home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
Tech Arsenal 1
/
Tech Arsenal (Arsenal Computer).ISO
/
tek-13
/
xvidoc.zip
/
SOURCE.LST
< prev
next >
Wrap
File List
|
1992-07-28
|
56KB
|
1,519 lines
Notes on the Xvi Source Code
Chris Downey
John Downey
Xvi (pronounced _e_c_k_s-_v_e_e-_e_y_e) is a free, portable,
multi-window implementation of the popular UNIX|- editor
vi.
This document contains information on how to port xvi
to systems not currently supported. It also explains
how the xvi source code is arranged into modules, and
explains some of the data structures which are used, so
that modifications may be made if and when necessary to
the editor itself.
_1. _I_N_T_R_O_D_U_C_T_I_O_N
Xvi is intended to be portable to just about any system. This is
one of the central reasons for its existence; the authors wish to
be able to use the same editor everywhere.
The main body of the editor is (supposedly) fully portable,
relying only on standard facilities defined by the White Book,
and on a set of _p_r_i_m_i_t_i_v_e_s which are provided by a set of one or
more modules for each operating system. If ___S_T_D_C__ is defined,
certain ANSI C facilities will be used, but the editor will
compile with non-ANSI compilers.
Therefore, in order to port xvi to a new system, all that is
necessary is to provide the defined set of _p_r_i_m_i_t_i_v_e_s, and then
build the editor. Or at least, that's the idea; we have refined
the set of primitives as we port the editor to new environments,
and it's getting pretty easy now.
The rest of this document is divided into sections as follows:
_S_e_c_t_i_o_n _2: _S_y_s_t_e_m-_S_p_e_c_i_f_i_c _M_o_d_u_l_e_s
This section deals with the layout of source files and
makefiles which you will have to deal with when porting xvi.
_S_e_c_t_i_o_n _3: _P_r_i_m_i_t_i_v_e_s _P_r_o_v_i_d_e_d _b_y _x_v_i
Discusses what primitives are provided by the main body of
the editor source code for use by the system interface code.
_________________________
|- UNIX is a trademark of Bell Laboratories.
_2_5_t_h _S_e_p_t_e_m_b_e_r _1_9_9_2 _P_a_g_e _1
_2 _X_v_i _S_o_u_r_c_e _C_o_d_e _N_o_t_e_s
_S_e_c_t_i_o_n _4: _S_y_s_t_e_m _I_n_t_e_r_f_a_c_e
Explains the primitives which need to be provided in order
to make xvi work.
_S_e_c_t_i_o_n _5: _D_a_t_a _S_t_r_u_c_t_u_r_e_s
Details the internal data types used in the editor, and any
functions available for operating on those types.
_S_e_c_t_i_o_n _6: _S_o_u_r_c_e _F_i_l_e_s
Lists the source files comprising the editor, and explains
what functionality is provided by each one.
_2. _S_Y_S_T_E_M-_S_P_E_C_I_F_I_C _M_O_D_U_L_E_S
The system-specific code normally consists of three (or more)
files; a ".c" file, a ".h" file, and a makefile. For example:
_q_n_x._c
_q_n_x._h
_m_a_k_e_f_i_l_e._q_n_x
comprise the system-specific module for the QNX operating system.
In most cases, the system-specific code is divided into two or
more modules, where one (called the _s_y_s_t_e_m _i_n_t_e_r_f_a_c_e _m_o_d_u_l_e) is
concerned with general interactions with the operating system and
the other (called the _t_e_r_m_i_n_a_l _i_n_t_e_r_f_a_c_e _m_o_d_u_l_e) is designed for
a specific interface to a display and keyboard (and possibly, a
mouse).
For example, the generic UNIX implementation has _u_n_i_x._c and
_u_n_i_x._h for the system interface module, and _t_e_r_m_c_a_p._c and
_t_e_r_m_c_a_p._h for the terminal interface module; this should work
reasonably with any full-duplex terminal that can be described in
the _t_e_r_m_c_a_p database. On consoles with memory-mapped displays,
or systems with graphic user interfaces, however, it may be
possible to achieve faster display updating, and perhaps other
benefits, by replacing the _t_e_r_m_c_a_p module with another one that
makes better use of whatever facilities are available. For
instance, there is an experimental version for SunView, which
allows mouse input on Sun workstations running the SunView window
system.
On the other hand, the _t_e_r_m_c_a_p-specific routines might
conceivably be useful on some other operating systems (such as
VMS), so in general it seemed a good idea to make the _t_e_r_m_c_a_p-
specific routines a separate module.
The current MS-DOS implementation has a separate terminal
interface module, which is designed specifically for IBM PC
compatible computers. This is in the files
_P_a_g_e _2 _2_5_t_h _S_e_p_t_e_m_b_e_r _1_9_9_2
_X_v_i _S_o_u_r_c_e _C_o_d_e _N_o_t_e_s _3
_i_b_m_p_c__a._a_s_m
_i_b_m_p_c__c._c
_i_b_m_p_c._h
The first of these is written in assembly language because there
are not enough routines common to the various MS-DOS C compilers
which reliably access the display and keyboard at a low enough
level.
The hardware-independent system interface module for MS-DOS is in
_m_s_d_o_s__a._a_s_m
_m_s_d_o_s__c._c
_m_s_d_o_s._h
The first of these is written in assembly language for the same
reason as is _i_b_m_p_c__a._a_s_m.
Theoretically, different terminal interface modules could be
written for MS-DOS systems running on hardware which is not IBM-
compatible but, unfortunately, such systems seem to be virtually
extinct nowadays.
Sometimes more than one makefile is provided, as in the case of
UNIX, where different versions work in slightly different ways.
It is, of couse, not necessary to provide all - or any - of these
files for a particular implementation; this is just a convention.
The makefile(s) for each system determine what files are used in
the compilation of the editor.
