow for open() to return TOS's (word) negative special file
    handles.
    make TTY's line buffered instead of unbuffered, so that the
     editing functions in read.c can work. (this  mod is not
    complete: it is in preparation for upcoming dev driver
    interface for character devices  ++jrb)

sbrk.c:: ++jrb
    the asm for sbrk was in the wrong place.

    ( (thanks edgar)
    a general warning about the asm dirty trick for multiple
    entry points: if the routine has strings you cannot use it
    as the compiler compiles strings into text area, and these
    are put before the function entry point, so the asm() would
    fall into the strings. suggested work around:
      asm(".text; .even; .globl _fooalias,_foo; _fooalias: jra _foo");
    seems to be the safest way of doing this. (warning: if there
    is no intervening text between _fooalias: and _foo: then
    you'll get a 0 offset error from gas. if you use sun /bin/as it just
    generates a bad opcode! when this is the case our old trick
    is the one to use.)
    )

main.c:: ers
    changed so stderr doesn't get set redirected if we're called from
    a sensible shell (i.e. one that understands the standard ARGV
    conventions)

assert.h:: ++jrb
    slightly improved (per gcc v1.37)

fprintf.c:: ++jrb
    adjusted __eprintf for improved assert.h

gnulib2.c:: ++jrb
    bug fixes (per gcc v 1.37)

limits.h:: ++jrb
    replaced: with output of config

float.h:: ++jrb
    replaced: with output of config (still lies a little bit! the
    output is from running `config -f' on a Sun)
    (we really need to fix up some of the float stuff -- where are
     you Kai-Uwe)

Acknowledgements: most of the fixes below are from Edgar Roeder from
diff he sent me for emacs with this lib. Thank you very much!! we will
be incorporating more of his ideas as the pluggable char devices
interface takes shape.

atof.c:: ++jrb
    dont clear errno (edgar)
    recognizes trailing F or L

qsort.c:: ++jrb
    fixed signed/unsigned compare problems (i cant believe i made
    this mistake)


errno.h, perrror etc etc:: ++jrb(edgar)
    made all errno's positive. so much more convenient! why did we
    have this brain damage in the first place?

grep errno *.c:: ++jrb(edgar)
    return positive errno's

fcntl.h, ioctl.h,keycodes.h,setjmp.h,signal.h,support.h,sysvars.h
tchars.h,times.h:: ++jrb(edgar)
    adjustments for emacs (some of the stuff in there is not fully
    implemented , but will be with the upcoming device interface).
    Thanks to edgar for all the emacs changes.

*: ++jrb
    got rid of RCS headers wherever i found them. they just cause
    lots of unnecessary diffs.

crt0.c:: ++jrb (thanks edgar)
    - acc support
    - dump friendly
    New acc convention:
    crt0 now detects acc. if a program wants to run as an acc (or
    both as an acc or app), then it must
        - set _heapbase to bottom of heap+base
        - set _stksize to size of heap+base
        - at startup: sp set to the top and malloc() will
          happen from heap.
        (note: malloc() *not* Malloc())

    New dump convention:
    undumpable programs set _initial_stack to desired value.
    _stksize is set from this at each startup, and malloc()
    happens from own heap. (which implies
    _initial_stack == heap+stack size)

crt0.c:: ++jrb
    - bug fix: the from area of old style cmdline was not being
    terminated properly. (showed up when fast bit is set)
    - fix up env to new style when we get a old style env.
    (bill -- tell me how lharc does from arcshell now)

sysvars.h:: ++jrb
    New convention:
    all sysvars defined as a address pointer of the correct type.
    to access their value, you just have to dereference in
    super mode.

sysvar.c:: ++jrb
    Long get and set a sysvar. Note the use of volatile. guess
    what happens if you dont use it!

all makefiles:: ++jrb
    adjusted for new targets

abort.c:: ++jrb
    for now raise SIGIOT. once the signal stuff is complete we'll
    raise SIGABRT as it should.

access.c:: ++jrb
    check owner modes, now that we do *real* modes. see stat.c

clock.c:: ++jrb
    returns # of ticks since start of process. new variables
    _start_time initialized in _main(). _child_time and _sys_time
    are maintained by spawnve.c (_sys_time is 0 for now always)
    this info is stuffed into the tms buffer by times().

dirent.c:: ++jrb
    opendir(): run name through unx2dos() before Fsfirst'ing.
    ino's are initialized from a up-counter (note: see stat too)
    we should probably be looking at a way to set them to their
    fat #.
    (note: d_reclen is not posix, its bsd'ism)

dup.c:: ++jrb
    set __open_stat on dup too.

fflush.c: ++jrb
    copy fp->_cnt into local before writing/clearing in case the write is
    interrupted.

fhandle.c:: ++jrb
    adjustments for new __open_stat struct

fopen.c:: ++jrb
    adjust open mode

main.c:: ++jrb
    avoid all static initializations.
    initialize clock variables
    make stderr rw for brain dead people
    exit with return value of main as be ansi C

malloc.c: ++jrb
    exchanged entry point and their aliases.
    merged in edgars changes to shrink heap if free'd block is at
    the top of the heap.

open.c:: ++jrb
    better mode handling.
    set additional fcntl flags in __open_stat

sbrk.c:: ++jrb
    return -1 on error
    if doing heapallocs and  n == 0 return _heapbase

scanf.c: ++jrb
    handles short values correctly
    add %n

setvbuf.c:: ++jrb
    if bp is null, alloc it

signal.c:: ++jrb
    _init_signal() : initializes sig_array instead of static
    initialization. (it'll do more soon). called from crt0.

sleep.c:: ++jrb
    sleep, usleep use clock()

spawnve.c:: ++jrb
    keep track of child_runtime

stat.c:: ++jrb
    *real* mode bits

tchars.c:: ++jrb
    added (yet unsupported) tchars

times.c:: ++jrb
    stuff _chaild_runtime etc from variables and clock()

osbind.h:: ++jrb
    another major re-hack:
    -- the basic reason: there was just no reliable
    way to get the definitions (inline or not does'nt matter) to
    fully evaluate the args before we changed the sp from under it.
    (if -fomit-frame-pointer is *not* used, then most of the time
     we dont need to do this, as things will just reference off of
     a6, but this is not true all of the time).
    my solution was to use local vars in the body of the statement
    exprs, and initialize them from the args of the statement expr block.
    to force the evaluation of the args before we change sp from
    under gcc's feet, we make the local vars volatile. we use a
    slight code optimization heuristic: if there are more than 4
    args, only then we make the local volatile, and relax
    the "r" constraint to "g". otherwise, we dont put the volatile
    and force the evaluation by putting the "r" constaint. this
    produces better code in most sitiations (when !__NO_INLINE__
    especially), as either the args are already in a register or
    there is good chance they will soon be reused, and in that
    case it will already be in a register.
    it may (the local vars, especially when no volatile)
    look like overhead, but in 99% of the situations gcc will just
    optimize that assignment right out. besides, this makes
    these defines totally safe (from re-evaluation of the macro args).

    -- as suggested by andreas schwab (thanks!)
     ([e-mail]) all the retvalues are now
     local register vals (see the extentions section in the info file)
     this really worked out great as all the silly "movl d0,%0" at
     the end of each def can now be removed, and the value of
     retvalue ends up in the correct register. it avoids all the
     silly "mov d0,[d|a]n" type sequences from being generated. a real win.
     (note in the outputs "=r"(retvalue) still has to be specified,
     otherwise in certain situations you end up loosing the return
     value in d0, as gcc sees no output, and correctly assumes that the
     asm returns no value).

    -- all the n's (the function #'s for the traps) are now given
    the more relaxed "g". This again results in better code, as
    it is always a constant, and gcc turns the movw %1,sp@- into
    a movw #n,sp@-. we could have given them a "i" constraint too,
    but "g" gives gcc more breathing room, and it does the right
    thing. note: the n's still need to have "r" constraints in the
    non-inline form (function form), as they are no longer constants
    in the function, but a normal arg on the stack frame, and hence
    we need to force evaluation before we change sp. (see osbind.c)

    -- straps.cpp and traps.c are history. we dont need no stinking
    non-reentrant bindings (straps) or incorrect ones (traps.c :-)

osbind.c:: ++jrb
    new file. contains non-inline bindings for both 16 or 32 bit
    lib. straps.cpp and traps.c are history.

*akefile:: ++jrb
    adjustments for above

string.h:: ++jrb
    inline __strcat was'nt returning dest.

isatty.c:: ++jrb
    check return value of Fseek against -EBADARG since errnos are +ve
    Fseek with SEEK_SET instead of SEEK_END
    (btw: atari's example in the T1.4 supplementary docs for
    isatty() is wrong)

rename:: ++jrb
    typo, (should be if(rval < 0) errno = -rval)
    if newname exists, delete it first (this is BSD'ism, buts thats
    ok because my handy ansi draft says "if a file new exists prior
    to call, the behavior is impl. defined"

crt0.c:: ++jrb
    __data_start removed as it can be determined from basepage.
    return from acc now calls _exit() directly not exit().
    its just too bad if an acc falls out like this. not much we
    can do.

times.h, time.h:: ++jrb
    time_t guarded by _TIME_T rather than __TIME_T_DEFINED to
    be consistent with size_t beging guarded by _SIZE_T

param.h:: ++jrb
    added definition of HZ. users may want to adjust
    the definition for local standards.

times.c:: ++jrb
    use definition of HZ from param.h

sleep.c:: ++jrb
    units for usleep were incorrect. HZ was a misnomer there.

memccpy.c, memcpy.c:: ++jrb
    size_t is unsigned, so <= 0 type comparisons dont make sense
    (and limit range)

minimal.h:: ++jrb
    _exit() now in crt0

gmon.c:: ++jrb
    true profiling comes to tos-gcc. profil(2), monitor(3)
    and friends. gprof distributed in a seperate dir.

crt0.c:: ++jrb
    added #ifdef for making gcrt0.o for profiling. the front
    end gcc.ttp, will automatically link in gcrt0.o for the
    startfile, when `-pg' option is used. NOTE: unlike under
    unixen, cc -pg will not want profiled libraries. if you
    want to link in profiled libraries, you have to explicitly
    link them in.

*akefile:: ++jrb
    adjust for above

qsort.c:: ++jrb
    turns out we were over-collapsing the walls! check left and right
    pts dont cross hi and lo resp.


rename.c:: ers
    fix so return value is always -1 on error

fork.c,vfork.c:: ers
    completely new way of doing vfork(), combined with a way to emulate
    fork(). of course, the parent has to wait until the child is
    done, but they both get copies of the same data space.

exec.c:: ers
    new file; emulate execve and execv (execl* and exec*p to follow).

spawnve.c:: ers
    changed to implement P_OVERLAY (this doesn't really work yet; all
    it does is _exit(spawnve(P_WAIT, ...))).

*makefile*:: ers
    fixed iio.olb (before, the files that were in it weren't usually
    linked, since they weren't needed yet; added fprintf.o, sprintf.o,
    etc.)

memory.h, stdlib.h:: ++jrb
    proto for alloca only if alloca() is !defined. typically we
    #define alloca __builtin_alloca.

dflonum.c, sflonum.s, fixnum.s:: ++jrb
    obsolete: please mv these files to subdir notused/

all float/double/long support routines:: ++jrb
    thanks to Kai-Uwe Bloem ([e-mail]) for the complete
    replacements. these routines really do work well. to see the
    difference try pml/pmltests/* before and after. thanks kai-uwe!
    (see README.FP and the _*.s files for more details).

*akefile*:: ++jrb
    adjusted for above

gmon.c::
    use etv_timer as it saves reggie and does the clock/4 etc,
    less overhead. the old code (that used timer C intr) is still there:
    to activate it -D_USE_TIMER_C_.
    when un-installing, dont check for GNUC signature, just _XBRA_MAGIC
    any handlers found  before myself that dont use XBRA are nuked as
    they may end up calling the handler we are de-installing. If
    they happen to have another XBRA friendly handler as their next,
    which is not the one we are de-installing, its tough luck, as
    we cant find their next, we just install our next into the exception
    vector.
    Thanks to edgar for the suggestions.

_normdf.s _normsf.s:: kub
 export ___normdf entry to C language. Rename the internal entry to a name
 not accessible from C to prevent crashes

_addsubdf3.s _addsubsf3.s _divdf3.s _divsf3.s _extendsfdf2.s
_floatsidf.s _muldf3.s _mulsf3.s _truncdfsf2.s:: kub
 adjust for the re-named entry points in _normdf.s and _normsf.s

_divmodsi3.s _udivmodsi3.s:: kub
 small bug fix: numbers having bits 30 or 31 set would eventually fail.

_normdf.s. *akefile:: ++jrb
    change interface of __normdf to expects ints instead of shorts
for exp, sign, and rbits. easier to interface this to !-mshort code
this way. change name of _normdf.s to _normdf.cpp, and preprocess
it in the makefile for #ifdef __MSHORT__.

atof.c, frexp.c, modf.c:: ++jrb
    use the new __normdf() entry point instead of norm.

norm.c: ++jrb
    redundant and obsolete. mv to notused/

mincl.32, mincl.16, gmakefile.16, makefile.16:: ++jrb
    make was attempting to make $(GLIB2) over and again. added
dummy target gl216done in turn depends on gnulib2.c. it is touch'ed
once GLIB2 has been made. (which means you must have or write
touch.ttp). in *akefile.16 added CLFLAGS, the flags required for
things that will be compiled with 32 bit intergers in the -mshort lib.

norm.c, _normdf.cpp, atof.c, frexp.c, modf.c:: ++jrb
    resurrect norm.c from not used (it refuses to die!).
__normdf() is doing something strange with the C entry point, and loosing
the sign bit. (the asm entry seems to be fine). in atof/frexp/modf
#define __normdf norm and bring norm.c back.

fwrite.c:: ++jrb
    after fflush()'ing when space goes to zero (but we are'nt finished
necessarily with the write request), if _IORW set _IOWRT flag, otherwise
we loose what we write in the buffer subsequently if read or seek happens
before the next flush. zoo showed up this bug. i can hardly belive it survived
this long! ARRRRRRRRGGGGGGGGGGGGGGGHHHHHHHHHHHHHHH!! (i feel much better now)

ldexp.cpp, frexp.cpp, modf.c:: ++jrb
    replace the .c files of the same name.

*makefile:: ++jrb
    adjust for above

crt0.c:: edgar
    detect prog/acc by testing sp instead of a0 for the benfit of
pack/unpack and friends that dont clear a0 when launching an app.

ctype.c:: edgar
    added two location before _ctype[] to catch isXXX(EOF) tests.

stat.h:: edgar
    make st_mode unsigned short type (some programs use st_mode as
an index into an array, negative indicies make them blow up).

atof.c:: ++jrb
    cleared up my confusion with __normdf(). i was forgetting
about denormalized numbers, and needed to call ldexp() instead of stuffing
the binary exponent directly into the result at the bottom of __adjust().
norm.c can now be safely mv'ed to unused.

*akefile:: ++jrb
    adjusted for above. in mincl.* took out the dummy glib*done targets
they we not required. the problem bill was seeing was due to the incorrect
def for GLIB2 in mincl.16. it should just be "gnulib2.o".

stddef.h:: ers
    changed definition of EXIT_FAILURE from (-1) to (1) to agree with
    Unix

sys/dir.h, dirent.h:: ers
    merged into one file, and updated to allow caching of directory
    information for stat()

stat.c:: ers
    stat() now communicates with the dirent functions, so that a "stat"
    on a file in an open directory doesn't have to go back to disk
    (opendir() gets all the information stat() needs for free, so we
    might as well save it).

grp.c, grp.h:: ers
    primitive group id/name functions now provided

stat.h:: ers
    include <types.h> if it hasn't already been included

string.h:: ers
    provide inline version of strdup(), for benefit of the library

perror.c:: ers
    all the error messages were in the wrong order now. oops.

lots of places:: ers
    added new code to handle symbolic links. whether or not to use
    them is up to the user, via the environment variable UNIXMODE
    (which also controls other forms of directory mapping, too).
    still to do: code to prevent conflicts for long filenames,
    and proper documentation for UNIXMODE.

chmod.c:: ers
    Changed to silently ignore attempts to chmod() a directory (TOS
    doesn't give any real permissions to directories anyways).
    umask() moved to open.c

console.c, signal.c:: ers
    Only steal keypresses if we really have to check for signals.
    This removes aggravating inability to type ahead while looking
    at output from utilities.

crt0.c:: ers
    remove all occurences of "int" keyword, add comment to remind us
    not to use ints, since crt0.o should be the same for 16 and 32
    bits.

dirent.c, symlink.c, stat.c:: ers
    improve "shadowing" code so that aliases for long filenames work
    exactly like the original files, and the originals don't appear
    in directories if _lAUTO is set. Also allow the ".dir" file to
    be hidden.

mincl.16:: ers
    GLIB2 should just be gnulib2.o, I think (as in mincl.32). Perhaps
    that's what was causing the wierd loops.

open.c:: ers
    added a more useful umask().

spawnve.c:: ers
    oops, was clobbering "mode" when its value would be needed later.
    exec() should work now!

symdir.h, unx2dos.c:: ers
    cleaned up flags, removed unused ones. Added "H"ide (hides the
    .dir file) and "r"oot (used to specify a default root drive;
    e.g. "rC" causes "/bin/sh" to translate into "C:\BIN\SH" no
    matter which drive is current. Useful if the "r" drive has
    symlinks like "usr->D:", so "/usr/bin" is "D:\BIN".