The following porting modules are available at present:
_2_5_t_h _S_e_p_t_e_m_b_e_r _1_9_9_2 _P_a_g_e _3
_4 _X_v_i _S_o_u_r_c_e _C_o_d_e _N_o_t_e_s
________________________________________________________________
| System | Makefile | Source Files |
|________________________|_______________|_________________________|
| UNIX | | |
| BSD | makefile.bsd| unix.[ch] termcap.[ch]|
| System V |- | makefile.usg| unix.[ch] termcap.[ch]|
| AIX | makefile.aix| unix.[ch] termcap.[ch]|
| ULTRIX | makefile.ult| unix.[ch] termcap.[ch]|
| Xenix |- | makefile.xen| unix.[ch] termcap.[ch]|
| POSIX (e.g. BSDI) | makefile.pos| unix.[ch] termcap.[ch]|
| SunOS | makefile.sun| unix.[ch] termcap.[ch]|
| SunView | makefile.sv | unix.[ch] sunview.h |
| | | sunfront.c sunback.c |
| | | xvi.icn |
|________________________|_______________|_________________________|
| | | |
| MS-DOS | | msdos_c.c msdos.h |
| | | ibmpc_c.c ibmpc.h |
| Microsoft C 5.* | makefile.msc| 8086mm.inc ibmpc_a.asm|
| & MASM 5.* | | msdos_a.asm |
| | | |
| Microsoft Quick C | makefile.qc | 8086mm.inc ibmpc_a.asm|
| & MASM 5.* | | msdos_a.asm |
| Zortech C++ 2.* | makefile.zc2| 8086mm.inc ibmpc_a.asm|
| & MASM 5.* | | msdos_a.asm |
| | | |
| Zortech C++ 3.* | makefile.zc3| 8086mm.inc ibmpc_a.asm|
| & MASM 5.* | | msdos_a.asm |
| Zortech C++ 3.* | | |
| 386 protected mode | makefile.386| pc386.[ch] |
|________________________|_______________|_________________________|
| | | |
| OS/2 |- | | |
| Version 1, text mode| | |
| Microsoft C 5.1 | makefile.os2| os2vio.[ch] |
| & MASM 5.1 | | i286.asm |
|________________________|_______________|_________________________|
| QNX | | |
| Version 2/3 (CII) | makefile.qnx| qnx.[ch] |
| Version 4 (Watcom C)| makefile.qn4| unix.[ch] termcap.[ch]|
|________________________|_______________|_________________________|
| | | |
| TOS |- | | |
| Lattice C | makefile.tos| tos.[ch] tos.lnk |
|________________________|_______________|_________________________|
|- Versions marked with |- probably do not work, as systems have
not been recently available to the authors for testing.
_P_a_g_e _4 _2_5_t_h _S_e_p_t_e_m_b_e_r _1_9_9_2
_X_v_i _S_o_u_r_c_e _C_o_d_e _N_o_t_e_s _5
_3. _P_R_I_M_I_T_I_V_E_S _P_R_O_V_I_D_E_D _B_Y _X_V_I
_3._1. _G_e_n_e_r_a_l _D_e_f_i_n_i_t_i_o_n_s
The file _x_v_i._h should be included by all system-specific modules;
this file should also be edited so that a system-specific header
file (or files), as determined by a predefined keyword, will be
included.
For instance, under UNIX, the word _U_N_I_X is defined by passing the
-_D_U_N_I_X flag to the C compiler from the makefile, and _x_v_i._h
contains the following lines:
#_i_f_d_e_f _U_N_I_X
# _i_n_c_l_u_d_e "_u_n_i_x._h"
#_e_n_d_i_f
in order to obtain the UNIX-related definitions from that header
file.
Among the definitions in _x_v_i._h are the following:
bool_t
A Boolean type having values _T_R_U_E or _F_A_L_S_E.
const
volatile
These are defined out when ___S_T_D_C__ is not defined, so that
it is always safe to use them.
_x_v_i._h also includes various other header files which are needed.
The following system header files are always included:
_s_t_d_i_o._h
_c_t_y_p_e._h
_s_i_g_n_a_l._h
_s_t_r_i_n_g._h
These files are included if ___S_T_D_C__ is defined:
_s_t_d_d_e_f._h
_s_t_d_l_i_b._h
_l_i_m_i_t_s._h
and if ___S_T_D_C__ is not defined, _x_v_i._h will provide its own
definitions for the following:
_I_N_T__M_A_X
_I_N_T__M_I_N
_U_L_O_N_G__M_A_X
_F_I_L_E *_f_o_p_e_n();
_c_h_a_r *_m_a_l_l_o_c();
_c_h_a_r *_g_e_t_e_n_v();
_2_5_t_h _S_e_p_t_e_m_b_e_r _1_9_9_2 _P_a_g_e _5
_6 _X_v_i _S_o_u_r_c_e _C_o_d_e _N_o_t_e_s
Finally, one of the following header files will be included:
_s_t_d_a_r_g._h
_v_a_r_a_r_g_s._h
depending on whether ___S_T_D_C__ is defined or not. In order to
make coding of _v_a_r_a_r_g_s functions easier, a macro _V_A__S_T_A_R_T() is
defined, which takes the same arguments as the ANSI-style
_v_a__s_t_a_r_t(), but which is also available in non-ANSI environments
(e.g. BSD).
In order to make it possible to use ANSI-style prototypes for
function declarations, but still allow compilation under non-ANSI
environments, the following macro is provided:
#_i_f_d_e_f ___S_T_D_C__
# _d_e_f_i_n_e _P(_a_r_g_s) _a_r_g_s
#_e_l_s_e
# _d_e_f_i_n_e _P() ()
#_e_n_d_i_f
so that function declarations may be specified thus:
_e_x_t_e_r_n _F_I_L_E *_f_o_p_e_n _P((_c_o_n_s_t _c_h_a_r *, _c_o_n_s_t _c_h_a_r *));
Please use this facility when you provide declarations for your
system primitives, unless your system always uses an ANSI
compiler.
_3._2. _P_a_r_a_m_e_t_e_r_s
An important facility provided for use by system-specific modules
is access to the editor's parameter table. This is achieved by
means of some apparent functions, and a set of #_d_e_f_i_n_ed token
values. The functions are:
void set_param(int n, val)
This function sets the indicated parameter to the passed
value, which must be of an appropriate type. Parameter
values may be obtained by means of the following functions
(actually macros):
char *Ps(int n)
return value of string parameter
int Pn(int n)
return value of numeric parameter
bool_t Pb(int n)
return value of boolean parameter
char **Pl(int n)
return value of list parameter (a NULL-terminated array of
character pointers)
_P_a_g_e _6 _2_5_t_h _S_e_p_t_e_m_b_e_r _1_9_9_2
_X_v_i _S_o_u_r_c_e _C_o_d_e _N_o_t_e_s _7
int Pen(int n)
return numeric value (index) of enumerated parameter
char **Pes(int n)
return string value of enumerated parameter
In all cases, the int n argument is the index of the parameter in
the table; a set of #_d_e_f_i_n_es is provided, of the form:
_P__n_a_m_e
which map the parameter names into integral values. Thus, for
example, we might obtain the value of the _c_o_l_o_u_r parameter:
_c_o_l_o_u_r = _P_n(_P__c_o_l_o_u_r);
or set the value of the _h_e_l_p_f_i_l_e parameter:
_s_e_t__p_a_r_a_m(_P__h_e_l_p_f_i_l_e, "/_u_s_r/_l_i_b/_x_v_i/_h_e_l_p");
_4. _S_Y_S_T_E_M _I_N_T_E_R_F_A_C_E
_4._1. _I_n_t_r_o_d_u_c_t_i_o_n
There follows a list of the primitives which must be provided
either by the system interface module or by the underlying OS.
Note that it is perfectly acceptable to implement functions or
external variables as macros so long as they "look the same" as
the definitions below. As a guideline, anything which is (a) in
capitals, or (b) is a const variable, will be implemented as a
#_d_e_f_i_n_e for most systems.
When you want to actually do the port, it is highly recommended
that you copy the system-specific files for the system which
seems closest to your own, and modify those files, rather than
starting from scratch.
All the following symbols should be defined in the system
interface module, or by standard header files already included by
_x_v_i._h, or by other header files explicitly included by the
system-specific header file:
const unsigned int MAXPATHLEN
The maximum number of characters in a pathname.
const unsigned int MAXNAMLEN
The maximum number of characters in a filename.
int remove(char *filename)
Remove the named file as per ANSI.
_2_5_t_h _S_e_p_t_e_m_b_e_r _1_9_9_2 _P_a_g_e _7
_8 _X_v_i _S_o_u_r_c_e _C_o_d_e _N_o_t_e_s
int rename(char *old, char *new)
Rename the file old to new as per ANSI.
void sleep(unsigned int seconds)
Put the process to sleep for the given number of seconds.
const char * const DIRSEPS
The pathname separators supported for system calls (e.g.
"\\/" for MS-DOS).
FILE *fopenrb(char *file)
FILE *fopenwb(char *file)
Like the standard _f_o_p_e_n() library call, but they both open
files in "binary" mode (i.e. no conversion of cr/lf/crlf is
done), for reading and writing respectively.
bool_t exists(char *filename)
Returns _T_R_U_E if the named file exists.
bool_t can_write(char *filename)
Returns _T_R_U_E if the named file can be written, i.e. if a
fopenwb(filename) will succeed.
char *fexpand(char *filename)
Returns a filename-expanded version of the passed filename.
#define SETVBUF_AVAIL
const unsigned int READBUFSIZ
const unsigned int WRTBUFSIZ
If _S_E_T_V_B_U_F__A_V_A_I_L (or ___S_T_D_C__) is defined, these constant
values are used to set I/O buffer sizes (using the setvbuf()
function) for reading and writing files. Note that if
buffers of these sizes are unavailable at runtime, the
editor will try to allocate smaller buffers by iteratively
halving the buffer size until the allocation succeeds. It
is therefore acceptable for these values to be quite large.
char *tempfname(const char *filename)
Create a unique name for a temporary file, possibly using
filename as a base (this will be used by _d_o__p_r_e_s_e_r_v_e() to
create a backup file for the file named by _f_i_l_e_n_a_m_e). The
string returned must have been allocated using _m_a_l_l_o_c();
_N_U_L_L can be returned if there is no more memory available.
int call_system(char *command)
Invoke the given command in a subshell. This is used for
shell escapes from xvi. The command string may contain
metacharacters which are expected to be expanded by a
command interpreter, e.g. UNIX /_b_i_n/_s_h, MS-DOS _c_o_m_m_a_n_d._c_o_m.
Return value is 0 for success. In many environments, this
call may safely be #_d_e_f_i_n_ed as _s_y_s_t_e_m(_c_o_m_m_a_n_d).
_P_a_g_e _8 _2_5_t_h _S_e_p_t_e_m_b_e_r _1_9_9_2
_X_v_i _S_o_u_r_c_e _C_o_d_e _N_o_t_e_s _9
int call_shell(char *shell)
Invoke the named shell. This is used for the :_s_h_e_l_l
command. It may be mapped into _c_a_l_l__s_y_s_t_e_m(), but is
separate on some systems for efficiency reasons (i.e. not
invoking two shells to get one). Return value is 0 for
success.
bool_t
sys_pipe(char *cmd, int (*wf)(FILE *), long (*rf)(FILE *))
Used for the ! command. The first parameter is the command
to invoke, while the second and third are functions which
should be called with an open file pointer in order to write
out old, or read in new lines (respectively). Note that if
"real" pipes are not available, it is acceptable to
implement this function using temporary files, but the wf
function must obviously be called before rf.
void sys_exit(int code)
Exit with given exit status. This routine must not return.
The editor is considered "dead" once it has been called, and
no further calls to editor functions should be made.
void delay(void)
Delay for a short time, about a fifth of a second. This is
used for showing matching brackets when showmatch is set.