    Fix up compatibility code to be really compatible (the _unixmode
    variable never gets set until runtime).

symlink.c:: ers
    add a (fake) link() call. clean up lstat() to take better
    advantage of stat's intelligence.

stat.h:: ers
    add definitions for major(), minor(), and S_IFBLK

string.h:: ers
    nuke inline strdup(); there were conflicts over the type of
    "malloc()".


frexp.cpp, ldexp.cpp, _normdf.cpp:: kub
    better support for 32 bit ints (when used as exponents that is)

device.h, stat.h, types.h,
chdir.c, mkdir.c, open.c, rename.c, rmdir.c, spawnve.c
stat.c, symdir.h, symlink.c, unlink.c, unx2dos.c, utime.c:: ers
    Here are diffs to support the "/dev/" directory. Although the "device.h"
    file claims to describe user definable devices, that isn't done yet;
    for now only the console, printer, and rs232 are supported, as well as
    Edgar's /dev/A/ alternative to A:, etc. There are some minor cleanups
    to the _unx2dos code to recognize device names; also, stat() has been
    fixed to deal with devices (so "ls -l CON:" works now!).
    _unx2dos now returns symbolic constants defined in symdir.h.
    note the name change of unx2dos to _unx2dos all over.

*akefile*:: ++jrb
    s/CPPFLAGS/PPFLAGS/g because gnumake insisted on automatically
    adding CPPFLAGS to CFLAGS (this behavior is a little strange)
    thanks to bill shorka.

*akefile:: ++jrb
    i finally got sick of dealing with all these makefiles.
    merged all the common parts into a new include files "mincl"
    adjusted makefile.16, makefile.32, gmakefile.16, gmakefile.32
    these are now used for both native atari and unix based
    cross-compiling using gnumake. i have a Makefile that i
    use for compiling that invokes gnumake on makefile.*
    and gmakefile.*  with CC, AR, CPP etc as args.

    please remove CGMakefile.* CMakefile.*, mincl.16 and mincl.32

stat.c:: ++jrb
    was not setting st_nlink for regular files

unx2dos.c:: ++jrb
    was not covering the case "/dev/D" (its odd to me atleast to
    type "ls /dev/D/", instead of "ls /dev/D")

_normdf.cpp, ldexp.cpp, frexp.cpp:: ++jrb
    quotes in comments were thowing gcc-cpp -P off (sorry i forgot
    about this).

dirent.c, stat.c :: ++jrb
    dta was not being reset back to orig dta before returning in
    two cases. this may cause problems as the dta was set to a local
    structure.

unlink.c:: ++jrb
    initialize dir = NULL at the top, so that _free_symdir()
    doesnt get junk when !ent and it is called. thanks to allan
    who reported and suffered thru this bug!
    Check in all files for any more instances of the same thing. they
    seem to be ok.

time.h:: ++jrb
    changed components of struct tm to be of type int rather than
    short as per ansi sec 4.12.

main.c:: ++jrb
    let the implementation dependent exit value of main be EXIT_SUCCESS
    if main() fall thru.
    this saves us a lot of headaches while porting, and is compliant.

gemdefs.h, obdefs.h:: ++jrb
    new files. aliases for gemfast.h to make porting easier.

close.c:: ers
    allow for device file names (open keeps track of them for fstat now)

ftw.c:: ers
    try to be smarter about whether to use '\' or '/' in file names

rename.c:: ers
    watch out for renaming when the new file name has the same _unx2dos
    result as the old one

stat.c,rmdir.c, unlink.c:: ers
    if links are active, insist that the proper file names are given
    (so the user doesn't accidentally delete "foo" when s/he asked to
    delete "Foo")

stat.c, symlink.c:: ers
    fixed up some comments, moved lstat to stat.c, fixed fstat to
    not try to search through links (it already has the DOS name
    given)

unx2dos.c:: ers
    if _lAUTO is on and a filename is given that needs changing for
    DOS or that already exists as a link, make sure that its DOS
    name is unique. Doing this here saves a *lot* of hassle in
    functions all over the place. Not extensively tested yet,
    but seems to work. We can probably do better than the present
    algorithm.

chdir.c:: andreas schwab
    fixed up if expression.

linea.h:: andreas schwab
    fixed up typo in cast for linead (last arg)

sys/ioctl.h:: ++jrb
    another alias for ioctl.h

getcwd.c:: ers
    Don't add the '\' for the root directory. This makes everything
    consistent; getcwd() always returns a path without trailing '\'.
    If symbolic links are on, use _full_unx2dos (see below) to
    look for long names, etc.

link.c, symlink.c, mincl:: ers
    link() moved to a separate file, so that programs like "tar" that
    supply a copying version of link won't have problems

unlink.c:: ers
    major bug in handling of automatic link deletion corrected (this
    bug caused a crash when an error occured during unlink()).

unx2dos.c:: ers
    add a routine, _full_dos2unx, which searches along the path looking
    for automatic symbolic links to the components, so that the unix
    name reflects what the user wanted.

ftw.c:: ers
    fixed the algorithm for deciding what kind of slash to use; it works
    now if no slash appears in the original name

symlink.c:: ers
    fixed readlink() to work correctly when the link being read points
    to another symbolic link

close.c, ioctl.c, isatty.c, open.c, read.c, unx2dos.c, write.c:: ers
    allow user defined "plug-n-play" devices

screen.c, screen.h, mouse.h:: ers
    sample plug-n-play mouse and screen devices, like the Minix gkernel
    /dev/mouse and /dev/screen

minimal.h:: ers
    add _init_signal so as to avoid dragging in the signal routines,
    and change _main to pass argc and argv on to main (since crt0.c
    parses the arguments anyways, we might as well let people use them!)

dirent.c:: ers
    fix a bug whereby failure to open the ".dir" file when auto links
    are active could cause a crash (actually, this only happened when
    the error exit code is not ENOENT)

spawnve.c:: ers
    allow mode == P_NOWAIT for microRTX and MiNT

spawnvp.c:: ers
    put error code in "errno" and return -1 on failure, instead of
    returning the (now positive) error code

findfile.c::
    terminated a string that didn't used to be terminated

write.c::
    made all terminal writes go through console_write_byte
    (so signals will always be checked); also checked and
    turned off '\n' -> '\r\n' translation when the terminal
    is in RAW mode (curses, for example, seems to expect this)

keycodes.h::
    fixed definitions of the ALT_1 through ALT_0 keys

getw.c:: ++jrb
    getl() had an incorrect mask. thanks to Rainer,[e-mail]

atof.c:: ++jrb
    __adjust should have been static

filbuf.c:: edgar
    do not remember EOF if input comes from a tty (er)

atof.c:: edgar
    change overflow condition in tem_mul(). ( was causing
    truncation at the 21'st digit)

fopen.c:: ++jrb
    freopen(): if the file is already open, close it. If it was closed
    its fine. (used it be: it failed if file was closed -- thnaks bill)

sysvars.h:: ++jrb
    added OSHEADER
    (note: atari (tos 1.4 addendum) docs for os_date incorrectly states
    that the format is $YYYYMMDD --  it is really $MMDDYYYY)

linea.h:: ++jrb
    adjusted logic ops. thanks alan.

main.c:: ++jrb
    make at_exit() a little safer by not loosing _at_exit[] when
    realloc fails. thanks allan.

main.c:: ++jrb
    exit with returned value from main(). if its junk, then the
    users main() needs to be fixed (no way to find out if returned
    value is junk or not).

main.c:: edgar
    initialize __open_stat for the standard file descriptors

close.c:: edgar
    check __open_stat.filename before freeing

dirent.c:: ++jrb
    dont check return value of Fsetdta (it does'nt have one)

exec.c, spawnvp.c::
     make execve more robust by checking X bit of stat.
     add execl and execvp

getuid.c:: edgar
    add geteXXX()
    for uid's return the value of var uid, instead of hardcoded 1

getpid.c:: edgar
    add getppid()

grp.*:: edgar
    add passwd field to grp structure

stat.c:: ++jrb
    dont check return value of Fsetdta (it does'nt have one)

basepage.h::
    elaborate it further

dirent.h::
    change defn of DIRSIZE

gemfast.h::
    mf_fg/bg were swapped in MFORM str.

osbind.h:: edgar
    adjust return value types of various bindings.
    added some system str types

pwd.h:: edgar
    add defn for pw_comment

vt52.h:: frank
    new file, thanks frank

assert.h::
    change defs. pass __LINE__ as a long. __eprintf is expects a long.

fprintf.c::
    adjust types for args to __eprintf

errno.h::
    merged in many aliases. not sure if we a blatently currupting
    the name space.
fcntl.h::
    minor adjustments. added the flock defs etc for MiNT. ERS i am
    not sure i understand why __SMALLEST_VALID_HANDLE should
    be 0 when !MSHORT (i know that handles are only word -ve, but
    when you assign to int, it will get sign extended).

gemfast.h::
    choose between defferent defs of struct MFDB by using
    __MFDB__ or __ALT_MFDB__ pre-processor symbols. (cp this
    to gemlib src directory too).

ioctl.h::
    add defs for EVENP, ODDP, change def of LCASE. ers: why did
    you del all the defs for baud rates?

limits.h:
    added the POSIX defs. they dont mean much.

linea.h::
    fix defs for linea7, and linead. fix proto for linead when no inline.

linea.c::
    do the same in linea.c

memory.h::
    use compiler.h, get rid of silly compatibility names

minimal.h::
    make everything go thru extern func __exit() instead
    (note not _exit(), otherwise main.o gets pulled in, which
     is not what we want)

osbind.h::
    add TT bindings.

osbind.c::
    same here. also fixed return retvalue; all over.
    i guess people must never __NO_INLINE__ otherwise this
    should have shown up. (actually no, since retvalue happens
    to be in d0 already!).

param.h::
    rm old args for NCARGS -- now its as much as free mem
    ERS note NO change def of HZ to  200. there is too much
    code assuming HZ == 60, so we must scale the 200 ourselves.

siglist.h, signal.h::
    replace with ers's files

stat.h::
    correct types
    add def for S_IFIFO

stddef.h::
    use compiler.h

stdio.h::
    use compiler.h
    use alt inline keyword when inlining getc

stdlib.h::
    use compiler.h

string.h::
    use compiler.h
    use alt keyword __inline__ when inlining

time.h::
    use compiler.h
    typedef for timeval

types.h::
    use compiler.h

compiler.h mintbind.h resource.h unistd.h wait.h regexp.h::
    new files (mintbind, resource and wait.h are #ifdef __MINT__ protected)

sys/errno.h     sys/resource.h  sys/signal.h    sys/wait.h::
    new files

ctype.c:: ++jrb
    ctype should accomodate all UCHAR max characters.
    make the type array UCHAR+2 long, accomodating EOF as
    the first element. that way you dont have to depend on
    a seperate array being contiguous right before the type
    array, to accomodate EOF. make ctype a pointer into
    this type array[1]. thanks jens.

ctype.h:: ++jrb
    adjust decl of _ctype

fork.c:: andreas
    move the code to save the stack data after the point where
    newbase is initialized.  needed due to diff in the way reg allocation
    is done in gcc ver > 1.37. thanks andreas.

_addsubfs3:: andreas
    It bites you when you add an insignificant number
    to another one, so that the difference between the exponents is
    greater than 24. In this case the result is in the wrong register. It
    seems that someone has `optimized' this in.

mktemp.c:: ++jrb
    dont always start tryng to make a temp file @ XXXXX0, instead
    start trying at XXXXX0+startat. this ensures that consequtive
    calls to mktemp() with actually creating files, yields unique
    names (in the same process of course). for MiNT incorporate pid
    or something.

osbind.c:: ++jrb
    adjusted proto for trap_14_wwwl

strstr.c:: ++jrb
    need proto for strncmp not strcmp

atof.c:: ++jrb
    added tests for 1 and 19/20 sig digits

linea.c:: ++jrb
    adjust proto for linead()

makefile.*, mincl:: ++jrb
    add 020 targets for mc68020/030 etc targets.

rename some long filenames as requested:
    ./Changelog ./Changelo
    ./PatchLevel.h ./PatchLev.h
    ./_addsubdf3.s ./_addsubd.s
    ./_addsubsf3.s ./_addsubs.s
    ./_divmodsi3.s ./_divmods.s
    ./_extendsfdf2.s ./_extends.s
    ./_fixunsdfsi.s ./_fixunsd.s
    ./_floatsidf.s ./_floatsi.s
    ./_truncdfsf2.s ./_truncdf.s
    ./_udivmodsi3.s ./_udivmod.s
    ./gmakefile.16 ./gmakefil.16
    ./gmakefile.32 ./gmakefil.32
    ./localtime.c ./localtim.c
    ./setlocale.c ./setlocal.c

Makefile, mincl : ++jrb
    adjust for the renamed files

Makefile:: ++jrb
    added 020 targets and defs

stksiz.c:: ers
    default symbol _stksize defn. if the user provides no def, this
gets pulled in. neat idea ers.

gemfast.h:: ++jrb
    added TurboC compatibility def for OBJECT (controlled by
#ifdef __TCC_COMPAT__ ).

access.c chdir.c chmod.c console.c dup.c fflush.c filbuf.c fputc.c
fread.c fseek.c fungetc.c fwrite.c getbuf.c getcwd.c getlogin.c
getopt.c getpid.c getpw.c getuid.c gmon.c isatty.c lseek.c malloc.c
mkdir.c open.c popen.c psignal.c read.c rename.c rmdir.c sbrk.c
setbuf.c sgtty.c sleep.c stat.c textio.c time.c unlink.c utime.c
write.c:
    s/unixlib.h/unistd.h/

close.c fhandle.c isatty.c open.c stat.c unlink.c:
    s/N_HANDLES/__NHANDLES/

allover:: ++jrb
    fixed function proto args to pass -Wstrict-prototypes.

assert.h:: ++jrb
    - if !stdc make sure __LINE__ and __FILE__ are def'ed
    - make sure that __LINE__ is printed using %ld not %d
      and cast (long)__LINE__

dirent.h:: ++jrb
    track name change for DIRSIZE/DIRSIZ from mntlib

osbind.h:: ++jrb
    make all the defined structs grow a leading _ (_KBDVEC etc)
    to avoid clashed with user defined types.

siglist.h:: ++jrb
    as in mntlib, move the siglist and signal names to siglist.c
    declare them extern here. (ers, why did you delete signal_names?)

siglist.c:: ++jrb
    new file with defn for sys_siglist and signal_names

mincl:: ++jrb
    add target for siglist.o

stdlib.h:: ++jrb
    add EXIT_FAILURE ifndef as in mntlib

unistd.h::
    ers, why did you make the counts unsigned?? this causes problems
with the return type, on a successful read you will get <= count or 0 or -1
but is count is unsigned, the err condition potentially becomes
undistinguishable.

alphasor.c:: ++jrb
    new file from mntlib, the sorting predicate for scandir().

mincl:: ++jrb
    - add target for alphasor.o
    - make it consistent with mntlib mincl (targets for GNULIB2 and
      rule for .cpp)

lib.h:: ++jrb
    got from mntlib, added console_input_status proto.

chdir.c:: ++jrb
    include lib.h, void cast return value of Dsetdrv as per mntlib

chmod.c:: ++jrb
    merged in change from mntlib that ensures archive bit is not
    cleared.

crt0.c:: ++jrb
    make _stksize extern that is picked up from user or if not
    specified by the user, from stksiz.c (where it is zero).
    Nice idea ers!

fflush.c:: ++jrb
    get rid of unused local c, per mntlib

allover:: ++jrb
    add #include "lib.h" as appro.

ftw.c:: ++jrb
    replaced with one from mntlib (stat should take care of UNIXMODE
    too, no?)

localtim.c:: ++jrb
    replace with one from mntlib

scandir.c:: ++jrb
    new file, from mntlib

mincl:: ++jrb
    add target for scandir.o

stat.c:: ++jrb
    look again-- add in real inode # code (edgar sent me, its
    here somewhere)

system.c:: ++jrb
    replaced with ers/pratt fork()'less code. (ers will re-direction
    in the parents context work just as well?? actually its probably
    ok, since gemdog is so foobar'ed already).


support.h, lib.h:: ++jrb
    consolidate the two. needs  to be checked.

dirent.h:: ++jrb
    replace NAME_MAX with limits.h, per mntlib

ioctl.h:: ++jrb
    add new MiNT proc ioctls

stdio.h:: ++jrb
    track renaming of vars in __getc__ from mntlib

string.h:: ++jrb
    incorporate apratts super/duper empirical findings

osbind.h:: ++jrb
    added definitions for Bconmap() and Bconctl()

malloc.c:: ++jrb
    test return value of sbrk == -1 for no mem, as the TT can ret
    memory with high bits set, per mntlib

sbrk.c:: ++jrb
    ers why did you increase slush from 64 to 256??
    changes to return -1L when no mem, as per mntlib

spawnve.c:: ++jrb
    copy argv[0] insted of path into env following ARGV=, per mntlib

stat.c:: ++jrb
    - i was resetting the dta in the if part, changed it to
    be consistent with mntlib.
    - deal with root dir of logical drive, per mntlib

times.c:: ++jrb
    ers can we switch back to HZ == 60 and scale as appro. there
is too much stuff out there that blindly assumes this, without looking at
param.h.