It is acceptable to just return if implementing this is not
easy.
_4._2. _S_c_r_e_e_n _C_o_n_t_r_o_l
An instance of the following structure must be defined in order
to allow screen output to take place:
_2_5_t_h _S_e_p_t_e_m_b_e_r _1_9_9_2 _P_a_g_e _9
_1_0 _X_v_i _S_o_u_r_c_e _C_o_d_e _N_o_t_e_s
_t_y_p_e_d_e_f _s_t_r_u_c_t _v_i_r_t_s_c_r {
_g_e_n_p_t_r *_p_v__w_i_n_d_o_w;
_i_n_t _p_v__r_o_w_s;
_i_n_t _p_v__c_o_l_s;
/* _p_u_b_l_i_c: */
_V_i_r_t_S_c_r *(*_v__n_e_w)(_V_i_r_t_S_c_r *);
_v_o_i_d (*_v__c_l_o_s_e)(_V_i_r_t_S_c_r *);
_i_n_t (*_v__r_o_w_s)(_V_i_r_t_S_c_r *);
_i_n_t (*_v__c_o_l_s)(_V_i_r_t_S_c_r *);
_v_o_i_d (*_v__c_l_e_a_r__a_l_l)(_V_i_r_t_S_c_r *);
_v_o_i_d (*_v__c_l_e_a_r__l_i_n_e)(_V_i_r_t_S_c_r *);
_v_o_i_d (*_v__g_o_t_o)(_V_i_r_t_S_c_r *, _i_n_t _r_o_w, _i_n_t _c_o_l);
_v_o_i_d (*_v__a_d_v_i_s_e)(_V_i_r_t_S_c_r *, _i_n_t _r_o_w, _i_n_t _c_o_l,
_i_n_t _i_n_d_e_x, _c_h_a_r *_s_t_r);
_v_o_i_d (*_v__w_r_i_t_e)(_V_i_r_t_S_c_r *, _i_n_t _r_o_w, _i_n_t _c_o_l, _c_h_a_r *_s_t_r);
_v_o_i_d (*_v__p_u_t_c)(_V_i_r_t_S_c_r *, _i_n_t _r_o_w, _i_n_t _c_o_l, _i_n_t _c_h);
_v_o_i_d (*_v__s_e_t__c_o_l_o_u_r)(_V_i_r_t_S_c_r *, _i_n_t _c_o_l_o_u_r);
_i_n_t (*_v__c_o_l_o_u_r__c_o_s_t)(_V_i_r_t_S_c_r *);
_v_o_i_d (*_v__f_l_u_s_h)(_V_i_r_t_S_c_r *);
_v_o_i_d (*_v__b_e_e_p)(_V_i_r_t_S_c_r *);
/* _o_p_t_i_o_n_a_l: _n_o_t _u_s_e_d _i_f _N_U_L_L */
_v_o_i_d (*_v__i_n_s_e_r_t)(_V_i_r_t_S_c_r *, _i_n_t _r_o_w, _i_n_t _c_o_l, _c_h_a_r *_s_t_r);
_i_n_t (*_v__s_c_r_o_l_l)(_V_i_r_t_S_c_r *, _i_n_t _s_t_a_r_t, _i_n_t _e_n_d, _i_n_t _n_l_i_n_e_s);
} _V_i_r_t_S_c_r;
The first three fields in this structure are "private", for use
only within the implementation of the "public" functions. The
remaining fields are all function pointers, and are described
below. Note that all functions have at least one parameter,
which is a pointer to the instance of the VirtScr in question.
This is always referred to as vs below. Note also that the top-
left-hand corner of the window is taken to be (0,0).
v_new(vs)
Obtain a new VirtScr, and return a pointer to it. This is
not used at present, and should return _N_U_L_L.
v_close(vs)
Close the window to which vs refers.
v_rows(vs)
Return the number of rows in vs.
_P_a_g_e _1_0 _2_5_t_h _S_e_p_t_e_m_b_e_r _1_9_9_2
_X_v_i _S_o_u_r_c_e _C_o_d_e _N_o_t_e_s _1_1
v_cols(vs)
Return the number of columns in vs.
v_clear_all(vs)
Clear the window completely.
v_clear_line(vs, int row, int col)
Clear the specified line, from the given column to the right
hand edge of the window, inclusive.
v_goto(vs, int row, int col)
Move the cursor to the specified row and column.
v_advise(vs, int row, int col, int index, char *str)
This function is called when the editor is about to produce
some output on the same line as the last output, but
separate from it by one or more characters. The destination
position is the coordinate pair (row, col + index), and str
contains the string of characters which are in the window
starting at position (row, col). Where there is a cost
incurred by moving the cursor to a specific screen position,
the terminal interface module may decide to write the
intervening characters to the screen rather than using a
specific "move cursor" sequence, in order to minimise the
number of characters written to the terminal.
Note that for many environments, the cost of re-positioning
the cursor is nil, and under these circumstances this
function need not do anything.
v_write(vs, int row, int col, char *str)
Write the specified string of characters into the window,
starting at the specified row and column. The parameters
will be such that the string will always fit into a single
line of the window, i.e. no line-wrapping is necessary;
however, it is quite possible for the string to end on the
last character of a line, and some implementations will need
to take special precautions to handle this correctly.
v_putc(vs, int row, int col, int ch)
This is like v_write but for a single character.
v_set_colour(vs, int colour)
Set the colour for all subsequent output (including clearing
of lines or the whole window) to the specified colour. The
meaning of the value is system-specific.
v_colour_cost(vs)
Return the number of extra characters which are taken up in
the window by a colour change. This is almost always 0, but
there exist some terminals for which it is not (see the "sg"
_t_e_r_m_c_a_p capability).