strlen.c:: ++jrb
    use the same pratt hack as in string.h (return ptr diff instead
of count -- and pray that the gcc code selection never changes  half :-)

strftime.c:: ++jrb
    replace with one from mntlib

regexp.c regsup.c regexp.h: ++jrb
    new files from mntlib.

mincl:: ++jrb
    add targets for regexp.o regsup.o

access.c:: ++jrb
    replace with one from mntlib

tmpnam.c:: ++jrb
    changed mask for mktemp from $$XXXXXX to TTXXXXXX. $$ was
    causing too many name problems (especially with shells).

dup.c:: ++jrb
    added upperbound check for handles as per mntlib

fgets.c:: ++jrb
    initialize c to EOF.

findfile.c:: ++jrb
    replaced with one from mntlib

fopen.c:: ++jrb
    ignore close status when freopening as per ansi/mntlib

getenv.c:: ++jrb
    replaced with one from mntlib.
    NOTICE new function putenv()

unistd.h:: ++jrb
    add proto for putenv()

gnulib2:: ++jrb
    bdiv is suspect:: replace with my one after test.

lseek.c:: ++jrb
    put #ifdef __GNUC__ around builtin_alloca

main.c:: ++jrb
    put in MiNT detection.
    NOTICE new var __mint;
    invoke MiNT detect
    or in __default_mode__ for the standard descriptors per mntlib

minimal.h:: ++jrb
    add _exit(), since signal and system call it.

malloc.c:: ++jrb
    ers: i liked the sentenials very much.
    include <unistd.h> for sbrk
    change _sbrk() to sbrk()

popen/pclose: ++jrb
    i dunno about these two.

ers:: i am totally out of sync with read/write and console routines.
i think i will leave the devices in for now. have to decide what strategy
would be best for the lib, and yet try to converge towards MiNTs idea
of these things... let me know if you get any inspirations.


setjmo.cpp:: ++jrb
    test for MiNT

scanf.c:: ++jrb
    ignore the following warnings:
scanf.c: In function _scanf:
scanf.c:123: warning: argument passing between incompatible pointer types
scanf.c:124: warning: argument passing between incompatible pointer types
scanf.c:216: warning: argument passing between incompatible pointer types
scanf.c:227: warning: argument passing between incompatible pointer types
scanf.c:364: warning: argument passing between incompatible pointer types
scanf.c:383: warning: argument passing between incompatible pointer types

main.c:: ++jrb
    at exit, only flush std streams that are not tty's. this is
important for some shells that have re-directed the streams, and
are going to un-redir them on exit (ERS: need to incorporate this in
mntlib too).

bsearch.c:: andreas schwab
`size_t' is unsigned long and ((size_t)-1) is greater than 0. That
leads to an endless loop (and index out of range) if the key is less
than the smallest element in the array. (in deciding dir)

unistd.h:: ++jrb
    add protos for random(), srandom() etc.

qsort.c:: ++jrb
    adjust for non-stdc compiler, add <alloca.h> for sun sparc
    yes, all those casts are actually required for some
    dumb-shit compilers

tmpnam.c:: michal
    use TxXXXXXX as the template, TTXXXXXX is more likely.
    use \ instead of / for shells and other things
    that dont understand unixmode

unistd.h:: ++jrb
    change read/write to conform to posix:
    int read(int, void *, unsigned int nbytes);
    int write(int, const void *, unsigned int nbytes);

    max nbytes == 0xffffe (0xffff fffe), return value all F's
    on error, with errno set. the user is responsible for
    handling the sign of the returned value as appro.
    (this really sucks).

read.c, fread.c,..., write.c, fwrite.c...:: ++jrb
    adjust for above. NOTE: in _read()/_write() we dont accept
    counts greater than LONG_MAX. i dont think is an unreasonable
    restriction at all :-)

osbind.h:: andreas schwab
    in the definition for Flopfmt, the return value has the wrong cast.
    the bad sector list is returned in the buffer pointed to by `buf',
    in d0 the usual error code is returned.

    in the `trap_...' definitions where the asm statements are splitted,
    (across two asm's -- ed:jrb)
    i have deleted the `volatile' for the parameters which are pushed in
    the first asm statement. this gives slightly better code, and i
    don't think that it causes any harm, because the stack pointer isn't
    changed yet.

dirent.c:: andreas schwab
    function opendir()
    the dirent list should be linked forward, so that `ls +sort none'
    gives the correct order.

    i have also slightly improved the closedir function (i like double
     pointers :-)

everywhere:: ++jrb
    convert to conditional macros _PROTO() style for all protos.

open.c, *scanf*.c, *printf*.c:: ++jrb
    convert all these to true <stdarg.h> style for var args
    instead of faking them out. get rid of the #ifdef __SRC__
    kludges everywhere (i think these os only one instance
    left for which there really is'nt a clean workaround).

everywhere:: ++jrb
    sync in with mntlib #9. change all _PROTO to __PROTO and
    _EXTERN yo __EXTERN. Also, use that compiler.h
    ers: note i added cplusplus too.

st-out.h:: ++jrb
    add defn for A_LNAM a gst compat long symbol (activated by ld -G)

atof.c:: ++jrb
    ok my arm got twisted enough: make the return pointer
    char ** even though you pass it a const char *.
    (another shining example of design by committee).
    adjust casts.

scanf.c:: ++jrb
    shut the compiler up with casts

localtim.c:: ++jrb
    make indst take a const pointer as it should.
    adjust casts.

qsort,c:: ++jrb
    adjust a few casts

strftime.c:: michal
    replaced file with code from michal. thanks!

atof.c:: ++jrb
    use HUGE val from math.h

sbrk.c:: ++jrb
    round up size passed to HeapAlloc to the next octet (same min
    alignment as in malloc()). It was possible for the heap
    to become unaligned. Thanks for michal jaegermann for catching
    this.

keycodes.h:: michal
    add scan codes for shifted func keys (i did'nt even know
    they were any). thanks michal.

doprnt.c:: michal
    Characters from formats which are echoed literally
    are not counted.
    Formats 'g' and 'G' do not pay attention that some significant
    digits were possibly produced in an integer part and are ready
    to insert any numbers of 0's before next non-zero will be reached.
    Thanks michal.

limits.h:: andreas
    When char unsigned, define the char limits appropriately.

limits.h, stat.c, dirent.[ch], unx2dos.c, sym*.c: andreas
    your latest updates introduced a bug in unx2dos.c: in limits.h
    NAME_MAX is defined to _POSIX_NAME_MAX, which is 14, but in unx2dos.c
    NAME_MAX is used to dimension the internal arrays for the file name
    conversion. thus if you pass an file name with a length of at least
    14, some very dangerous things happen. it took me some hours of
    debugging to find it out.

    Thanks andreas, i changed all dependencies of NAME_MAX in
    the library to LIB_NAME_MAX, so that for instance the MiNT
    library can define it appro for the library. the def of
    LIB_NAME_MAX is 32, since that is what it was assumed as.
        Anything less will blow up stuff in unx2dos as andreas
    pointed out, and also in dirent.c, symdir.c and stat.c.

RENAME:: frank
    _fixunsd.s -> fxunsd.s

RENAME:: ++jrb
    all over in the pml lib and include files:
        pmluser.h ->  math.h

atof.c, strtol.c, strtoul.c:: michal
    If you will look closer at strtod function you will find out that,
    contrary to ANSI specifications, there is no way to tell if a
    string passed to it was "0.0", or a string like "+" or "-" or
        consisting only from a white space.  The problem is that it does
    not set *endptr to nptr, even if a conversion was not really performed.
    It will also happily accept something like that "0.0.0.0"
    without stopping at the second dot.

    A similar bug affects also strtol and strtoul (and I do not know
    what possibly else).

osbind.h:: frank
    the new TT defs were returning int instead of short

stdio.h:: jrb
    change getc to a # define instead of an inline when !__NO_INLINE__
    this keeps gdb much happier, and also produces better code in
    some situations (but never worse).

unistd.h:: ++jrb
    add proto for system()

all over:: ++jrb
    fix up for -Wall: add casts, include files, add/sub protos etc

strtol.c, strtoul.c:: michal
    arrrgh! the above changes at Patchlevel 66 broke these. Michal fixed
    it, and also extended them for ansi conformance to accept  bases
    between 2 and 36 and 0. NOTE: if you built gcc @ patchlevel 66,
    please BUILD IT AGAIN. Due to broken strto*l the code will be wrong.

atof.c:: michal
    removed possibility of an infinite loop

ident.c:: ++jrb
    make it hold a static string identifying the Patchlevel so that ident(1)
    can pick it up

mincl:: ++jrb
    add ident.o target

perror.c: michal
    adjust strings for EDOM and ERANGE to correspond to <errno.h>

perror.c:: ++jrb
    - calculate sys_nerr based on sizeof sys_errlist
    - no need to check for return value of strerr for null

textio.c:: michal
    - a buffer with only CRs would call read() with 0 bytes to read,
    and the resulting zero count (correctly) returned by read() gets
    interpreted as an EOF. Thanks michal!

strtol.c strtoul.c:: michal ++jrb
    fix it up. should be close to ansi requirements, sans some debatable
    cases like where the endptr be for "0xGXXXX".

crt0.c:: ++jrb
    get rid of unused label

printf*, scanf*:: ++jrb
    fix up protos. note we fake the _scanf/_doprnt protos to fit
    the situation.

malloc.c:: ++jrb
    dummy should always be there to avaopi bra with zero offset

textio.c:: michal
    fixup the case causing premature EOF when read returned a good count
    but nbytes became 0 because it was skipping over \r's.

chdir.c, chmod.c, dirent.c, mkdir.c, open.c, rename.c, rmdir.c, spawnve.c,
 stat.c, symdir.c, symlink.c, system.c, unlink.c, utime.c, lib.h:: ++jrb
    use unx2dos() instead of _unx2dos, similarly dos2unx() instead
    of _dos2unx(). that way if the user re-vectors the mapping functions
    the users routines gets called.

unx2dos.c:: ++jrb
    maybe we should be checking bounds on the filenames against
    _LIB_NAME_MAX. i dunno, this is not pascal, soo GIGO :-)

stdio.h:: ++jrb
    use safe def for __fp

---------------------------------- Patchlevel 68 ---------------------------

read.c:: andreas
    echo chars >= 0x80 as normal characters for users with national
    chars.

sscanf.c, fprintf.c:: andreas
    delete spurious arg decl

device.h:: ++jrb
    define mkdev/major/minor

device.c:: ++jrb
    NEW FILE, this code was previously in unx2dos.c, but it needs
    to be seperate so that people can see it (and use it :-).

mincl:: ++jrb
    add target for device.o

all*.h:: ++jrb
    add in extern "C" {} 's for g++

---------------------- PatchLevel 69 -----------------------------

bcmp.c, bzero.c, div.c, malloc.c, read.c, sbrk.c, strchrc.c,
strrchr.c, unlink.c, write.c:: andreas
    equivalence names with .stabs, rather than the old dirty trick
    which could not guarantee that a string constant would not
    be generated by the compiler before the entry point of the
    code that the label was being equvalenced to. Neat idea!


gmon.c: andreas
    clear histo buffer (dont depend on a clear bss)

gprof/gprof.c:: andreas
    fix up for long GST symbols (aka gcc -G)

fscanf.c:: ++jrb
    add vscanf(), vfscanf()

sccanf.c:: ++jrb
    add vsscanf()

stdio.h:: ++jrb
    add protos for v*scanf()

setjmp.h:: ++jrb
    dont proto catch/throw for g++

---------------------- PatchLevel 70 -----------------------------

makefile.??, gmakefil.??:: ++jrb
    i forgot these, organize them better (thanks allan)

---------------------- PatchLevel 71 -----------------------------

strtol.c, strtoul.c:: michal
    fix up for case "0xGYYY" (0x followed by invalid hex digits)

fflush.c:: ++jrb
    when _IOREAD and lseek fails on a device (_IODEV), ignore it.

signal.c:: ++jrb
    - __check_signal should be an extern (it is declared in console.c
    or xconsole.c when using termcap).
    - make __check_signal counting for  either or !SIG_IGN

fcntl.c:: ++jrb
    new file, limited implementation of fcntl() functionality
    fcntl cmds F_DUPFD, F_GETFD, F_SETFD, F_GETFL, F_SETFL
    note: DUPFD has very limited semantics, about all we can do in TOS
          GET/SETFL: get and set the __open_stat flags, unlike unix
          semantics for these calls.

close.c, fcntl.h, spawnve.c:: ++jrb
    add support for close on exec fcntl flag.

mincl:: ++jrb
    add fcntl.o target

write.c:: ++jrb
    add support for O_APPEND flag. (seek to end before each write).

null.c:: ++jrb
    new file, implemntation of /dev/null using plug'n play device
        interface. use install_device() before using device to install
    /dev/null as a legitimate device.

stat.c:: ++jrb
    check for device in fstat, pass full unixname to stat if so.
    stat already handled devices.

mincl:: ++jrb
    add target for null.o
---------------------- PatchLevel 72 -----------------------------

symdir.c: andreas
    delete the .dir file if there are no links
    some cleanups in _cache_lookup (yes, it produces some fewer
    bytes of code :-)

unx2dos.c: andreas
    change the dot conversion strategy
    the old code had difficulties with filenames of the form
    asdfgh.tar.Z:
    _adjust("asdfgh.tar.Z") -> "ASDFGH,T.Z", the linked-to name
    becomes "asdfgh.t.z". but when later retrieved, _adjust is
    called with the linked-to name
    _adjust("asdfgh.t.z") -> "ASDFGH.T,Z"
    with the new coding:
    _adjust("asdfgh.t.z") -> "ASDFGH,T.Z"
    there are yet some problems with filenames ending in '.', the
    period will be lost after conversion

crt0.c:: ++jrb
    revert back to testing A0 (instead of SP) after controversy on the net.
    packers will just have to fix themselves.
    in addition to testing A0 check long A0@(36) (parents basepage). for an
    acc this should be NULL.

crt0.c:: frank celler
   before fixing up env if launched from the desktop, be careful that this
   is not really the value of a env variable with a NULL value.

fcntl.h:: ++jrb
    adjust constants to match MiNT.

console.c:: Scott Kolodzieski (uunet!sparc1.stevens-tech.edu!scott )
    new function flush_key_buff(): enables you to stuff the
    currently accumulated (in the system)  chars to the console
    buffer. this is important in the situation
    where you do a fork(), and chars accumulate while in the child
    process.

ctype.h:: michal
    make sure isprint/isgraph dont become true for chars > 0x7f
    (they were before). thanks michal.

symdir.c:: scott
    _del_symdir_cache(): routine to flush out symdir cache. please
    note the interaction between symdir cache and unx2dos before
    using this.

crt0.c:: andreas
    the amount of space for the environment and the arguments reported by
    the function is sometimes four bytes too small, so that the first
    malloc() clobbers the NULL pointer at the end of the argument list.
    this is a bad idea especially for programs which rely on this NULL
    pointer (which is completely legal).

fork.c:: andreas
    there is a subtle bug in _fork (fork.c). suppose you are doing
    something like this (which is essentially what gnumake does):

  if ((pid = vfork()) == 0)
    {
      /* children side */
      do_something();
      execv(...);
    }
  else
    {
      /* parent side */
      /* ... */
    }

    the problem is, if do_something() does too much malloc()'s, more
    memory must be fetched from TOS (via sbrk()) and is inserted into the
    free memory list by malloc(). but later on the parent's side, this
    memory is no more valid, but still will be returned by malloc(). in
    the case of gnumake, it gets overwritten by subsequent fork()'s and
    *bang*. here it is read_symdir (symdir.c) which sometimes wants too
    much memory (it is called when the program to be executed is searched
    through PATH).
    the solution i suggest is to malloc() a big block and free() it again
    so that there is enough memory left to be handled by malloc(), without
    having to sbrk() more until the "child" process does the execv(). i
    have tested this with gnumake running it over night (making TeX fonts)
    and no more bombs had appeared yet :-)

pipe.c:: Scott Kolodzieski (uunet!sparc1.stevens-tech.edu!scott )
    implementation of pipe for TOS. thanks scott!
    everyone, do yourself a favor and check out scotts port of
    bash. its great!

close.c, open.c, fcntl.h:: scott
    chnages to support pipes.

unx2dos.c:: ++jrb
    translate /dev/D\foo correctly. the code was looking
    for only '/' after /dev/D (after the drive char).

memcpy.c:: ++jrb
    stab hack in memmove() at bill shrokas suggestion. the
    memcpy() handled overlap, so its ok.

string.h:: ++jrb
    add proto for memmove()

malloc.c:: realloc() michal
    if space cannot be allocated by realloc, the orig object
remains unchanged (and a NULL is passed back).