_2_5_t_h _S_e_p_t_e_m_b_e_r _1_9_9_2 _P_a_g_e _1_1
_1_2 _X_v_i _S_o_u_r_c_e _C_o_d_e _N_o_t_e_s
v_flush(vs)
Flush all screen output, and move the cursor on the screen
to the correct position. The screen need not actually be
updated until either this function is called, or
xvi_handle_event() returns.
v_beep(vs)
Beep. It is acceptable to flash the screen or window if no
audio facility is available.
v_insert(vs, int row, int col, char *str)
This function inserts the given string at the given
position, pushing any other characters on the same row to
the right. If such a facility is not available, the
function pointer should be set to _N_U_L_L.
v_scroll(vs, int start, int end, int nlines)
This function scrolls the set of lines between start and end
(inclusive) by nlines lines. If nlines is positive, _n_o_r_m_a_l
scrolling should be done, i.e. the lines should be moved
upwards with respect to the window. If nlines is negative,
scrolling should be in the reverse direction. The lines
which are left by the scrolling should be cleared. The
function should return non-zero if the scrolling was
successful, otherwise 0.
If scrolling is not available, the function pointer should
be set to _N_U_L_L.
_4._3. _P_a_r_a_m_e_t_e_r_s
Default values should be #_d_e_f_i_n_ed for certain parameters as
follows:
__________________________________________
| Parameter Name| Type | #define name |
|________________|__________|________________|
| syscolour | numeric| DEF_SYSCOLOUR|
| colour | numeric| DEF_COLOUR |
| statuscolour | numeric| DEF_STCOLOUR |
| roscolour | numeric| DEF_ROSCOLOUR|
| helpfile | string | HELPFILE |
|_f_o_r_m_a_t__________|__s_t_r_i_n_g___|__D_E_F___T_F_F________|
_4._4. _F_i_l_e _F_o_r_m_a_t_s
The functions in xvi which read and write text files are aware of
several different newline conventions (for example, "\n" on UNIX,
"\r\n" on MS-DOS, and so on), so that any version of the editor
can read and write any of the supported formats. The value of
the format parameter (which can be set to "unix", "msdos",
"macintosh", etc.) determines which format is currently being
used. If you are porting xvi to a system with a newline
convention which isn't one of those currently supported (see the
table called _t_f_t_a_b_l_e in _f_i_l_e_i_o._c) you may have to add a new entry
_P_a_g_e _1_2 _2_5_t_h _S_e_p_t_e_m_b_e_r _1_9_9_2
_X_v_i _S_o_u_r_c_e _C_o_d_e _N_o_t_e_s _1_3
to the table.
Unfortunately, the current design is not as general as it ought
to be. If you happen to be porting to VMS, or some other system
which doesn't use either a single character or a consecutive pair
of characters to represent a newline, you will have quite a lot
of work to do if you want to retain the facility for converting
between file formats within the editor.
In any case, your system interface module should define _D_E_F__T_F_F
to be the index of the entry in tftable which represents the
default format for your system. This is the value for
_P_e_n(_P__f_o_r_m_a_t) which will be compiled into the parameter table.
_4._5. _N_o_t_e_s _o_n _T_e_r_m_c_a_p _I_m_p_l_e_m_e_n_t_a_t_i_o_n
There exists a termcap implementation of the terminal interface,
currently only used for the UNIX port. This module could quite
easily be re-used for other systems if desired; the following
routines would need to be defined by the system module:
void foutch(int c)
Output a single character to the terminal. This must be
implemented as a function, not a macro, because it is passed
as a parameter into the _t_e_r_m_c_a_p library.
void moutch(int c)
Same as _f_o_u_t_c_h() except that it can be implemented as a
macro. This will be used by the _t_e_r_m_c_a_p interface module to
write characters to the screen.
void oflush(void)
Flush buffered output to the terminal.
_4._6. _E_n_t_e_r_i_n_g/_L_e_a_v_i_n_g _V_i_s_u_a_l _M_o_d_e
Some facility is commonly necessary for the system interface
module to be able to tell the terminal interface module to enter
or exit _v_i_s_u_a_l mode. This might mean changing the terminal state
between "raw" and "cooked" modes, or switching display pages. No
specific interface for this is defined, although the standard
UNIX and MS-DOS implementations do use such a facility, and the
interface functions for both systems are identically defined.
_4._7. _F_u_n_c_t_i_o_n _K_e_y_s/_M_o_u_s_e _H_a_n_d_l_i_n_g
Function key values are coded into a set of #_d_e_f_i_n_ed constants in
the file _a_s_c_i_i._h; e.g. the value _K__U_A_R_R_O_W might be given as input
when the keyboard up-arrow key has been pressed.
If the global variable _S_t_a_t_e is not equal to _N_O_R_M_A_L, all function
keys except for a backspace key are invalid input. If an invalid
key is pressed, the safest strategy may be to beep and wait for
another key to be pressed. _N_O_R_M_A_L is defined in _x_v_i._h.
_2_5_t_h _S_e_p_t_e_m_b_e_r _1_9_9_2 _P_a_g_e _1_3
_1_4 _X_v_i _S_o_u_r_c_e _C_o_d_e _N_o_t_e_s
Another facility which may be provided is handling mouse input on
systems where it is available. The strategy for interpreting
mouse input is controlled by the _m_o_u_s_e_c_l_i_c_k() function (in
_m_o_u_s_e._c); the idea is to make the strategy independent of any
specific device interface. If a mouse button is pressed before a
keyboard key is pressed, the following routine should be called:
_m_o_u_s_e_c_l_i_c_k(_i_n_t _r_o_w, _i_n_t _c_o_l_u_m_n);
where row and column are the current co-ordinates, counted in
character positions, of the mouse pointer within the screen or
editing window. If the mouse is moved while a button is held
down, the routine
_m_o_u_s_e_d_r_a_g(_i_n_t _s_t_a_r_t_r_o_w, _i_n_t _e_n_d_r_o_w, _i_n_t _s_t_a_r_t_c_o_l_u_m_n, _i_n_t _e_n_d_c_o_l_u_m_n);
should be called with co-ordinates describing the movement. If
the global variable _S_t_a_t_e is not equal to _N_O_R_M_A_L, mouse input can
be ignored altogether.
All this will be considerably tidied up at a later stage, when we
have proper _x_v_E_v_e_n_t types for function keys and mouse actions.
_4._8. _M_a_i_n
Finally, the system interface module must provide a main()
function. This function must call xvi_startup(vs, argc, argv,
env) at startup, with parameters as follows:
VirstScr *vs;
This is a pointer to the VirtScr structure for the first
window, or for the terminal screen.
int argc, char **argv;
These are as for a main() function.
char *env;
This is an environment string, normally the return value
from getenv("XVINIT"). If the concept of environment
variables does not exist, a string of the form
"source _f_i_l_e_n_a_m_e" may be passed instead, so as to allow
users to localise their usage of the editor.
The return value from xvi_startup() is a pointer, which will be
used in future to identify the window for input events. For now,
it should be stored in the VirtScr's pv_window field.
Having called xvi_startup(), input events may then be passed to
the editor by calling xvi_handle_event with a pointer to an
xvEvent structure as the sole argument. This structure is
defined as follows:
_P_a_g_e _1_4 _2_5_t_h _S_e_p_t_e_m_b_e_r _1_9_9_2
_X_v_i _S_o_u_r_c_e _C_o_d_e _N_o_t_e_s _1_5
_t_y_p_e_d_e_f _s_t_r_u_c_t _e_v_e_n_t {
_e_n_u_m {
_E_v__c_h_a_r,
_E_v__t_i_m_e_o_u_t
} _e_v__t_y_p_e;
_u_n_i_o_n {
/* _E_v__c_h_a_r: */
_i_n_t _e_v_u__i_n_c_h_a_r;
/* _E_v__t_i_m_e_o_u_t: */
} _e_v__u;
} _x_v_E_v_e_n_t;
#_d_e_f_i_n_e _e_v__i_n_c_h_a_r _e_v__u._e_v_u__i_n_c_h_a_r
The ev_type field is a tag which identifies the type of event
which has occurred. At present, only two events are supported:
an input character from the user, and a timeout. The union which
follows contains data associated with each event type; currently
only the type Ev_char requires data, as may be seen. The #_d_e_f_i_n_e
for ev_inchar is provided purely for convenience.
The return value from xvi_handle_event() is a long integer value
which is the time in milliseconds for which the editor is
prepared to wait for more input. If no input arrives within that
time, the function should be called again with an event of type
Ev_timeout. The timeout value returned may be 0L, indicating
that no timeout is necessary. It is very important that timeouts
should actually be implemented because they are needed for the
_p_r_e_s_e_r_v_e facility.
Currently, if a keyboard interrupt is received,
_x_v_i__h_a_n_d_l_e__e_v_e_n_t() need not be called (it should, in any case,
never be called from an asynchronous interrupt or signal handler)
but the global variable _k_b_d_i_n_t_r should be set to a non-zero
value.
_5. _D_A_T_A _S_T_R_U_C_T_U_R_E_S
Structures used in xvi are all typedef'd, and all begin with a
capital letter. They are defined in _x_v_i._h. The following data
structures are defined:
_5._1. _L_i_n_e
This structure is used to hold a single text line. It contains
forward and backward pointers which are connected together to
form a two-way linked list. It also contains a pointer to an
allocated text buffer, an integer recording the number of bytes
allocated for the text, and the line number (an unsigned long).
The text is null-terminated, and the space allocated for it may
_2_5_t_h _S_e_p_t_e_m_b_e_r _1_9_9_2 _P_a_g_e _1_5
_1_6 _X_v_i _S_o_u_r_c_e _C_o_d_e _N_o_t_e_s
be grown but is never shrunk. The maximum size of this space is
given by _M_A_X__L_I_N_E__L_E_N_G_T_H.
The line number is used when showing line numbers on screen, but
this is secondary to its main purpose of providing an ordering on
lines; the ordering of two lines in a list may be established by
simply comparing their line numbers (macros are available for
this purpose; see later for details).
_5._2. _B_u_f_f_e_r
This structure holds the internal representation of a file. It
contains pointers to the linked list of lines which comprise the
actual text. We always allocate an extra line at the beginning
and the end, with line numbers 0 and _M_A_X__L_I_N_E_N_O respectively, in
order to make the code which deals with this structure easier.
The line numbers of Line structures in a Buffer are always
maintained by code in undo.c, which is the only module which ever
changes the text of a Buffer.
The Buffer structure also contains:
o+ flags, including readonly and modified
o+ current filename associated with the buffer
o+ temporary filename for buffer preservation
o+ space for the _m_a_r_k module to store information about marked
lines
o+ space for the _u_n_d_o module to store information about the
last change
o+ number of windows associated with the buffer
The following macros are used to find out certain information
about Lines within Buffers:
lineno(Buffer *b, Line *l)
Returns the line number of the specified Line, which belongs
to the specified Buffer.
earlier(Line *l1, Line *l2)
Returns _T_R_U_E if l1 is earlier in the buffer than l2.
later(Line *l1, Line *l2)
Returns _T_R_U_E if l1 is later in the buffer than l2.
is_lastline(Line *l1)
Returns _T_R_U_E if l1 is the last line (i.e. the extra line at
the end, not the last text line) of the buffer.
_P_a_g_e _1_6 _2_5_t_h _S_e_p_t_e_m_b_e_r _1_9_9_2
_X_v_i _S_o_u_r_c_e _C_o_d_e _N_o_t_e_s _1_7
is_line0(Line *l1)
Returns _T_R_U_E if l1 is the 0th line (i.e. the extra line at
the start, not the first text line) of the buffer.
_5._3. _P_o_s_n
This structure is very simple; it contains a Line pointer and an
integer index into the line's text, and is used to record a
position within a buffer, e.g. the current cursor position.