---------------------------- Patchlevel 73 ---------------------------

popen.c:: ++jrb
    use tmpnam() to get temp filename. That way $TEMP || $TMPDIR
get looked up.

main.c:: ++jrb
    moved clearing of errno to just before calling main(). otherwise
somehow is was !=0 at the start of main() in some cases.

unlink.c:: scott
    i cant believe this was still lurking in there. Thanks scott!!
    if the file to be unlinked was open, and we put it in the
    atexit chain, we were'nt strduping the filename, as a result
    all filenames in the chain were pointing to the same loc.
    arrgggh! now strdup() the name, and free it in do_deletes. if
    the strdup() fails, link it in the chain with a null filename,
    and in do_deletes() just skip over entries with null
    filenames. we need to do this as we cannot un-atext the entry,
    nor do we want to as they may be others in the chain before
    this one, or successive attempts may succeed.

main.c:: ++jrb
    call the atexit functions after closing all the open files.
    this is essential if the deferred unlinks() etc are going to
    work.

binmode.c:: ers
    new file. as per mntlib pl 13, __default_mode__ is now a
        global that the user may define. if not defined by the user
    it gets picked up from here, with a default value of 0 (text mode)
    the user may now define __default_mode__ = IOBIN; and that
    way stdin etc get opened in binary mode. thanks eric.

fdopen.c fopen.c main.c:: ers
    adjustments for __default_mode__.

mincl:: ++jrb
    add target for binmode.o

localtim.c:: ers
    sync with mintlib, use symbolic constants

main.c,spawnve.c,findfile.c: ers
      when starting up, convert the environment variable PATH to
      POSIX form (i.e. using ':' as a separator); when execing a
      new program, convert back to old form so old programs will
      work correctly. thanks eric.

fread.c:: jeffrey jackson ([e-mail])
    fread was'nt returning EOF on large reads. thanks jeffrey.

mktemp.c rename.c symlink.c:: ++jrb
    save errno around calls to access()

symlink.c:: jeffrey
    the length returned was wrong because we were doing a strncpy
    before strlen.

main.c:: ++jrb
    for stdin/out/err set open_stat[].filename to the unix device
    so that fstat() works. thanks to jeffrey for discovering this.

pipe.c, close.c::  scott
    minor fixes for pipes. bash-1.10 works real well now.

stat.c: jeff
    handle the case for fstat() when stdin/out etc are re-directed.
    (ie: open_stat[].filename == NULL).

unx2dos.c:: jeff
    overwrite __link_to only once when following links, so that
    lstat() works correctly.

symlink.c:: jeff
    readlink didn't work for symbolic links in root file system if
    the passed name had the slash.  This was because the slashed
    was nulled to create the directory name of the file (boundry
    condition type bug).

scanf.c:: ++jrb
    %n was'nt storing the # of chars read in so far, but was storing
    number of conversions done. thnaks for allan pratt  for reporting
    this.

scanf.c:: ++jrb
    old one had too many subtle problems. replaced with one from
    gnu libc with a few subtle corrections, and also a bug fix.
        NOTES:
    - interface is different from equivalent gnuC lib function
          it was munged to match our old _scanf().
    - if __NO_FLOAT__ is defined then the floating point stuff
      gets nuked (for iio*.olb) as per our old scanf.c.

       It is very important to read and understand the GNU Library General
       Public License. It specifies rights and conditions that are different
      from the GNU copyleft.

atof.c:: ++jrb
    strtod() was leaving the pointer one char too far when input
    had a tailing 'e' (for example after 100ergs, the endptr would
    be pointing to 'r' rather than 'e').

scanf.c:: ++jrb
    had to hack it quite a bit to get out bugs. the regression test
    is great now!

crt0.c:: ++jrb
     More meanings for _stksize (thanks to eric and allan for the feedback)

    _stksize            meaning
      -4L        keep 3/4, free 1/4, malloc from own heap
      -3L        keep 2/4 (1/2), free 1/2 malloc from own heap
      -2L        keep 1/4 of memory, free 3/4, malloc from own heap

    NOTE: all of the following will do malloc from Malloc() first,
    when that fails, malloc() will do further mallocs from
    our own heap. This lets us use the maximum amount of
      memory in traditional ST's as well as newer split address
    STs.

      -1L        keep all of memory (except MINFREE at top) and do
            mallocs from own heap, with heap grown upwards towards
            stack, and the stack growing down towards heap,
            with a minimum slush between them so that they
            dont meet (only checked while malloc'ing). With
            this model, further spawning is not possible, but it is
            well suited for programs such as gcc-cc1 etc.
            Thanks to Piet van Oostrum & Atze Dijkstra for this idea

    0L        keep minimum amount of memory. this is also the
            case when _stksize is undefined by the user.
    1L        keep 1/4 of memory, free 3/4 ( as in Alcyon GEMSTART)
    2L        keep 2/4 (1/2), free rest
    3L        keep 3/4, free 1/4
    other        keep that many bytes
    -other        keep |other| bytes and malloc from own heap


    Also note that when _initial_size is specified, malloc happens from own
    heap only.

crt0.c:: ++jrb
    New convention for ACCs: (the old convention is still supported):

    If _heapbase is not specified by the user, _stksize bytes will
    be Malloc()'ed from the system, and _heapbase will be set to
    this area, and sp will be set to the top of the allocated
    area. All malloc()'s of course will happen from this local heap.

    The old convention for ACCs where the _heapbase and _stksize
    were specified, and sp was set to _heapbase+_stksize will
    continue to work.

malloc.c, sbrk.c:: ++jrb
    implement _split_mem for _stksize >= -1L. get mem from
    Malloc first, after that is exhausted, start carving out of
    own heap (the heap will be in the other address space of the
    split address space). (ie: other to the one the program is
    launched in). The way this is done is that when _stksize >= -1L
    in crt0.c the vars _split_mem and _heapbase are set in crt0.c.
    then whenever we sbrk, we first check _split_mem, which if true
    causes us to Malloc(). Then when Malloc() fails, and if _split_mem
    was on, we turn it off and that causes sbrk() to start allocing from
    heap.

crt0.c:: ++jrb
    in _start1() the local vars and args MUST be in registers because
    we are playing games with SP and the FP. to ensure this use
    __asm("REG") facility of gcc. thanks allan.

isatty.c:: ++jrb
    put in a check so that PRN: always returns false. i believe
    that this will fix the problem of GS and gnuplot,
    where fwrite() to a file opened in binary mode to PRN: was munging
    the bytes on the way to the printer. Unfortunately, i dont
    have a printer attached to my ST to check.
    Tim says its ok now. GS works!

unistd.h, mkdir.c:: andreas schwab
    mkdir takes a second mode argument (currently ignored).
    thanks andreas.

basepage.h:: ers
    make undocumented parts of the BP hard to get at.

compiler.h:: ers
    as per mntlib PL16. note that wchar_t is now an int.

time.h:: ers
    per mntlib pl 16

sys/timeb.h:: ers
    new file

timeoday.c:: ers
    New file.
    New BSD-ish ftime(), gettimeofday() and settimeofday() system calls.

mincl:: ers
    add target for timeoday.

spawnve.c:: ers
    use TOS_ARGS instead of NCARGS

pipe.c:: ers/scott
    Eric beet down the MiNT/PIPE bug, and I must say with two trivial
    changes the dup2/pipe interface works quite well. Under MiNT
    file handles are not automatically shared with children. So
    the dup2 in bash/emacs was failing.

getpass.c, mincl, unistd.h :: ers
    new file from mintlib for getpass(). Add target to mincl. Add
    proto to unistd.h. Note: you need to put the final '\n' yourself,
    after the call to getpass().

getpass.c:: ++jrb
    Null terminate return string (was being passed back with \n)
    If fget retuns a Null pass back Null (per man on the sun).

gmon.c:: ++jrb
    we collectively decided that all xbra's should be chained via
    the next's. make adjustments to unlink_handler.

---------------------------- Patchlevel 74 ---------------------------

scanf.c:: Michael Ritzert
    - incorrect return value if fewer data items than specified
    could be read: the supplied _scanf always returns -1 in this case.
    Instead, it should return EOF if the error occured before
    the first data item to be read and the number of items
        read with success later on.
    - the %c format passes the character following the one to be read.

fscanf.c:: ++jrb
    #define fungetc to ungetc for pre Ansi sunOS 4.0.*

ctype.[ch]:: andreas
    allow negative characters (also handles 0xff correctly)

fopen/fdopen:: andreas
    set errno to EMFILE if there are no more FILE slots free
     (gawk uses it)

fputc:: andreas
    don't return EOF if c == (signed char)0xff

malloc.c:: andreas
    split memory didn't work
    the memory allocation routines if using the split memory
    model, sbrk can return values which are out of order, due to switching
    from system heap to own heap.

lstat:: andreas
    unx2dos already stored the link flags in __link_flags; no
     need to call unx2dos twice

system:: andreas
    command line argument can be arbitrary long, hope no longer
    than BUFSIZ

unx2dos:: andreas
    the link contents can be a full pathname in symbolic links

write.c:: andreas
    handle chars with high bit set correctly

fcntl.h:: andreas
    GEMDOS returns file handles <= 80 (for TOS 1.4, don't know
    about newer TOS's)

main.c:: andreas
    indices into __open_stat were mixed with file handles.

system.c:: frank
    skip over quotes.

pwd.h, getpw.c:: frank, ++jrb
    adjust types for uid and gid. they should be uid_t and gid_t
    respectively.

pipe.c:: ++jrb
    cleanup: get rid of -Wall warnings.

localtim.c:: ++jrb
    if the time given is weird, return the epoch. this is a questionable
fix: we used to return NULL, but that seems to cause problems, as things
that use the pointer returned by localtime() never seem to check for
the possibility of a NULL. my handy ANSI reference is mum on what to
do when the  time given is weird. this was causing zoo to blow up in
zoosrc/nixtime.i when trying to archive files with weird timestamps (some
people seem to have a weird kbd processor that time warps!).

ostruct.h:: Bill Shroka
    NEW file. it contains most of the popular OS (non Aes/Vdi) structures
    Thanks bill: this is a good beginning to this file, it more or
    less corresponds to franks Documentation/varcard.tex. We can add
    on more stuff as needed.

osbind.h:: Bill Shroka
    use <ostruct.h>

pipe.c:: ++jrb
    All fprintf() in here needed to be conditional upon DEBUG. Thanks
to michal for finding this. his code was unexpectedly pulling in all
kinds of stuff due to external references tp fprintf() here.

pipe.c, close.c:: ++jrb
    take out direct call to pipeclose() from close to avoid pulling
in all the pipe stuff into programs that dont use pipes. instead
go thru a pointer that is initially nulled, and if the program calles
pipe() is patched in with the address of pipeclose(). this way only the code
that actually uses pipes will link in the pipe code.

crt0.c:: ++jrb
    cleanup.

float.h: mjr
    cleanup values to better reflect reality.


***** NOTE ***** ***** NOTE ***** ***** NOTE ***** ***** NOTE *****

_addsubd.s _addsubs.s _divdf3.s _divsf3.s _extends.s _fixdfsi.s _fxunsd.s
_floatsi.s _muldf3.s _mulsf3.s _normdf.s _normsf.s _truncdf.s modf.s:: mjr

    NOTE: these .S files replaced with corresponding .CPP files.


mincl:: ++jrb
    adjust targets for above

math-68881.h:: ++mjr
    removed any error checking code (use the library versions if
    error checking is required)

frexp:: ++mjr
    y = frexp(0,ip) now corectly returns 0 both in y and *ip

    sfp004 - version written but not tested due to lack of hardware
    my sfp004 board died, so only a "plausibility test" could be made.
    most probable sources of bugs:
    the fscale instruction has been inserted to the old function body
    after the dead of the board.
    The test if the 1st arg is 0, too.

doprnt.c:: ++mjr
    NAN and INFINITY conditions are caught by analysing only the most
    significant long word (uniqe).
    this also serves as a workaround to the conversion bug of
    gcc-as
    (DBL_MAX is converted to 0x7ff00000 0x00000002
            instead of   0x7fefffff 0xfffffffe
    ) which sometimes causes doprnt to crash.

_fixdfsi.cpp:
_fixunsdfsi.cpp:
    optional error checking added.
    All three situations (no fpu, sfp004, TT) supported.

_normsf.cpp:
_divdf3.cpp:
_divsf3.cpp:
    returns signed infinity in case of an exception.
    if option ERROR_CHECK is enabled:
    _errno is set to ERANGE.
    The EDOM situation 0/0 is not tested explicitly.
    All three situations (no fpu, sfp004, TT) supported.

_normdf.cpp:
    if option ERROR_CHECK is enabled:
    _errno is set to ERANGE.
    message is printed.
    value of ERANGE corrected to 62 (was 34 = EPATH)

ldexp.cpp:
    value of ERANGE corrected to 62 (was 34 = EPATH)
    optional ERROR_CHECK inserted.
    underflow is no longer considered an error condition to
    stay consistent with pml.
    if ERROR_CHECK is enabled, a message is printed on overflow.

_addsubd.cpp:
_addsubs.cpp:
_muldf3.cpp:
_mulsf3.cpp:    ++mjr:
    sfp004 code and software floating point code merged into one file.

---------------------------- Patchlevel 75 ---------------------------

getopt.c, random.c, ctime.c:: michal
    get rid of dependency on doprnt().

random.c:: ++jrb
    warn about needing an ANSI pre-processor for str concat.

bcopy.s, memcpy.c:: ++jrb
    DELETE these files.

bcopy.s, memcpy.s:: michal
    replace with these. note now memcpy and memmove are NOT
     equivalenced. memmove will correctly handle overlap, while memcpy
     will not.

bcopy.s, memcpy.s:: ++jrb
    make the cnt check unsigned.

string.h:: ++jrb
    adjust protos for bXXXX() functions to size_t for counts.

mincl:: ++jrb
    adjust dependcy for memcpy()

memcpy.s:: ++jrb
    was missing label setup:

osbind.h:: michal
    Add new Waketime() binding for ST Book.

osbind.h:: [e-mail]
    change cast for Kbdvbase() as g++ wont grok void (**)

ctype.[ch]:: ers/++jrb
    change _ctype[] back to pointer so that locales may switch pointers.
    keep andreas's changes for unsigned index into ctype in. that
    way EOF is covered too. (by the 0xff entry).

compiler.h file.h unistd.h assert.h ostruct.h:: ++jrb
    sync up with mntlib pl18 changes for soz etc.

bcmp.c, bzero.c, qsort.c: ers
    Ansification (casting lvalues is a no-no :-).

doprnt.c getcwd.c memccpy.c mkdir.c tmpnam.c fdopen.c
getenv.c memchr.c sbrk.c ftw.c memcmp.c sleep.c crt0.c
getbuf.c malloc.c memset.c sprintf.c strdup.c system.c sysvar.c:: ++jrb
    merge in soz portability changes as per mntlib 18

mkdir.c:: ++jrb
    check for existance before attempting to make per mntlib 18

doprnt.c:: ++jrb
    t was not being initialized in one case. do it at the top.

gnulib2.c:: ++jrb
    add support routines for gcc-2.0: note that the lib needs
    to be re-compiled BEFORE you try to compile gcc-2.0,
    otherwise you will end up chasing your tail!

ldexp.cpp: mjr++
    rewrote the sfp code.

doprnt.c: mjr++
    fixed my bug in the treatment of NaNs and Infinities.

mktemp.c:: ++jrb
    make mktemp() more MiNT friendly by using pid when __mint.

---------------------------- Patchlevel 76 ---------------------------

memcpy.s:: ++jrb
    in the upcopy loop setup for the oddeven case, the count was being
    moved with a movw instead of movl.

---------------------------- Patchlevel 77 ---------------------------

spawnve.c:: Marcus Nick
    cmlen was'nt being initialized before counting the chars in
    args loop.

osbind.h:: ++jrb
     -- another re-hack needed for gcc-2.0: the optimization that we
        used earlier for traps with more than 4 args, making them volatile
    and using "g" constraints no longer works, because gcc has become
    so smart! we now remove the volatile, and give "r" constraints
    (just like traps with <= 4 args). that way the args are evaled
    before we change the stack under gcc, and at appropriate times
    put into reggies and pushed (or as in most cases, they are evaled
    straight into reggies and pushed -- and in even more common cases
    they are already in reggies, and they are just pushed). not doing
    this with -fomit-frame-pointer was causing the temps (from evaluing
    the args) to be created on the stack, but when we changed sp
    from under gccs feet, the offsets  to the temps ended up being wrong.

osbind.c:: ++jrb
    do the same.

fread.c fwrite.c ioctl.c system.c textio.c
compiler.h ioctl.h osbind.h ostruct.h stdlib.h:: ++jrb
    sync with mntlib pl19 changes -- thanks ers
    osbind.h had incorrect fn# for Xbtimer
    other changes mostly for soz compatibility.

filbuf.c getlogin.c lseek.c read.c:: ++jrb
    fixed up a few warnings.

crt0.c:: ++jrb
    round the temporary stack to a long word boundary.

minimal.h:: michal
    tidy up protos
    (in case you did'nt know <minimal.h> is used to make small executables
     that dont use any stdio etc. it avoids pulling in _main() and all
     the stdio, atexit etc etc related code).
    for instance here is a minimal hello world:
    #include <minimal.h>
    #include <osbind.h>
    main()
    {
      Cconws("Hello World\r\n");
    }

atof.c:: Michael
    merge diff for atof

---------------------------- Patchlevel 78 ---------------------------
signal.c, console.c: alexander lehmann
  fix for a bug where console i/o would get stuck when a interrupt
  was caught and it occured when the input buffer was not empty.


fread.c, fwrite.c: alexander lehmann
  I rewrote fread and fwrite to do CR/LF conversion when in text mode.
  Additionally I fixed a small bug in eof detection.