These functions are available for operating on Posn structures:
gchar(Posn *)
Returns the character which is at the given position.
inc(Posn *)
Increments the given position, moving past end-of-line to
the next line if necessary. The following type is returned:
_e_n_u_m _m_v_t_y_p_e {
_m_v__N_O_M_O_V_E, /* _a_t _b_e_g_i_n_n_i_n_g _o_r _e_n_d _o_f _b_u_f_f_e_r */
_m_v__S_A_M_E_L_I_N_E, /* _s_t_i_l_l _w_i_t_h_i_n _s_a_m_e _l_i_n_e */
_m_v__C_H_L_I_N_E, /* _c_h_a_n_g_e_d _t_o _d_i_f_f_e_r_e_n_t _l_i_n_e */
_m_v__E_O_L, /* _a_t _t_e_r_m_i_n_a_t_i_n_g '_\_0' */
};
dec(Posn *)
As for inc() but decrements the position.
lt(Posn *p1, Posn *p2)
Returns _T_R_U_E if the position specified by p1 is earlier in
the buffer than that specified by p2.
_5._4. _X_v_i_w_i_n
This structure maps a screen window onto a Buffer. It contains:
o+ a pointer to the Buffer structure which it is mapped onto
o+ the cursor's _l_o_g_i_c_a_l position in the buffer (a Posn
structure)
o+ the cursor's _p_h_y_s_i_c_a_l position in the window (row and
column)
o+ information about size and location of screen window
o+ current text of status line
o+ forward and backward pointers to other windows
Note that there is at least one Xviwin for every Buffer.
When the editor was modified to support buffer windows, many
_2_5_t_h _S_e_p_t_e_m_b_e_r _1_9_9_2 _P_a_g_e _1_7
_1_8 _X_v_i _S_o_u_r_c_e _C_o_d_e _N_o_t_e_s
global variables were moved into the Buffer and Xviwin
structures; some were left as globals. For instance, the _u_n_d_o
and _m_a_r_k facilities are obviously buffer-related, but _y_a_n_k is
useful if it is global (actually static within its own module);
it was decided that _s_e_a_r_c_h and _r_e_d_o should also be global.
Some modules have their own internal static data structures; for
instance, the _s_e_a_r_c_h module remembers the last pattern searched
for. Also, certain modules use data structures which are
included in more global ones; e.g. each Buffer structure contains
some data used only within _u_n_d_o._c. This is not very well
structured, but in practice it's quite clean because we simply
ensure that references to such structures are kept local to the
module which "owns" them.
_5._5. _M_a_r_k
This data structure records a mark (defined by the m command).
It contains a Posn and a character field to hold the letter which
defines the mark. Each Buffer contains an array of these
structures for holding alphabetic marks, plus one for the
previous context mark (as used by the '' and `` commands). The
file _m_a_r_k._c deals with marks.
_5._6. _C_h_a_n_g_e
This structure records a single change which has been made to a
buffer. It also contains a pointer, so that it may be formed
into a list. See the discussion of _u_n_d_o._c below for further
details.
_5._7. _F_l_e_x_b_u_f
This structure is used to store text strings for which the length
is unknown. The following operations are defined for this type.
All functions take a Flexbuf pointer as a parameter.
flexnew(f)
Initialise a Flexbuf; not needed for static Flexbufs.
flexclear(f)
Truncate a Flexbuf to zero length, but don't free its
storage.
flexdelete(f)
Free all storage belonging to a Flexbuf.
flexempty(f)
Return _T_R_U_E if the Flexbuf is empty.
flexlen(f)
Return the number of characters in the Flexbuf.
_P_a_g_e _1_8 _2_5_t_h _S_e_p_t_e_m_b_e_r _1_9_9_2
_X_v_i _S_o_u_r_c_e _C_o_d_e _N_o_t_e_s _1_9
flexrmchar(f)
Remove the last character from a Flexbuf.
flexpopch(f)
Remove the first character from a Flexbuf and return it.
flexgetstr(f)
Return a pointer to the string contained in the Flexbuf.
flexaddch(f, c)
Add the character c to the end of the Flexbuf.
lformat(f, fmt, ...)
A subset of sprintf() but for Flexbufs.
vformat(f, fmt, va_list)
A subset of vsprintf() but for Flexbufs.
The last two functions are especially useful, since they avoid
the usual problems with the lack of bounds-checking in sprintf().
All code in the editor itself now uses Flexbufs to avoid the
possibility of buffer overruns, and to reduce the size of the
executable. Some OS-specific modules, however, may still use the
printf() family. The subset of printf-like format specifiers
implemented includes those for integers and strings, but not for
floating-point numbers.
_5._8. _b_o_o_l__t
A simple Boolean type; has values _T_R_U_E and _F_A_L_S_E, which are
defined as 1 and 0 so as to be compatible with C comparison
operators.
_5._9. _x_v_E_v_e_n_t
This type is defined in the previous section, since it forms part
of the porting interface.
_5._1_0. _V_i_r_t_S_c_r
This type represents a virtual screen, and is constructed in a
similar way to a _c_l_a_s_s. It contains some function pointers which
may be used to manipulate the screen in various ways, and some
private data which is used by the implementation of the class.
The old terminal interface, which consisted of a set of disparate
functions, is being replaced by the VirtScr interface; the first
step in this process has been accomplished by the provision of a
default VirtScr implementation using the old primitive functions.
New, native, VirtScr implementations may now be coded, which will
increase the efficiency of screen output.
As the final stage, a windowing implementation of the VirtScr
class will be provided, using the underlying VirtScr
implementations, and the window-handling code in the editor will
_2_5_t_h _S_e_p_t_e_m_b_e_r _1_9_9_2 _P_a_g_e _1_9
_2_0 _X_v_i _S_o_u_r_c_e _C_o_d_e _N_o_t_e_s
be modified to that each occurrence of an Xviwin references its
own VirtScr. It will then be possible to build a version of the
editor which operates in a true windowing environment by using a
separate screen window for each buffer, instead of the current
vertical-split method.
A full definition of the VirtScr type will be found in the
previous section.
_5._1_1. _G_l_o_b_a_l _V_a_r_i_a_b_l_e_s
There are only a few global variables in the editor. These are
the important ones:
curbuf
pointer to the current Buffer
curwin
pointer to the current Xviwin
State the current _s_t_a_t_e of the editor; controls what we do with
input characters. The value is one of the following:
NORMAL The default state; vi-mode commands may be
executed
INSERT Insert mode, i.e. characters typed get inserted
into the current buffer
REPLACE Replace mode, characters in the buffer get
overwritten by what is typed
CMDLINE Reading a colon-command, regular expression or
pipe command
DISPLAY Displaying text, i.e. :p command, or :set or
:map with no argument
echo This variable controls what output is currently
displayable. It is used at various points within the
editor to stop certain output which is either undesirable
or sub-optimal. It must always be restored to its
previous value after the code which changed it has
finished what it is doing.
kbdintr
This can be set to a non-zero value to indicate that an
asynchronous user-generated interrupt (such as a keyboard
interrupt) has occurred. See the discussion of event
handling in the previous section.
_P_a_g_e _2_0 _2_5_t_h _S_e_p_t_e_m_b_e_r _1_9_9_2
_X_v_i _S_o_u_r_c_e _C_o_d_e _N_o_t_e_s _2_1
_6. _S_O_U_R_C_E _F_I_L_E_S
The header file _x_v_i._h contains all the type definitions used
within the editor, as well as function declarations etc.
The following source files form the primary interface to the
editor:
startup.c
Entry point for the editor. Deals with argument and
option parsing and initial setup, calling module
initialisation functions as necessary.
events.c Contains the routine xvi_handle_event(), which is the
entry point for handling input to the editor; input is
passed to different routines according to the State
variable. Timeouts on input are also handled here, by
calling appropriate routines in map.c or preserve.c.
edit.c Deals with insert and replace modes.
normal.c Handles normal-mode commands.
map.c This file is responsible for all input mapping (both
set up by the :map command and internally for
function-key mappings; it also implements a stuff-
characters-into-the-input-stream function for use
within the editor. This is used, for example, to
implement command redo (but _n_o_t to implement "undo" and
"put" as in STEVIE).
Colon (ex-type) commands are handled by this group:
cmdline.c
Decodes and executes colon commands.
ex_cmds1.c
File-, Buffer- and Xviwin-related colon commands.
ex_cmds2.c
Other colon commands (e.g. shell escape).
Screen updating is done within the following files:
screen.c Screen updating code, including handling of line-based
entry (for colon commands, searches etc) as they are
typed in, and display-mode stuff (for parameter
displaying, :_g/_r_e/_p etc).
cursor.c This file contains the single function cursupdate(),
which is responsible for deciding where the physical
screen cursor should be, according to the position of
the logical cursor in the buffer and the position of
_2_5_t_h _S_e_p_t_e_m_b_e_r _1_9_9_2 _P_a_g_e _2_1
_2_2 _X_v_i _S_o_u_r_c_e _C_o_d_e _N_o_t_e_s
the window onto that buffer. This routine is not very
optimal, and will probably disappear in due course.
defscr.c This file contains the default implementation of the
VirtScr class, on top of the old terminal/system
interface.
status.c Functions to update the status line of a window; there
are different functions to display file information
(name, position etc.) and error/information messages.
These files deal with specific areas of functionality:
find.c Search functions: all kinds of searches, including
character-based and word-based commands, sections,
paragraphs, and the interface to "real" searching
(which is actually done in _s_e_a_r_c_h._c).
mark.c Provides primitives to record marks within a Buffer,
and to find the marks again.
mouse.c Code to handle mice moving the cursor around and
resizing windows.
param.[ch]
Code to handle setting of, and access to, parameters.
(These are things like tabstops, autoindent, etc.)
pipe.c Handles piping through system commands.
preserve.c
File preservation routines.
search.c Code for pattern-searching in a buffer, and for
substitutions and global execution. Uses regexp.[ch]
for the actual regular expression stuff.
tags.c Routines to handle tags - for :tag, -t and ^].
undo.c Code to deal with doing and undoing; i.e. making and
unmaking changes to a buffer. This is one of the more
complex and delicate files.
yankput.c
Code to deal with yanking and putting text, including
named buffers.
while these files provide lower-level functions:
alloc.c Memory allocation routines.
_P_a_g_e _2_2 _2_5_t_h _S_e_p_t_e_m_b_e_r _1_9_9_2
_X_v_i _S_o_u_r_c_e _C_o_d_e _N_o_t_e_s _2_3
ascii.[ch]
Deals with the visual representation of special
characters on the display (e.g. tabs, control chars).
buffers.c
Routines dealing with the allocation and freeing of
Buffers.
fileio.c File I/O routines; reading, writing, re-editing files.
Also handling of the format parameter.
flexbuf.c
Flexible-length character-buffer routines.
misccmds.c
Miscellaneous functions.
movement.c
Code to deal with moving the cursor around in the
buffer, and scrolling the screen etc.
ptrfunc.[ch]
Primitives to handle Posn structures; including various
operators to compare positions in a text buffer.
regexp.[ch], regmagic.h
Regular-expression stuff, originally written by Henry
Spencer (thanks Henry) and slightly hacked for use
within xvi.
signal.c Handling of terminal-generated signals in an ANSI
environment.
virtscr.h
Virtual Screen interface definition. This is a new
part of xvi, and is not yet fully completed. When it
is finished, it will provide the ability to implement
"native" versions of xvi under various windowing
systems, in a clean and wholesome way. Currently there
is a single instance of the VirtScr class, which is
defined on top of the old system/terminal interface.
windows.c
Code to deal with creating, deleting, resizing windows.
version.c
Contains only the version string.
_2_5_t_h _S_e_p_t_e_m_b_e_r _1_9_9_2 _P_a_g_e _2_3