  ( this pertains to the example when you are fread'ing say 20 bytes
    from a file that contains 18 bytes: in the old coding
    you would first succeeed with a count of 18, and on subsequent
    freads you would get eof. In this coding, you get a count of 18
    and also eof flag:
    Well this was not really a bug, just when eof shows up. My ansiC
    doc is mum on this point) -- ++jrb

---------------------------- Patchlevel 79 ---------------------------

gnulib2.c:: ++jrb
    add gcc 2.0 support routine __fixunssfsi()

bblink.c:: ++jrb
    new file to support gcc -tcov (-tcov is not working as yet, but
    when it does, we are ready!)

mincl:: ++jrb
    add targets for above

doprnt.c:: michal/++jrb
    had the same problem as osbind for -fomit-frame-pointer, in ICONV
    change constraints for stuff that is pushed before the sp is reset
    from "g" to "r".
    Also, michal optimized ICONV a bit to do the long div only while
    the number > 65535.

signal.h sysvars.h
div.c doprnt.c gmon.c main.c sbrk.c system.c: Michal
    shut off gcc 2.X warnings.

***************  RENAME     **********************************************

    rename file _floatsi.cpp to _fltsi.cpp

**************************************************************************


**********************    REMOVE ******************************************
        bzero.c, memset.c. memcpy.s
*************************************************************************

bcopy.s, bzero.cpp:: alexander lehmann
    efficient replacements for bcopy/bzero/memmove/memset/memcpy

bcopy.s:: ++jrb
    minor adjustment.

mincl:: ++jrb
    adjust targets for above

fread.c:: ++jrb
    remove redundant cast in 'data=(char*)data + m;' since data
    is a char * now.


_cmpdf2.s _cmpsf2.s  _fixsfsi.cpp _fltsisf.cpp gnulib2.c:: andreas
    * gnulib2.c: nothing needed for gcc-2.0 except __fixsfsi and
    __floatsisf
    * _fixsfsi.cpp, _fltsisf.cpp: new files
    * _cmpdf2.s, _cmpsf2.s: new entry points for gcc2

    in gnulib2.c i #if 0'ed out the not used portions instead of deleting
    them, just so that we have a handy cross reference around. also note
    that at the bottom there is still one new gcc 2.X support routine
    around.  ++jrb

_fixdfsi.cpp:: andreas
    some cleanups

_floatsi.cpp:: andreas
     code for __M68881__ was bogus (is this needed anyways?)

alloca.s:: andreas
    remove some wasted bytes

console.c:: andreas
    (raw_in): allow typeahead (which is only available at GEMDOS level).

doprnt.c:: andreas
    -- remove some misuse of NULL; handle %g correctly.
    -- a version of iconv for -m68020 and also for non __GNUC__
        -- i merged in earlier iconv change, with one diff: when
    using the long div, make d1 come out into 'i' by declaring
    d1 as the reggie for i as andreas had done in his change, rather
    than use a movw d1, %1 as i had it earlier   ++jrb

getenv.c putenv.c :: andreas
    * getenv.c (putenv): put it in separate file as it is independent
    of getenv
    * putenv.c: new file

getuid.c unistd.h:: andreas
    * getuid.c: use uid_t and gid_t
    * unistd.h: ditto
    note that sete?[gu]id() has int as parameter type so that it
    matches the old-style definition

mkdir.c:: andreas
    * mkdir.c: stat() on an already unx2dos'd filename is a no-no (and
    also wastefull), use Fattrib() instead

read.c:: andreas
    (_delchar, str_length): handle ^@ (^@ == (int)0 ++jrb) correctly

rmdir.c:: andreas
    rewrite symbolic directory if dir was nonempty

scanf.c :: andreas
    improve handling of character class

stat.c:: andreas
    allow symbolic link to a deleted path

strcoll.c string.h :: andreas
    * strcoll.c: new file, implementation of strcoll and strxfrm that
    doesn't know about locales
    * string.h: add proto for strcoll and strxfrm

strftime.c:: andreas
    echo unknown format char ala printf (otherwise
    buf is uninitialized)

mincl: andreas
    add target for putenv.o
    mincl: add taget for strcoll.o
    adjust gnulib2 targets:
        add targets for _fixsfsi.o, _fltsisf.o
        remove targets for _lshrsi3.o,...,_lesf2.o

compiler.h float.h signal.h stddef.h stdio.h :: andreas
    avoid warning `cast to pointer from integer of different size'
    if -mshort.

    This is 'fixed' in gcc 2.1 (actually we had the fix for gcc 2.0 PL 2)
    to not issue this warning for 0 being promoted. but andreas's changes
    are applicable in any case. ++jrb

unx2dos.c:: ++jrb/ers/michal
    change default unixmode (if not user specified) from "/" to "/d"
    this takes care of problems dues to caonicalization of PATH to
    posix form, where D:\foo is mapped internally to "/dev/D/foo. if
    there was no 'd' in the default unixmode, then a casual
    user would be surprised why non-spawning application could'nt
    find stuff in such PATH's (like when using findfile).

---------------------------- Patchlevel 80 ---------------------------

bcopy.s:: michal
    I looked at this new bcopy from the latest update and decided to check
    if it is faster only theoretically or maybe also in practice.  Reason
    beeing that my friend claimed that he tried a similar code on TT and
    after much coding it was actually slower.  So I got curious.  Well, at
    least on my Mega it is faster.  For some cases significantly so.
    Although nearly twice as big as the previous one.  Still I found that
    a small change speeds it up BUT in my experiments only for misaligned
    data.  In such situations I got up to 10% shorter times for a little
    bit longer data (break-even seems to be around 25 bytes; for
    smaller sizes the modified code is a tad slower but differences in timing
    are truly neglibible).  I still wonder why a similar change in a loop
    control for aligned data did not bring any real speed variations.

    Another surprise - two tests with branches for the last three
    bytes are actually better than a tight dbra loop.  This may look
    quite a bit different on a small instruction cache of 68020.

    [ jrb: yes, also it will be interesting to see the difference on
    68k's with loop-mode (is the 010 the only one??)].

localtim.c:: ++jrb
    rename dst to _dst and make it non-static. timeoday.c uses it.

ctype.c:: ++jrb
    break out toupper(), tolower() into seperate file so that
    users can re-define these and yet pull in ctype stuff, whout
    clashing.  (me2 does this).

toxxx.c:: +=jrb
    new file comtaining toupper()/tolower()

mincl:: ++jrb
    add target for toxxx.o

mintbind.h:: ++jrb
    add def for Psigpause() per MiNT 0.95 pl 1

osbind.h: ers
    Bconmap should be XBIOS call 0x2c, *not* 0x2b.
    0x2b is DMAwrite. There's obviously a potential for serious
    lossage here :-(. (mntlib has the same bug)

doprnt.c:: ers
    -- conditionalize prototypes so Sozobon can compile this
    -- use "+Inf", "-Inf", "NaN" instead of " Infinity ",
        " -Infinity ", and " Not A Number ". I'm told that
        ANSI specifies the shorter strings, and some programs
        rely on these.

*.h and some *.c:: ers/++jrb
    merge mntlib and lib headers. a few .c files needed adjustments
        because of this.

obstack.c/obstack.h:: ++jrb
    sanitize the ones from gcc 2.1 distribution for 16/32 ints and
        add to lib.

mincl:: ++jrb
    add target for obstack.c

bzero/bcmp/bcopy string.h:: ++jrb
        get rid of the silly old lbXXX aliases.
        add the MiNT _bxxx() aliases. should be useful with people like
        the soz 1.33 author who dont distinguish between an int and the size
        of an addressable object (under the guise of efficiency). big sigh!

malloc.c:: ++jrb
    split off realloc and calloc into their own files. put common
    definitions in lib.h so that it may be included in realloc.c
    and calloc.c.

realloc.c/calloc.c:: ++jrb
    new files.

mincl:: ++jrb
    add targets for realloc.o and calloc.o

lib.h:: ++jrb
    add common alloc defs
    cleanup

mincl: ++jrb
    in all the rules, change any explicitly specified source filename
    to $<, so that when VPATH is set the correct thing happens.

---------------------------- Patchlevel 80.5 ---------------------------
*:: ers/++jrb
    start of merge of mnt/tos libs. all headers done. lots
    of common files done. lots of minor adjustments all
    over.

console.c:: scott
        We were doning a dev=LOOKUP(handle) in console_write_byte, than again
        in flush_key_buff()

eprintf.c:: ++jrb
    turn say into a routine that calculates the strlen. there was already
    a mistake there.

---------------------------- Patchlevel 81 ---------------------------

keycode.h, unistd.h:: michal
    change #if defined() to #ifdef

*.h:: ++jrb
    make all decls for alloca consistent

getlogin.c, lseek.c, qsort.c, gnulib2.c, stat.c:: ++jrb
    #include <memory.h> for alloca.

all over:: ers, ++jrb
    change unx2dos/dos2unx to _unx2dos/_dos2unx to avoid posix names
    space corruption in the library. not that there is still a user
    callable unx2dos/dos2unx.

atol.c:: ers
    conditionalize proto

getuid.c:: ers
    let default uid/gid be root/bin

lib.h:: ers
    adjust protos

gnulib2.c:: alexander lehmann
    _fixunssfsi(): I did some investigating and discovered that the
    function did almost always return 0xffffffff. However the function
    doesn't look wrong in any sense. I rewrote the function a bit
    (without really changing anything) and now it works. The problem arises
    from a bug in GCC, that causes certain negative integer constants to
    become positive when cast to float. e.g. -2^31 (LONG_MIN) becomes
    2^31. This constant was used in the _fixunssfsi function and caused the
    problem.

---------------------------- Patchlevel 82 ---------------------------

crt0.c:: ++jrb
    add extern references to ident strings for all libraries
    so that the libraries that are linked in have their ident
    strings linked into the executable. (so that ident <executable>
    reveals all Patchlevels of all things linked in).
    For satisfying the references of the libraries that are not linked
    we include dummy symbols in gnu.olb.

il.c ic.c iw.c ig.c ip.c:: ++jrb
    the files containing dummies.

mincl:: ++jrb
    add targets for above.

limits.h: alexander
    I did some further investigating into the LONG_MIN problem and
        I found out why the native gcc on my NeXT did create correct
    code while the ST version didn't. The new gcc installs a limits.h
    file that first includes the system limits.h and the redefines
    some of the constants. LONG_MIN is defined as -LONG_MAX-1 which
    is of course equal to -2<<31, but somehow compiles correctly.
    I changed the ST limits.h defined the constant the same way and I
        added the definition for long long limits from the gcc2.x limits.h.
        This will probably be never used, but it can't hurt, I guess.

stricmp.c strnicmp.c:: alexander
    case-insensative verions of strcmp and strncmp. good for
    TC compatibility.

mincl:: ++jrb
    add targets for above

string.h:: ++jrb
    add protos for above.

binmode.c:: ++jrb
    pull __default_mode__ out into its own module.

defmode.c:: ++jrb
    new file, defining __default_mode__, so that it may be pulled in
    from here when undefined by user. previously it was
    bundled in with binmode, that caused binmode() to be pulled
    in along with it.

mincl:: ++jrb
    add target for  above.

fwrite.c:: ++jrb
    when writing out in text mode was returning len+#of CR's in
    buffer, instead of just len of buffer. this showed up with
    ghostscript 2.41. To fix this, we compensate by decrementing
    the counter `l' when we add a CR  since `l' is incremented
    at the bottom of the loop. We could have also fixed by checking
    `wrote_cr' before incrementing `l' but we dont because mostly
    this condition would be false and we would have more overhead
    in the common case.

---------------------------- Patchlevel 83 ---------------------------

ioctl.h, signal.h:: ers
    sync with mntlib changes.

main.c, ctime.c, getuid.c, lib.h, ltoa.c, sprintf.c:: ers
    sync with mntlib changes

frwbin.c:: ++jrb
    new file containing the default __FRW_BIN__ flag. this flag controls
    the default  behaviour fread/fwrite:
      when non-zero, fread/fwrite will not do CR processing
      even for text mode files (ie: ignore file mode).

      when zero fread/fwrite is process according to file mode

    this file is linked in when the user does not specify the variable.

fread.c, fwrite.c:: ++jrb
    incorporate flag.

mincl:: ++jrb
    add target for frwbin.o

compiler.h dirent.h float.h math.h mintbind.h osbind.h
ostruct.h regexp.h setjmp.h stdarg.h tcfloat.h tcmath.h
unistd.h varargs.h :: ++jrb/ers
    sync up with mntlib 21_2, mainly changes for Turbo/Pure C

access.c alphasor.c atof.c bcmp.c bsearch.c calloc.c chmod.c doprnt.c
findfile.c flonum.h fopen.c fseek.c ftw.c getbuf.c getenv.c getlogin.c
getopt.c getpid.c getpw.c getuid.c gnulib2.c grp.c lib.h localtim.c lseek.c
ltoa.c malloc.c putenv.c qsort.c random.c realloc.c  regexp.c sbrk.c setbuf.c
setvbuf.c strcat.c strchr.c strftime.c strncat.c strncpy.c strrchr.c strtol.c
strtoul.c system.c sysvar.c time.c timeoday.c tmpfile.c tmpnam.c utime.c
getcwd.c isatty.c spawnve.c sleep.c :: ++jrb/ers
    sync up with mntlib 21_2, mainly changes for Turbo/Pure C. nice thing
    about this merge and the earlier soz/lcc merges is that a lot of
    lint cleaning took place as a result.

compiler.h ioctl.h unistd.h signal.h setjmp.h osbind.h mintbind.h
math.h keycodes.h :: ++jrb/ers
    sync with mntlib 21_3

perror.c:: ++jrb/ers
    sync with mntlib 21_3

scanf.c:: michal
    sscanf (test,"%%Page: %d %d\n", &i1, &i2);
    after "%%" *fc equals 'P' but c was never updated and is still
    equal '%'.  Therefore test on line 232 fails and no conversions will
    be performed. [the fix was to get the next char into c after testing
    for '%' - ++jrb]
    thanks michal!

*.s, *.cpp, linea.c crto.c:: howard
    changes to support baserel code generation.

Makefile, mincl:: howard
     New integrated scheme elimiating makefile.16 makefile.32.
    thanks howard!

maikefile.16, makefile.32 :: ++jrb
    delete, no longer needed.

mintbind.h obsind.h obstack.h
doprnt.c getpass.c linea.c malloc.c obstack.c regexp.c
sbrk.c stricmp.c strnicmp.c strrchr.c system.c textio.c :: ers/+=jrb
    more cleanups from mntlib pl 22 (casts etc from compilers that
    are more fussy [some really overkill]).

compiler.h mintbind.h:: ers/++jrb
    sync up with mntlib pl 23

difftime.c:: uunet!gnu.ai.mit.edu!entropy
    operands were backwards

doprnt.c:: Michael ([e-mail])
  the g-format does not work correctly. In many circumstances, the output is
  a huge fp-number which is correct in all the 15 significant digits; but the
  length of the mantissa is roughly doubled and filled with garbage. This
  seems to be rounding problem - it occurs more frequently with the 68881,
  but only when the significant rest of the mantissa is a sequence of zeros.
  I observed it with  %g %1.6g %1.14g %1.15g.

  In my opinion, the g-formatter of cvt() and round() should be rewritten.
  As i don't have the time doing this, i hacked a kludgy fix:
  I save the original no of digits, keep track of the '.', and cutoff the
  buffer before the trailing zeros are stripped. At the same time, i supress
  unreasonable precision, i.e., digs = min( digs, 15 ).

doprnt.c:: Michael/++jrb
    Michael had a problem with ICONV when doprnt.o is linked
    from outside gnu.olb. (i have'nt figured out why this
    would make a difference). Anyways now there is a
    !defined(__NO_INLINE__) around _ICONV so that the C version
    may be picked istead of the inline asm version.

---------------------------- Patchlevel 84 ---------------------------

doprnt.c:: ++jrb
    Revert out michaels 15 sig digit hack. causing too many problems
    (michal jaegermann reports all kinds of failures from gawk, mawk).
    Revert out DBL_EPSILON hack too..
    Awaitings michals advise on how to fix mjr's problems...

_addsub*:: Olaff ([e-mail])
    To summarize the changes in both double and float arithmetic:

    * The "sticky" byte d2.b is now initalized properly. In the former
    version it was a deterministic random generator.

    * The rounding is now computed correctly. The rounding bits reside in
    d3.b.

    * Due to cancellation effects one get one extra bit. So a conditional
    jump had to be modified.

    One extra bug was included in the single precision arithmetic.

    * When the routine discovered that shifting can be done by swapping, it
    shifted the result one  additional digit.

_norm*:: Olaff ([e-mail])
    round up towards even.

bcopy.s, bzero.cpp: andreas
    dont use a4 (for -mbaserel)

linea.h:: ++jrb
    linea2() really needs to declare its output (i know its implicit, but
         unless you put it in the output list, g++ thinks linea2() does
    not output anthing, and complains on its usage).

doprnt.c:: ++jrb
    change condition for unreasonable precision for floats. (it should not
    have been MAXFRACT only as we do zero padding on the right).

regexp.c :: ++jrb
    - include <sys/types.h> instread of <types.h>
    - fix cast in malloc from (unsigned)  to (size_t) as it should be.

regexp.c:: ++jrb
    in regmatch(), there was over-enthusuatic conversion of
    int to size_t that caused regular expressions to break.
    please be careful!

    in the PLUS/STAR case, `no' needs to be a signed entity, other
    wise the loop is useless.

    tregex in the library tests showed this up.

doprnt.c:: ++jrb
    constraint ("g") on the source operand was wrong for divu. source
    has to be  data addressable ("dmi" constrain should cover).

---------------------------- Patchlevel 85 ---------------------------

_divdf3.cpp:: olaf
   prevent the tie rounding case if dividing is not exact.
     paranoia now says: "Division appears to round correctly"
     ** requires _normdf Version 1.6.1 or later

_divmods.s _mulsi3.s _udivmod.s:: ++jrb
    add c68 labels

stdarg.h/varargs.h:: uunet!ms2.maus.de!Bjarne_Pohlers (Bjarne Pohlers)/++jrb
    bjarne reported a problem with an example in the way we were
    pull off args from the stack. he had a fix that looked good. i
    went the gcc-way for the same fix, and used __builtin_next_arg()
    and let the compiler do the right thing. thanks bjarne.!

_addsubd.cpp _addsubs.cpp _divsf3.cpp _normsf.cpp: Olaf
    Problems solved:
      * tie breaking for single precision division
      * prevent core dump for single precision overflow

    Only one flaw is left for single precision arithmetic!

    I did a rewrite of the add/sub algorithm; now it is about 10-20% faster.

falcon.h: Howard
    bindings for Falcon. Thanks howard.

fread.c:: Andreas
    With the BSD-like pipes in MiNT 0.96/PL14 one cannot assume, EOF is
    reached if read() returns less bytes than requested; it could be a
    short read from a pipe.

    There's another `feature' in fread: if a big block is read in text
    mode (!_IOBIN) and the data actually doesn't contain CR's, fread ends
    up copying the data block onto itself.

Synch with mntlib Patchlevel 25: ers
    ioctl.h:
        ANYP means "no parity", not "any speed"

    difftime.c: [e-mail]
        The parameters were reversed. Ouch.
    bcmp.c: [e-mail]
        Change the ODD macro to return a 16 bit value (for
        Pure C)

    Following from andreas:
        compiler.h: for gcc2, use the builtin sizes
        gemfast.h: add XCONTROL messages and WF_COLOR defines
        mintbind.h, osbind.h: add the missing definitions
        screen.h: fix definitions for TT
        stdio.h: allow 32 open files as does MiNT
        wait.h (WSTOPPED): use expected value

---------------------------- Patchlevel 86 ---------------------------

_cmpsf2.s _divsf3.cpp _extends.cpp _mulsf3.cpp
_negsf2.s _normsf.cpp : olaf
    completed impl of NaN and infinities, even for floats.
    thanks olaf!

atof.c, doprnt.c:: andreas
    I have implemented improved floating point formatting/scanning
    routines for the mc68881. They use the builtin packed decimal
    format, which gives a precision of 17 decimal digits. Formatted and
    rescanned numbers now give identical results, and enquire reports no
    problems anymore.

string.h:: ++jrb
    add parens to shut up gcc.

doprnt.c:: Howard Chu
    The patch to doprnt makes it less vulnerable to crashing when handed
    a large floating point value. (Took me a while to figure out why I was
    crashing, around address 0x30303030!  }-)  Basically, my math routine
    returned a value like 0x7ff12345 0x60000000 which is larger than INF
    but less than NaN. The check I put in makes it print this range of
    numbers as NaN or -NaN, which is what the Sun C library does.

falcon.h: ++entropy
    Protected against multiple #include's.

string.h: ++entropy
    Added some parens to appease gcc -Wall with gcc 2.3.1.

getuid.c: [e-mail], ++entropy
    Only negative returns from Psetuid() and Psetgid() indicate
    an error.  Positive returns indicate success (the PID is
    returned from the OS).  Modified getuid() and setgid() accordingly.

a64l.c: [e-mail]
    NEW module for conversion between longs and base-64 ASCII strings.

a64l.c: ++entropy
    Use __PROTO, use compiler.h, etc.

support.h: ++entropy
    Added prototypes for a64l(), l64a().

support.h: [e-mail], ++entropy
    Fixed prototype for link().  Removed protoype for isatty()
    (it exists in unistd.h)

limits.h: ++entropy
    Added #define PASS_MAX 8.

getpass.c: ++entropy
    Fixed to use PASS_MAX, fixed off-by-one-error (was writing past end
    of static buf, very bad news).  Removed some small kludges.

fflush.c: [e-mail], ++entropy
    Return EOF instead of 0 for files that aren't open.

mincl: ++jrb
    add new target for a64l.o

alloca.s, bcopy.s: [e-mail]
    RENAME to alloca.cpp, bcopy.cpp so we can do conditional compilation.

alloca.cpp, bcopy.cpp, bzero.cpp: ++dsb
    Changes for HSC.

mincl:
    Account for the name changes.

compiler.h: ++dsb
    Rearrange the __SOZOBONC__ #defines, fix a typo.

ctime.c, fscanf.c, sscanf.c: ++dsb
    Make compatible with HSC.

compiler.h: ++hohmuth
    define __TCC_GEMLIB__ when used with Pure/Turbo C

aesbind.h, gemfast.h: ++hohmuth
    Make it useable with Pure's GEM library.  The special declarations
    in aesbind.h check for __TCC_GEMLIB__.  Improved compatibility with
    Turbo's GEM declarations when __TCC_COMPAT__ is defined.

osbind.h: ++hohmuth
    Fix typos, add macros for sloppier type checking when used with
    Pure C.

strupr.c: ++hohmuth
    NEW file:  add library call strupr().

string.h: ++hohmuth
    Add declaration of strupr().

mincl, lattice/mincl.lcc, purec/mintlib.prj, sozobon/makefile:
    Add strupr.c.

doprnt.c: [e-mail]
    Fix up the ARG macro (which tends to produce wrong results with Pure
    C (which uses a type conversion scheme within ?: that differs from
    GCC)).

regexp.c: ++hohmuth
    Make it Turbo/PureC-proof.

ctime.c: ++hohmuth

falcon.h: [e-mail]
    Remove superfluous equals sign in enum declaration.

fread.c: ++ulf
    Fix return value for text mode.

times.c, times.h: ++ulf
    Use long return value instead of int.  This allows larger values
    and agrees with the SysV man page. Returns realtime since epoch
    like sysV (rather than 0,1,-1 like bsd as it previously did)

strlwr/strupr.c:: ++jrb
    bulletproof a little.

a64l.c: [e-mail]
    Add to comments a note that prototypes are in support.h.

getpass.c:
    Read password from /dev/tty, instead of stdin, if MiNT is active.

support.h:
    Functions link(), symlink() now take "const char *" instead of
    "char *" parmeters.

access.c: ++boender, ++entropy
    Use real, not effective, user id's for access() checks.
    Fix so access() always returns success on existing files if real
    user id is 0 (root), regardless of file mode.

mintbind.h:
    Add new GEMDOS functions (macros) Dgetcwd() and Salert().

errno.h: ++boender
    Define ENODEV as alias for EUKDEV/EUNDEV.


---------------------------- Patchlevel 87 ---------------------------

support.h:: warwick
    s/new/newname/  otherwise it clashes with the new op of c++

findfile.c:: warwick
    "findfile" doesn't work if the path is "\" (or "/", or "\\'),
    while "E:\foo" works fine.  Can it handle trailing slashes?

_addsubd.cpp _addsubs.cpp _divdf3.cpp _divsf3.cpp _extends.cpp
_fixdfsi.cpp _fixsfsi.cpp _fxunsd.cpp _muldf3.cpp _mulsf3.cpp
_truncdf.cpp: andreas
    All floating point routines: my 68000 book says that bset #bitno,dn is
    faster than orl #imm,dn.  _fixsfsi.cpp: that was also my fault

atof.c:: andreas

ctime.c:: andreas

doprnt.c:: andreas
    Another bug that i introduced in doprnt.c:
    the m68020 version of _ICONV has the arguments of divull reversed

errno.h, strerror:: andreas
    include the new error EPIPE

filbuf.c:: andreas
    clr FILE->cnt on EoF or error.

fopen.c:: andreas
    dont close ttys for MiNT
    get new buffer if file was closed

strftime.c:: andreas
    strftime: typo in case 't', initialize buf if format
    directive is undefined

strtol.c:: andreas
    negating LONG_MIN gives overflow

lib.h, fcntl.h::
    The declaration for __open_stat etc. should be moved to lib.h, where
    the internals of the library belong to.

close.c isatty.c dup.c fcntl.c fhandle.c::
    adjust for above

strupr/lwr:: michal
    use <ctype> tolower/upper instead (in case we ever support 8 bit or
    wide chars, then the EOR trick of course will not work).

main.c:: andreas
    integrate c++ __main() (see changes in gcc 2.3.3 pl 3 (c-decl.c)).
    c++ low level support now moved to this lib from libg++

    define std in/out/err symbols for debugging.


gmon.c:: andreas
    fixup for gcc > 1

gbl-ctors.h, longlong.h: andreas
    new files in common

libgcc2.c:: andreas
    new file in common

osbind.h, falcon.h, mintbind.h:: andreas
    fix up for __GNUC__ > 1. note how the restriction of # of params
    to __asm() as been lifted, and now we use this feature for
    the larger binding. this is very benefecial as we dont have
    to bend over backwards to get a few binding correct (binding in
    which we would change the SP from under the compilers feet).

realloc.c: andreas
    In realloc, a block can only be split in two if there's enough room
    left for a struct mem_chunk (plus a bit more). [+ roundup ++jrb ]

---------------------------- Patchlevel 88 ---------------------------

memcmp.c:: andreas
    should compare with unsigned char

strr?chr.c:: andreas
    should not depend on signedness

memccpy.c and memcmp.c:: andreas
    simplified

math-68881.h:: ++jrb
    sync up with gnu version.

math.h:: ++jrb
    change all #ifdef _M68881 to
     #if defined(_M68881) ||  defined(__M6881__)
    __M68881__ gets automatically defined when gcc -m68881

doprnt.c:: michal
    Still, so far I found two bugs in floating point support.  One
    was introduced by Howard when he tried to catch garbled results from
    floating point operations.  He caught in the process also negative
    zero and that was part of the reason why Scott got NaN.

    [in the part that check for overflows]

_divdf3.cpp:: michal
    The other part was a wrong branch in a division routine which
    was giving a negative zero as a result of a division 0.0/1.0.

_divdf3.cpp, _divsf3.cpp:: michal, olaff
    Fixed sign for retinf, and retzero.

---------------------------- Patchlevel 89 ---------------------------

    michal did a tremendous amount of job tracking down modf and
    fixing the following. thanks michal.

    Wed May 26 20:35:34 1993  Michal Jaegermann  (michal at smok)

    * fixed norm_df and norm_sf not to set errno when processing
      legitimate zero argument
    * fixed wrong error codes and made sure that future changes
      in errno.h will be automatically reflected;
      this involves changes in errbase.h and errno.h
    * removed unused code from _divsf3.cpp
    * corrected bug in setting sign of Inf in _truncdf.cpp
    * made zeros returned from frexp and ldexp consistenly signed
      (they were signed or unsigned, depending on a way they were
       calculated)
    * renamed _cmpdf2.s to _cmpdf2.cpp and _cmpsf2.s to _cmpsf2.cpp
      and added preprocessor directives to split them into separate
      objects while compiling (required change in mincl)
    * replaced 68000 code for modf() with totally new, shorter and
      much faster version
    * added Olaf Flebbe fixes for sign of 0 in multiplication routines
    * general code cleanup all over the place in floating point support
      routines

scanf.c:: ++jrb
    fix bug reported by warwick, when "%Ns", the test for the end of
    the %s loop was incorrect. (left off 1 char too soon).

    added some decls for non ansi platforms.

    use long double type only when __M68881__ not with soft float.
    out soft float stuff knows nothing about long doubles.

---------------------------- Patchlevel 90 ---------------------------

_addsubs.cpp, _addsubd.cpp, _div*.cpp, _mul*.cpp:: michal
    following some discussions with Olaf I rewrote addition routines to
    conform - at least here - strictly to IEEE spec.  As a matter of fact
    they even decreased in size. :-)

    this stuff which I am mailing you right now corrects handling
    various cased when you add or multiply infinities or NaNs.
    To my best knowledge this now follows IEEE specifications, although
    do not quote me on that. :-)  There is also some code rearangement
    in division routines to reduce unnecessary branching.

    thanks michal!

osbind.h:: Markus Gutschke <[e-mail]>
    stack adjustment amount was wrong. was 6 should be 10. thanks
    markus.

osbind.h:: [e-mail]
    wrong stack adjustment for trap_1_wwwll was 16 should have
    been 14. thanks robert.

---------------------------- Patchlevel 91 ---------------------------

pgrp::
    NEWFILE from mintlib 31

    unistd.h:
    Added pause() and sigpause() prototypes.  Changed prototype of
    setpgrp() to reflect new SysV-compatible version, and make proto
    apply only if not _POSIX_SOURCE.  Changed prototype of getpgrp() to
    reflect return type of pid_t (POSIX).  Added setpgid(), setsid(),
    _bsd_setpgrp() and _bsd_getpgrp() protos.  To access the BSD
    functions by their usual names, define the macro _BSD_PGRP (and do
    _not_ define _POSIX_SOURCE) before including unistd.h (for example,
    gcc -D_BSD_PGRP -c foo.c).

    pgrp.c:
    NEW file, for process group related functions. Renamed setpgrp to
    _bsd_setpgrp().  Created new SysV-compatible setpgrp(). Created
    new BSD-compatible _bsd_getpgrp() (requires MiNT 1.03 or newer).
    Created new POSIX setpgid() and setsid().


putpwent.c::
    NEW file from mntlib 31

eprintf.c:: mntlib31
    throw away status buffer. [move buffer onto stk]

findfile.c:: mntlib31
    rehacked by Uwe Ohse, 28.4.93, to support reentrant spawn/popen.

getpw.c:: mntlib31
    sync up with what appears to be a rewrite

grp.c:: mntlib31
    ditto

linea.h, grp.h, ioctl.h, mintbind.h, osbind.h, ostruct.h pwd.h ::
    sync with mntlib 31.

---------------------------- Patchlevel 92 ---------------------------

main.c:: andreas
    the isatty status should be cleared for dumped programs

float.h:: andreas
     update for GCC 2.4/enquire 4.3; the values for
    LDBL_{EPSILON,MIN,MAX} are computed with bc and verified with enquire.

support.h: [e-mail]
    Remove two occurances of the string "new" for C++ compatibility.

ctype.h: ++entropy
    More POSIX namespace monkey business.

types.h: ++entropy
    Made lots of stuff #ifndef _POSIX_SOURCE.  Added a few POSIX types
    (nlink_t, mode_t).

doprnt.c: ++hohmuth
    Make a long constant explicit for Pure C.

setjmp.h: ++entropy
    Implemented BSD-compatible _setjmp(), _longjmp(), setjmp(), and
    longjmp().  Selectable by defining _BSD_SOURCE.  BUG: not compatible
    with __STRICT_ANSI__.

fcntl.h: ++entropy
    Add POSIX-required mask FD_CLOEXEC, to select the close-on-exec
    bit from the flags used in F_GETFD/F_SETFD fcntl().

unistd.h: ++entropy
    Do not define prototype for getopt() if _POSIX_SOURCE is defined.

types.h: ++entropy
    Do not #include <utime.h> if _POSIX_SOURCE is defined.

utime.h: ++entropy
    Added prototype for utime().

unistd.h: ++entropy
    Changed BSD process group selection switch from _BSD_PGRP to
    _BSD_SOURCE.

stat.h: ++nox, ++entropy
    Added S_ISCHR(), S_ISDIR(), S_ISBLK(), S_ISREG(), S_ISFIFO(),
    S_ISMEM(), S_ISLNK(), S_IRWXU, S_IRWXG, S_IRWXO.
    (btw, is S_IMEM a typo of S_IFMEM or was that intentional?)

setjmp.cpp, setjmp.h: ++entropy
    Implemented POSIX sigsetjmp() and setlongjmp().

fopen.c: ++entropy
    Always use O_NOCTTY flag in the call to open().

errno.h, strerror.c: ++entropy
    Add definition of EIO and ENOSPC.

support.h: ++entropy
    Prototype for _isctty() from the MiNT libs.

ctermid.c, stdio.h: ++entropy
    NEW file, POSIX ctermid() function.  Prototype and constant L_ctermid
    in stdio.h.

doprnt.c: [e-mail]
    HSC choked on a string with implicit newlines (that is,
        "like
        this"
    as GCC is wont to do); I replaced it with explicit "\n"'s and string
    concatenation.

fread.c, fwrite.c: ++dsb
    Some judicious changes from size_t to unsigned long to allow reads
    and writes of >=64K with HSC.

getpw.c: ++dsb
    Minor tweak to work around a foible in HSC.

compiler.h: ++dsb
    Changes for HSC v2.00 and higher.

utime.h: ++dsb
    Changed "#endif _UTIME_H" to "#endif /* _UTIME_H */" to keep HSC
    from complaining.

fcntl.h: ++entropy
    O_NOCTTY definition (for the MiNT libs).

pgrp.c: ++entropy
    Made a more reasonable attempt at setsid().  Kludge a way to
    "disassociate" the controlling tty by forcing fd -1 to /dev/null.

fwrite.c:: michal
    With unbuffered IO fwrite miscounts number of characters written.

a64l.c, calloc.c, chdir.c, chmod.c, eprintf.c, grp.c,lib.h, linea.c,
psignal.c, putenv.c, raise.c, random.c, regexp.c, regsup.c, strlwr.c,
strrev.c, strupr.c, strvar.c, textio.c, timeoday.c, *.h:: entropy
    sync up with mntlib pl 33. most changes are cleanups.
    (what does linea2 return?)

--- sync with mntlib34:
ffs.c: ++frog
    Much faster algorithm.

Makefile, _fixsfsi.cpp, _fltsisf.cpp,
frexp.cpp, modf.cpp : [e-mail] (Stefan Steyer)
    Bug fixes for SFP004 coprocessor code.  Add sfp004 targets.

doprnt.c: [e-mail] (Andreas Schwab)
    The decimal precision in a printf format does not include the sign or
    hex prefix, e.g. printf ("%-2.5d", -18) should print "-00018".

time.h: [e-mail]
    Add prototype of select() to time.h (is there a better place?)

ctype.h, grp.h, dirent.h, stddef.h, string.h: ++entropy
    Cleaned up for POSIX.

wait.h, pwd.h: ++entropy
    Cleaned up for POSIX (need more work).

compiler.h, limits.h, stdio.h: ++entropy
    Set stream limits for POSIX.

locale.h: ++entropy
    Define NULL if not already defined (POSIX).

limits.h: ++entropy
    Raise value of _POSIX_NGROUPS_MAX from 0 to arbitrary value of
    512 now that we have a decent getgroups().

fcntl.h, errno.h: ++entropy
    Add definition of F_SETLKW to fcntl.h.  Add ELOCKED to errno.h.
    If Fcntl() gets ELOCKED, return EACCES instead (POSIX).

unistd.h, stat.h, fcntl.h: ++entropy
    Moved many prototypes to the correct headers.

signal.h: ++entropy
    Fix some violations of POSIX namespace.

-- end sync

findfile.c:: michal
    try to accomodate tos style paths
    PATH=e:\bin,c:\bin,...

unistd.h:: ++jrb
    take out setlinebuf proto, it does'nt really belong here.

---------------------------- Patchlevel 93 ---------------------------

falcon.h:: howard
    corrections from howard

math-68881.h:: ++jrb
    -- Put extern "C" around file when __cplusplus
    -- Make hypot use internal version of sqrt (because signature
       clash in libg++:xfix )
    -- Define internal version (_sqrt) of sqrt.

osbind.h: [e-mail]
    chand addw #n,sp to lea n(s),sp which is 4 cycles shorter

dirent.c::  Christoph Koerner ([e-mail])
    sync up dirent.c with changes in dirent.h (DIRENTSIZ -> __DIRENTSIZ)

fprintf.c:: ++jrb
    break out vfprintf, printf and vprintf into their own .o's. that
    way the user can replace any of them.

printf.c, vfprintf.c, vprintf.c:: ++jrb
    NEW files

mincl:: ++jrb
    add targets for above

gnulib2.c:: Steven Ourada ([e-mail])
    #define WORD_BIG_ENDIAN for long long.

sync with mntlib35:

doprnt.c: [e-mail]
    Add support for printing/scanning long longs.

div.c: ++pvt1-117
    Add div functions for compilers other than GCC.

bsearch.c: ++dsb
    Fix bug reported by Thorsten Roskowetz: the order arguments to
    bsearch()'s comparison function was reversed.

ioctl.h: [e-mail] (Jan-Hinrich Fessel)
    Make it possible to include both ioctl.h and filesys.h.

pgrp.c: ++schwab
    Correctly check the MiNT version in _bsd_getpgrp().

doprnt.c: ++pvt1-117
    Changes to keep Lattice C from generating annoying warnings.

math.h, process.h, support.h, unistd.h, findfile.c, textio.c: ++schwab
    Add 'const' to declarations where appropriate.

malloc.c:  ++schwab
    Make malloc() round all memory requests to the page size to be able
    to use all the available memory.

setjmp.cpp, setjmp.h: ++schwab
    Some cleanup; one element of jmp_buf[] can be saved by using the
    fact that signal 0 cannot be masked; why was the type char*[] insead
    of long[]?

fopen.c: ++schwab
    Make fopen() use 0666 as the default file mask (use umask() if you
    want something else); make mkdir() respect the umask setting.

utime.c: [e-mail]
    Pass through a NULL tset argument unchanged to the filesystem
    so that it can determine permissions correctly.

falcon.h: [e-mail], [e-mail]
    Change several erroneous macros to correctly cast return to short.

errno.h: ++Ole_Arndt
    Add errors ENOTDIR and ELOOP.

signal.h: [e-mail] (Ole Arndt)
    Fix definition of _SIGSET_MAX_INDEX to correct several bugs
    in the POSIX signal mask functions.

stdio.h:
    Add vsscanf() and _getbuf() protos (yech).

dirent.h:
    Add alphasort() proto (excluded by _POSIX_SOURCE).

utmp.h:
    Rename write_utmp() to _write_utmp().  Rename write_wtmp() to
    _write_wtmp().  Add protos (and other cleanups) to utmp.h.

atof.c, fdopen.c, fputs.c:
    Move 'register' to beginning of declarations (shuts up gcc -Wall).

fwrite.c: ++michal
        With unbuffered IO fwrite miscounts number of characters written.

findfile.c: ++michal
    Try to accomodate tos style paths PATH=e:\bin,c:\bin,...

unistd.h: ++jrb
    Take out setlinebuf, doesn't really belong here.

ffs.c: [e-mail] (Michal Jaegermann)
    An even faster algorithm.  Yow!  Are we optimized yet?

support.h:
    Added _exit() and gethostname() protos. Removed write_utmp() and
    write_wtmp() protos.

lib.h, crtinit.c:
    Moved prototypes from crtinit.c to lib.h, and added more to lib.h.

abort.c, lib.h:
    Make abort() close file handles in exactly the same manner
    as exit().

signal.h: [e-mail], ++Oskar, ++entropy
    Fix type of sa_mask in struct sigaction (sigset_t, not long).

falcon.h: [e-mail], ++entropy
    Add missing traps.

unistd.h, support.h:
    Cleaned unistd.h for easier reading.  Moved back the link(),
    symlink(), and readlink() protos from support.h (sorry if I seem
    really indecisive on where some of these protos belong).

doprnt.c, open.c, putenv.c, utime.c:
    Include appropriate headers for protos (and fixed mknod() proto).

sys/statfs.h:
    Add statfs() proto.

sys/errno.h, sys/ioctl.h, sys/resource.h, sys/signal.h, sys/wait.h:
    Fix multiple-include protection to hopefully prevent possible
    infinite include loops (see Bugs file).

getpw.c, pwd.h: [e-mail] (Oskar)
    Use uid_t (_UID_T) for getpwuid() instead of int (POSIX).

=== end of merge

symlink.c, spawnve.c, spawnvp.c, exec.c: ++jrb
    adjust args to meet new protos in process.h, unistd.h

getpage.c:: ++jrb
    NEW file. defines getpagesize() (for use with malloc.c etc)

mincl:: ++jrb
    add proto for above

== merge with mintlib 36
_addsubs.cpp: ++nox
    Correct an inverted branch.
utime.c: ++schwab
    utime(NULL) uses time() to convert the current time to Unix format
    and immediately converts it back to GEMDOS format. Instead it should
    use the GEMDOS format directly.
    stime() must be changed because Tset{date,time} can now return a
    real error number. This depends on Tset{date,time} returning -1 if
    args are bad, which is true for all TOS versions, i think.
falcon.h: [e-mail]
    Add two omitted backslashes to fix new trap definitions.
chmod.c: [e-mail]
    chmod() unnecessarily calls stat() to check if the file is a
    directory; this information is already available through Fattrib().
nlist.c, nlist.h: [e-mail]
    NEW files nlist.c, nlist.h.  A version of nlist() for mntlib/toslib.
    Works OK with my binary config program but your milage may vary.
wcmb.c, stdlib.h: [e-mail]
    NEW file wcmb.c for ANSI wide char functions.
ctype.h: [e-mail]
    The toint() macro in ctype.h was translating hex digits wrong.

== end merge

mincl:: ++jrb
    add tagets for nlist and wcmb

== merge with mntlib 37

bzero.cpp, mincl: ++entropy
    Break bzero and memset into separate .o's (reduce drag).

string.h, strcmp.c: ++pvt1-117
    Allow use of builtin string functions.

file.h: ++entropy
    Add FREAD and FWRITE macros for TIOCFLUSH flushing modes (not yet
    implemented in MiNT).

unistd.h: ++entropy
    Add tcgetpgrp() and tcsetpgrp() protos.

ioctl.h: ++entropy
    Changes to account for definitions in common with termios.h,
    also add _RTSCTS flag definition.

termios.h: ++entropy
    NEW files, partially implementing the POSIX termios functions (in
    the MiNT library).

wait.h: ++entropy
    Add WSTOPSIG() macro (POSIX).

utime.c: ++entropy
    Add a kludge so that we pretend the utime() worked for directories.

stat.h: ++entropy
    Mode argument is now mode_t instead of unsigned (POSIX).

mkfifo.c, stat.h: ++entropy
    NEW file providing a fake mkfifo(), and a prototype for it.

resource.h: ++entropy
    Add BSD-compatible struct rusage members with 0 values for
    minimal compatibility.

errno.h: ++entropy
    Added EOPNOTSUPP as (temporary) alias for EINVAL.
    Added ECHILD as alias for ENOENT (POSIX).

doprnt.c, fprintf.c, printf.c, sprintf.c,
vfprintf.c, vfprintf.c, lib.h: ++schwab
    Change the interface of _doprnt to also pass a putchar function
    similar to _scanf.  This allows use of sprintf in a program which
    doesn't use stdio and it simplifies the implementation of sprintf a
    bit.

lseek.c: [e-mail]
    Make lseek() and tell() always return -1 on errors.  Make lseek()
    work better with unseekable devices.  Don't modify errno
    when no error has occurred.  General cleanup.

nlist.c: ++pvt1-117
    Add some typecasts to keep Lattice happy.

mintbind.h: [e-mail]
    Add some casts in the Lattice C inlines.

frexp.cpp: [e-mail]
    Bug fix: add a missing @.

ctime.c, doprnt.c, findfile.c, getpass.c,

localtim.c, strftime.c, strncat.c: [e-mail]
    Removed some dead assignments and increments.

stdlib.h: ++entropy
    Added NULL (POSIX).

== end merge

mincl:: ++jrb
    add new rules for bzero/memset
    new target for (dummy) mkfifo

== merge with mntlib 38

compiler.h: ++pvt1-117
    Make LCC do the right thing in limits.h.

termios.h: ++entropy
    Add VLNEXT and VWERASE control chars.

types.h: [e-mail]
    Add fd_set type and related macros.

falcon.h: [e-mail]
    Change return types of VsetMask() and Dsp_Reserve() from
    void to short.

_mulsi3.s, _umulsi3.s: ++schwab
    .lmul and .ulmul aren't equivalent.

wait.h: [e-mail]
    Make the wait prototype indirection more robust.  Also fix the
    definition of WIFSIGNALED().

falcon.h: ++pvt1-117
    Inline traps for Lattice.

wcmb.c: ++pvt1-117
    Use _wnull from wnull.c instead of a local variable.

wcscat.c, wcscmp.c, wcscpy.c, wcslen.c, wnull.c: ++pvt1-117
    NEW files for ISO wide char support.

stdlib.h: [e-mail]
    Add prototypes for new ISO widechar functions.
== end merge

mincl:: ++jrb
    add targets for wcscat.o, wcscmp.o, wcscpy.o, wcslen.o, wnull.o

abort.c:: ++jrb
    rid of redundant fclose_all_files(), exit() should do all that.

limits.h:: ++jrb at suggestion of michal
    make groups limit 0 for non-MiNT

---------------------------- Patchlevel 94 ---------------------------
=== synch with mntlib39

***** include/errno.h
----------------------------
revision 38.2
date: 1993/10/21 16:21:30;  author: entropy;  state: Exp;  lines: +1 -0
Add ENOTEMPTY as alias for EEXIST (POSIX).
=============================================================================
***** include/limits.h
----------------------------
revision 38.3
date: 1993/10/23 18:44:56;  author: entropy;  state: Exp;  lines: +17 -18
Raised some arbitrary limits (or changed to actual limit, as in the case
of _POSIX_OPEN_MAX).  Rearranged things a bit for consistency.
----------------------------
revision 38.2
date: 1993/10/23 18:28:52;  author: entropy;  state: Exp;  lines: +4 -0
Set _POSIX_NGROUPS_MAX to 0 when not __MINT__.
Thanks to [e-mail] for the bug report.
=============================================================================
***** include/time.h
----------------------------
revision 38.2
date: 1993/10/25 10:00:30;  author: entropy;  state: Exp;  lines: +7 -2
Update select() prototype to use fd_set * for arguments, and
make some stuff !_POSIX_SOURCE.
=============================================================================
***** include/types.h
----------------------------
revision 38.2
date: 1993/10/25 09:59:28;  author: entropy;  state: Exp;  lines: +4 -1
Allow fd_set definition to be gotten from time.h too.
=============================================================================
***** include/wait.h
----------------------------
revision 38.2
date: 1993/10/24 18:47:48;  author: nox;  state: Exp;  lines: +1 -1
Some older versions of GCC trip over the new wait macros...so
it's better not to use them for now.
=============================================================================

== end sync

read.c:: michal
    replace the echochar()/delchar() macros with safer ones, and
    ones that dont solicit warnings from the compiler.

mincl:: michal
    build bzero/memset together when building bzero.o. doing
    them seperately saves only 12 bytes, but causes jump
    to become long (again not that much of a big deal, as
    those jumps happen once)

osbind.h: ++jrb
    relax the constraints on the inputs of trap_14_wwwwwww (only
    Rsconf maps to this)  to "g" from "r", as these many "r" 's
    give gcc 2.>3.X heartaches (understandably). note this is ok
    since these args will never be expressions, and we never
    have to constrain hard enough to force eval before we change
    sp from underneath gcc.

qsort.c: ++jrb
    adjust order of inclusion of compiler.h so that INLINE may be
    conditionally overriden even when __GNUC__

== sync with mntlib 40

***** abort.c
----------------------------
revision 39.2
date: 1993/10/28 17:54:14;  author: entropy;  state: Exp;  lines: +20 -2
Fix for the TOS lib, and also remove a POSIX violation.
=============================================================================
***** utime.c
----------------------------
revision 39.2
date: 1993/10/27 15:01:02;  author: entropy;  state: Exp;  lines: +16 -1
Implement the Dcntl() versionof FUTIME in utime().  Comment out the
kludge for directories in the Fcntl() version.

==end sync

mincl: ++jrb
    undo the above change for memset, keep them seperate as there
    are util there that want one without the other. what the heck, its
    only about a  dozen bytes.

==sync with mntlib 41
***** nlist.c
----------------------------
revision 40.2
date: 1993/10/31 09:24:14;  author: hohmuth;  state: Exp;  lines: +3 -3
add a few typecasts for __MSHORT__
=============================================================================
***** strerror.c
----------------------------
revision 40.2
date: 1993/10/31 11:33:36;  author: entropy;  state: Exp;  lines: +59 -1
Add support for the socket lib.
=============================================================================
***** utime.c
----------------------------
revision 40.2
date: 1993/10/31 09:25:04;  author: hohmuth;  state: Exp;  lines: +1 -1
make assignments used as truth values clearer, to avoid
spurious warnings
=============================================================================
***** wcmb.c
----------------------------
revision 40.2
date: 1993/10/31 09:25:16;  author: hohmuth;  state: Exp;  lines: +2 -2
make assignments used as truth values clearer, to avoid
spurious warnings
=============================================================================
***** wcscat.c
----------------------------
revision 40.2
date: 1993/10/31 09:25:24;  author: hohmuth;  state: Exp;  lines: +2 -2
make assignments used as truth values clearer, to avoid
spurious warnings
=============================================================================
***** include/PatchLev.h
----------------------------
revision 40.2
date: 1993/10/30 10:30:02;  author: entropy;  state: Exp;  lines: +1 -1
-
=============================================================================
***** include/errno.h
----------------------------
revision 40.2
date: 1993/10/31 11:03:32;  author: entropy;  state: Exp;  lines: +40 -1
Add support for the socket library.
=============================================================================
***** include/osbind.h
----------------------------
revision 40.3
date: 1993/10/31 09:23:14;  author: hohmuth;  state: Exp;  lines: +3 -3
fix prototypes for Pure C: gemdos(), bios(), xbios()
=============================================================================

== end sync

math.h:: ++jrb
    allow define NO_INLINE_MATH to override _M68881_. Also, check
    for __GNUC_INLINE__ as a pre-requisite to inlining (so the  user
    may override globally with  __NO_INLINE__.

crt0.c:: ++jrb
    put a section change directive back to .data  at the end of the asm()
    block. in gcc 2.5.X the compiler does not reset the block type
    by emitting an appropriate .directive after a asm(). this was causing
    problems with -mbaserel (reloc of acc_argv was always out of range).

strerror.c:: ++jrb
    make sys_errlist[] a char *const like _sock_errlist[].

---------------------------- Patchlevel 95 ---------------------------

bzero.cpp: michal
    when Lmemset is defined and not Lbzero, the jump at the bottom
    of the memset setup reduces to a zero offset jump. put a
    #ifdef Lbzero (which amounts to  really #ifdef Lmemset && Lbzero)
    around the `jra do_set'.

---------------------------- Patchlevel 96 ---------------------------

==sync with mntlib42
***** abs.c
----------------------------
revision 41.2
date: 1993/11/24 20:23:16;  author: pvt1-117;  state: Exp;  lines: +7 -0
Inline abs() and labs() for LCC.
=============================================================================
***** bblink.c
----------------------------
revision 41.3
date: 1994/01/13 08:03:50;  author: schwab;  state: Exp;  lines: +29 -26
Fixed version of update for GCC 2.5
----------------------------
revision 41.2
date: 1993/11/30 18:51:36;  author: schwab;  state: Exp;  lines: +139 -2
Updated for GCC 2.5
=============================================================================
***** buffindf.c
----------------------------
revision 41.2
date: 1994/01/09 09:33:20;  author: schwab;  state: Exp;
NEW file.
The library uses the names findfile and buffindfile which are in the
user's namespace. findfile is not really used and should be move into
a separate file, and buffindfile should be renamed to _buffindfile.
I have renamed findfile.c to buffindf.c and moved the function
findfile into a new findfile.c, and changed all callers.
=============================================================================
***** fdopen.c
----------------------------
revision 41.3
date: 1994/01/13 09:37:36;  author: entropy;  state: Exp;  lines: +2 -0
*** empty log message ***
----------------------------
revision 41.2
date: 1993/11/24 20:12:04;  author: nox;  state: Exp;  lines: +1 -1
Use binary mode on tty's (under MiNT) to avoid extra ^M's.
=============================================================================
***** findfile.c
----------------------------
revision 41.2
date: 1994/01/09 09:29:32;  author: schwab;  state: Exp;  lines: +2 -126
The library uses the names findfile and buffindfile which are in the
user's namespace. findfile is not really used and should be moved into
a separate file, and buffindfile should be renamed to _buffindfile.
I have renamed findfile.c to buffindf.c and moved the function
findfile into the new findfile.c, and changed all callers.
=============================================================================
***** fopen.c
----------------------------
revision 41.4
date: 1994/01/09 11:12:06;  author: nox;  state: Exp;  lines: +2 -2
*** empty log message ***
----------------------------
revision 41.3
date: 1994/01/09 11:06:16;  author: nox;  state: Exp;  lines: +2 -0
In _fopen() we seek to the end when file mode "a" is used (the kernel
won't do this until the first write.)
----------------------------
revision 41.2
date: 1993/11/24 20:12:34;  author: nox;  state: Exp;  lines: +1 -1
Use binary mode on tty's (under MiNT) to avoid extra ^M's.
=============================================================================
***** getpw.c
----------------------------
revision 41.2
date: 1993/11/30 20:40:50;  author: nox;  state: Exp;  lines: +4 -2
Convert backslashes in the home directory in the passwd file, for backwards
compatibility.
=============================================================================
***** lib.h
----------------------------
revision 41.2
date: 1993/11/30 18:51:48;  author: schwab;  state: Exp;  lines: +1 -1
Use new way to declare a function that doesn't return, since
the old way wasn't ANSI conforming.
=============================================================================
***** libgcc2.c
----------------------------
revision 41.2
date: 1993/11/30 18:51:58;  author: schwab;  state: Exp;  lines: +35 -62
New entry points: L_builtin_new, L_caps_New and L_builtin_del are replaced
by L_op_new, L_new_handler and L_op_delete.
=============================================================================
***** longlong.h
----------------------------
revision 41.2
date: 1993/11/30 18:52:22;  author: schwab;  state: Exp;  lines: +25 -24
Updated for GCC 2.5
=============================================================================
***** include/assert.h
----------------------------
revision 41.3
date: 1994/01/09 09:02:56;  author: lux;  state: Exp;  lines: +4 -13
Make assert() macro more ANSI compliant by replacing the broken assert()
macro with what was previously given as assertval(), and remove
the assertval() macro.
----------------------------
revision 41.2
date: 1993/11/30 18:50:46;  author: schwab;  state: Exp;  lines: +1 -1
Use new way to declare a function that doesn't return, since
the old way wasn't ANSI conforming.
=============================================================================
***** include/compiler.h
----------------------------
revision 41.2
date: 1993/11/30 18:50:50;  author: schwab;  state: Exp;  lines: +10 -0
Use new way to declare a function that doesn't return, since
the old way wasn't ANSI conforming.
=============================================================================
***** include/ioctl.h
----------------------------
revision 41.2
date: 1993/11/24 21:09:20;  author: schwab;  state: Exp;  lines: +1 -0
Add definition of B134.
=============================================================================
***** include/math-68881.h
----------------------------
revision 41.2
date: 1993/11/30 18:50:58;  author: schwab;  state: Exp;  lines: +112 -48
Give inline functions const attribute instead of declaring as const.
Fix a bug in atan2 where the function from libpml gives a different result
on the arguments (0, -1).
=============================================================================
***** include/minimal.h
----------------------------
revision 41.2
date: 1993/11/30 18:51:10;  author: schwab;  state: Exp;  lines: +4 -3
Use new way to declare a function that doesn't return, since
the old way wasn't ANSI conforming.
=============================================================================
***** include/mintbind.h
----------------------------
revision 41.3
date: 1994/01/09 10:01:00;  author: pvt1-117;  state: Exp;  lines: +1 -1
Some changes for Lattice C.
----------------------------
revision 41.2
date: 1993/11/24 20:50:10;  author: hohmuth;  state: Exp;  lines: +1 -1
Fix return type of Pwaitpid() for Pure C.
=============================================================================
***** include/osbind.h
----------------------------
revision 41.2
date: 1994/01/09 10:01:28;  author: pvt1-117;  state: Exp;  lines: +3 -3
Some changes for Lattice C.
=============================================================================
***** include/stdlib.h
----------------------------
revision 41.3
date: 1993/11/30 18:51:16;  author: schwab;  state: Exp;  lines: +3 -3
Use new way to declare a function that doesn't return, since
the old way wasn't ANSI conforming.
----------------------------
revision 41.2
date: 1993/11/24 20:20:46;  author: pvt1-117;  state: Exp;  lines: +10 -0
Inline abs() and labs() for LCC.
=============================================================================
***** include/support.h
----------------------------
revision 41.3
date: 1994/01/09 09:28:44;  author: schwab;  state: Exp;  lines: +1 -1
Rename buffindfile() to _buffindfile().
----------------------------
revision 41.2
date: 1993/11/30 18:51:22;  author: schwab;  state: Exp;  lines: +2 -2
Use new way to declare a function that doesn't return, since
the old way wasn't ANSI conforming.
=============================================================================
***** include/termios.h
----------------------------
revision 41.2
date: 1993/11/24 21:10:38;  author: schwab;  state: Exp;  lines: +1 -0
Add definition of B134.
=============================================================================
***** include/unistd.h
----------------------------
revision 41.2
date: 1993/11/30 18:51:28;  author: schwab;  state: Exp;  lines: +1 -1
Use new way to declare a function that doesn't return, since
the old way wasn't ANSI conforming.
=============================================================================

==end sync

mincl: ++jrb
    adjust for above
    -- add target for buffindf.o
    -- adjust names for op_new etc

---------------------------- Patchlevel 97 ---------------------------

== sync with mntlib 43
***** libgcc2.c
----------------------------
revision 42.3
date: 1994/02/10 11:38:42;  author: entropy;  state: Exp;  lines: +1 -1
Only use XFmode extended real floating point if XFLOAT_ENABLE is defined.
----------------------------
revision 42.2
date: 1994/01/28 16:30:08;  author: schwab;  state: Exp;  lines: +1 -1
Don't try to use extended float mode (XFmode) with gcc before version 2.4.
Actually, this will also exclude gcc 2.4, since __GNUC_MINOR__ was
introduced just with gcc 2.5.
=============================================================================
***** longlong.h
----------------------------
revision 42.2
date: 1994/01/28 16:30:30;  author: schwab;  state: Exp;  lines: +24 -25
Revert the last change, we don't need a register prefix.
=============================================================================
***** include/assert.h
----------------------------
revision 42.2
date: 1994/02/12 00:01:04;  author: entropy;  state: Exp;  lines: +14 -14
Add back old statement-based assert macro, but now called __assert().
=============================================================================
***** include/osbind.h
----------------------------
revision 42.2
date: 1994/02/10 08:05:26;  author: entropy;  state: Exp;  lines: +2 -2
Relax constraints from "r" to "g" on trap_14_wllwwwwwlw (used
by Flopfmt()).
=============================================================================

==end sync

==sync with mntlib 44
***** a64l.c
----------------------------
revision 43.2
date: 1994/02/28 13:27:40;  author: entropy;  state: Exp;  lines: +0 -49
broke out functions
=============================================================================
***** abort.c
----------------------------
revision 43.2
date: 1994/02/27 13:16:42;  author: entropy;  state: Exp;  lines: +9 -7
A possibly more ANSI/POSIX compliant abort().
=============================================================================
***** abs.c
----------------------------
revision 43.2
date: 1994/02/28 13:33:08;  author: entropy;  state: Exp;  lines: +4 -9
broke out functions
=============================================================================
***** atoi.c
----------------------------
revision 43.2
date: 1994/02/28 13:33:32;  author: entropy;  state: Exp;
NEW file, broken out from atol.c
=============================================================================
***** atol.c
----------------------------
revision 43.2
date: 1994/02/28 13:34:10;  author: entropy;  state: Exp;  lines: +2 -8
broke out functions
=============================================================================
***** chmod.c
----------------------------
revision 43.2
date: 1994/02/28 13:36:30;  author: entropy;  state: Exp;  lines: +2 -24
broke out functions
=============================================================================
***** chown.c
----------------------------
revision 43.2
date: 1994/02/28 18:37:20;  author: entropy;  state: Exp;
NEW file, broken out from chmod.c
=============================================================================
***** div.c
----------------------------
revision 43.2
date: 1994/02/28 13:36:30;  author: entropy;  state: Exp;  lines: +3 -32
broke out functions
=============================================================================
***** fgetpos.c
----------------------------
revision 43.2
date: 1994/02/28 13:50:12;  author: entropy;  state: Exp;
NEW file, broken out from fsetpos.c
=============================================================================
***** fopen.c
----------------------------
revision 43.2
date: 1994/02/28 13:36:30;  author: entropy;  state: Exp;  lines: +20 -146
broke out functions
=============================================================================
***** fopen_i.c
----------------------------
revision 43.2
date: 1994/02/28 13:46:12;  author: entropy;  state: Exp;
NEW file, broken out from fopen.c
=============================================================================
***** fputs.c
----------------------------
revision 43.2
date: 1994/02/28 13:36:30;  author: entropy;  state: Exp;  lines: +15 -26
broke out functions
=============================================================================
***** freopen.c
----------------------------
revision 43.2
date: 1994/02/28 13:46:12;  author: entropy;  state: Exp;
NEW file, broken out from fopen.c
=============================================================================
***** fseek.c
----------------------------
revision 43.2
date: 1994/02/28 13:36:30;  author: entropy;  state: Exp;  lines: +5 -41
broke out functions
=============================================================================
***** fsetpos.c
----------------------------
revision 43.2
date: 1994/02/28 13:36:30;  author: entropy;  state: Exp;  lines: +17 -33
broke out functions
=============================================================================
***** ftell.c
----------------------------
revision 43.2
date: 1994/02/28 13:49:16;  author: entropy;  state: Exp;
NEW file, broken out from fseek.c
=============================================================================
***** getegid.c
----------------------------
revision 43.2
date: 1994/02/28 13:52:34;  author: entropy;  state: Exp;
NEW file, broken out from getuid.c
=============================================================================
***** geteuid.c
----------------------------
revision 43.2
date: 1994/02/28 13:52:34;  author: entropy;  state: Exp;
NEW file, broken out from getuid.c
=============================================================================
***** getgid.c
----------------------------
revision 43.2
date: 1994/02/28 13:52:34;  author: entropy;  state: Exp;
NEW file, broken out from getuid.c
=============================================================================
***** getpid.c
----------------------------
revision 43.2
date: 1994/02/28 13:36:30;  author: entropy;  state: Exp;  lines: +5 -2
broke out functions
=============================================================================
***** getppid.c
----------------------------
revision 43.2
date: 1994/02/28 13:50:12;  author: entropy;  state: Exp;
NEW file, broken out from getpid.c
=============================================================================
***** getuid.c
----------------------------
revision 43.2
date: 1994/02/28 13:36:30;  author: entropy;  state: Exp;  lines: +5 -63
broke out functions
=============================================================================
***** l64a.c
----------------------------
revision 43.2
date: 1994/02/28 13:27:04;  author: entropy;  state: Exp;
NEW file, broke out from a64l.c.
=============================================================================
***** labs.c
----------------------------
revision 43.2
date: 1994/02/28 13:28:50;  author: entropy;  state: Exp;
NEW file, formerly in abs.c
=============================================================================
***** ldiv.c
----------------------------
revision 43.2
date: 1994/02/28 13:42:52;  author: entropy;  state: Exp;
NEW file, broken out from div.c
=============================================================================
***** lib.h
----------------------------
revision 43.3
date: 1994/02/28 14:03:04;  author: entropy;  state: Exp;  lines: +12 -0
add some prototypes
----------------------------
revision 43.2
date: 1994/02/27 11:51:30;  author: entropy;  state: Exp;  lines: +1 -0
add _enoent() prototype.
=============================================================================
***** puts.c
----------------------------
revision 43.2
date: 1994/02/28 13:47:38;  author: entropy;  state: Exp;
NEW file, broken out from fputs.c
=============================================================================
***** regexp.c
----------------------------
revision 43.2
date: 1994/02/24 16:41:16;  author: schwab;  state: Exp;  lines: +1 -0
regcomp fails to compile "(a+|b)*" (from libtest/tregex.c).  The bug
is an uninitialized byte in reginsert.
=============================================================================
***** rewind.c
----------------------------
revision 43.2
date: 1994/02/28 13:49:16;  author: entropy;  state: Exp;
NEW file, broken out from fseek.c
=============================================================================
***** setegid.c
----------------------------
revision 43.2
date: 1994/02/28 13:52:34;  author: entropy;  state: Exp;
NEW file, broken out from getuid.c
=============================================================================
***** seteuid.c
----------------------------
revision 43.2
date: 1994/02/28 13:52:34;  author: entropy;  state: Exp;
NEW file, broken out from getuid.c
=============================================================================
***** setgid.c
----------------------------
revision 43.2
date: 1994/02/28 13:52:34;  author: entropy;  state: Exp;
NEW file, broken out from getuid.c
=============================================================================
***** setregid.c
----------------------------
revision 43.2
date: 1994/02/28 13:52:34;  author: entropy;  state: Exp;
NEW file, broken out from getuid.c
=============================================================================
***** setreuid.c
----------------------------
revision 43.2
date: 1994/02/28 13:52:34;  author: entropy;  state: Exp;
NEW file, broken out from getuid.c
=============================================================================
***** setuid.c
----------------------------
revision 43.2
date: 1994/02/28 13:52:34;  author: entropy;  state: Exp;
NEW file, broken out from getuid.c
=============================================================================
***** uidgid.c
----------------------------
revision 43.2
date: 1994/02/28 13:52:34;  author: entropy;  state: Exp;
NEW file, broken out from getuid.c
=============================================================================
***** include/PatchLev.h
----------------------------
revision 43.2
date: 1994/02/15 20:33:52;  author: entropy;  state: Exp;  lines: +1 -1
-
=============================================================================
***** include/ioctl.h
----------------------------
revision 43.5
date: 1994/02/27 09:42:40;  author: entropy;  state: Exp;  lines: +13 -8
Add some more TIOCM_* definitions
----------------------------
revision 43.4
date: 1994/02/21 19:31:36;  author: entropy;  state: Exp;  lines: +29 -0
Catch up with MiNT's file.h:  Add cursor ioctl's and some missing
process ioctl's.
----------------------------
revision 43.3
date: 1994/02/21 19:20:24;  author: entropy;  state: Exp;  lines: +9 -0
Add TIOCMGET (faked) and related bitmasks.
----------------------------
revision 43.2
date: 1994/02/19 14:25:04;  author: entropy;  state: Exp;  lines: +11 -9
Added TIOCSDTR, TIOCCDTR and cleaned up a bit.
=============================================================================
***** include/mintbind.h
----------------------------
revision 43.2
date: 1994/02/26 08:45:42;  author: entropy;  state: Exp;  lines: +8 -0
Add new MiNT system calls: Tmalarm(), Psigintr(), Suptime().
All are unofficial and subject to change.
=============================================================================
***** include/ostruct.h
----------------------------
revision 43.2
date: 1994/02/22 17:47:58;  author: entropy;  state: Exp;  lines: +4 -0
Complete the _PARAM structure used with Initmous().
=============================================================================

New files: atoi.c chown.c fgetpos.c fopen_i.c freopen.c ftell.c getegid.c
geteuid.c getgid.c getppid.c l64a.c labs.c ldiv.c puts.c rewind.c setegid.c
seteuid.c setgid.c setregid.c setreuid.c setuid.c uidgid.c

== end sync

---------------------------- Patchlevel 98 ---------------------------

falcon.h:: howard
    corrections from howard