RISC DISC 1

home *** CD-ROM | disk | FTP | other *** search

/ RISC DISC 1 / RISC_DISC_1.iso / pd_share / utilities / zap / LineEditor / !ReadMe next >

Wrap

Text File | 1994-08-30 | 18.0 KB | 414 lines

LineEditor Module release 2.63 ============================== *** Source, binaries and documentation Copyright (C) Oliver Betts, 1994 *** This software is provided as is - it has been tested to a reasonable extent, but probably has a few lurking bugs. I take no responsibility for any loss or damage, directly or indirectly caused. UNIX is a trademark of AT&T; 4DOS is a trademark of JP Software Inc. Incidentally, I highly recommend 4DOS if you have to use DOS. Versions with a letter suffix are test versions and may only be distributed with express permission from me. Other versions may be distributed freely provided all files are supplied intact and that no profit is made. You may bundle it with other freeware software provided these distribution restrictions are met, but please make sure you're supplying the latest version. If you want to distribute it in some other way, get in touch. ***************************************************************************** This module started out as some modifications to version 2.30 of Richard K. Lloyd's LineEditor module, to make it work better with TaskWindows (specifically to use the same *FX4 and *FX 220-228 settings as the Wimp). However, RKL seems not to respond to email or snail mail, and so not having his permission to release my modified version, I have entirely rewritten the module. In doing so I have fixed a number of bugs which were in the original and added many extra features, some of which were inspired by tcsh (for UNIX) and 4DOS (for DOS!). I'm still actively working on this program, so watch out for new versions. Please let me know if you find any bugs or problems (read this file first though!). I'll release the source with future versions. Files in this release: !ReadMe - this file LineEditor - the module RebindKeys - obey file to change key bindings Actions - a list of the actions which can be bound to keys DfltKeys - default key bindings OllyKeys - the bindings I use -- provided as an example Rebinder - used by RebindKeys -- don't run directly The keys file has a very similar layout to the !Zap keys file. The mechanism for rebinding is a bit nasty, and will be improved when I get some spare time. For a list of the default key bindings, load the LineEditor module and type "*Help Editor". Random note: Ctrl-N is now mapped to Next-line (same as cursor down), so to turn paged scrolling on, you need to type Ctrl-Q Ctrl-N. Similarly for some other useful Ctrl-<letter> combinations. I've bound csL csN csO csT and csZ to "vanilla" by default, so you can use Ctrl-Shift-N, etc instead. A few comments on use with !Zap (see also !Zap's !Help file): * To use function keys and Control + key you need to type Ctrl-Q first to "Quote" the key - otherwise !Zap will process the key. The cursor keys, TAB and PageUp/PageDown are passed through and so don't need quoting. * COPYing works - activated by the command COPY which is bound to COPY and Shift-COPY by default. * TaskWindows under RISC OS 2: The TaskWindow module supplied with !Edit on RISC OS 2 Application Disc 1 (v 0.03) won't work with !Zap - you need the one supplied with the C/Asm DDE (about 0.29) or above. 0.03 only works with !Edit 1.00 and not very well. The RISC OS 3.1 TaskWindow module works best of all, but you can't legally get it without having RO 3.1... * With the default keymap supplied with !Zap 1.10, Enter on the numeric keypad acts like Ctrl-J in TaskWindows, ie the line isn't stored in the history. You can make it act like Return by altering the appropriate line in the keys file from: kENTER &166 RETURN to: kENTER &166 CHAR &0D which makes Enter behave like Return in TaskWindows, and in other modes. * Due to a bug in !Zap 1.10, Ctrl-Q Ctrl-@ in a TaskWindow gives an `R', rather than setting the mark. A work around is to use Ctrl-Q Ctrl-SPACE instead in TaskWindows. Use with !Edit/!SrcEdit: * !Edit/!SrcEdit TaskWindows don't "understand" the codes to move the cursor left and right (&08 and &09) so things don't work terribly well. Also, the code for a beep (&07) is ignored, so failing filename completions don't beep. * My version and RKL's seem to work equally well/badly (simple recall and filename completion work okay). * I don't think it's possible to work around not having VDU 8 - get !Zap if you want to use TaskWindows much - it's much faster too! Use with RISC OS 2: * I've added code to versions 2.34 and above to read the task handles of TaskWindows under RISC OS 2, so local history buffers should now work. * The TaskWindow modules supplied with RISC OS 2 and the Acorn C DDE (0.03 and 0.29 anyway) have a problem - if the LineEditor is loaded, closing a TaskWindow doesn't release the memory allocated to the task. This would seem to be a TaskWindow bug [it occurs if you soft-load 0.03 or 0.29 with RISC OS 3.1, and with RKL's version]. My best advice (other than to get RISC OS 3.1) is to reduce the next slot before starting a TaskWindow and not to close task windows that you might want again. * There's a problem with the cursor keys producing symbols rather than being interpreted as cursor keys with some setups when using TaskWindow 0.29. Comments and suggestions are greatly encouraged - email will get a reply (if I can get through), snail mail will eventually, but don't hold your breath. If email bounces, I read the usenet group comp.sys.acorn regularly. Please try to give as much detail as possible in bug-reports - steps to demonstrate the bug from loading in the module are very helpful indeed. Thanks to Andy Hayward, David Moore and Dom Symes for testing and comments. Also to Clive Jones for suggesting how to find the current task's handle under RISC OS 2 (though his idea didn't work ;) ). Also, thanks to everyone who's reported bugs. You can obtain the latest version by anonymous ftp from ftp.mantis.co.uk in the directory /pub/users/olly - the file should be le???.arc If you can't ftp, send me an email with "req-lineed" in the subject line, and my mailer will automagically uuencode and mail you the latest version. Olly Betts, 1994.08.30 Internet email: olly@mantis.co.uk Indirected snail mail: Olly Betts, 24 Morgans Road, Hertford, Herts, SG13 8BS, UK Improvements/bugfixes over RKL's 2.30: ====================================== More TaskWindow compatible: --------------------------- Changed to use same *FX4 and *FX 220-228 settings as Wimp, so using the line editor in a TaskWindow doesn't screw up the cursor keys. More values are stored in registers or blocks rather than absolute locations, which greatly improves behaviour with multiple TaskWindows. Each TaskWindow now has a separate currently-recalled-line pointer. If Global history is on, when lines are deleted (by Shift-Insert, or when the history buffer gets full), the pointers for each TaskWindow are adjusted appropriately. Function keys work separately in each TaskWindow. You can't *RMKill the LineEditor module while a line is being entered. Miscellaneous bugs fixed: ------------------------- Ctrl-H performs backspace (RKL's interprets it as Ctrl-Delete and erases the entire contents of the history buffer!) Backspace in overtype mode didn't delete characters on the screen. RKL's gave incorrect underline cursor in non-BBC gap modes (eg MODE 17). Escape now moves cursor to end of line before exiting. You can now enter all top-bit-set characters; Eg With RKL's version, you can't enter LOAD"Ý" (to type "Ý" hold down ALT and type 221 on the numeric keypad, then release ALT - RKL's reads it as Shift-Insert). Filename completion code now copes with long leafnames (RKL's would break for leafnames >11 characters eg AUTOEXEC/BAT is 12). RKL's used CMOS locations 37 and 38, but his !Trash also uses bits 0-2 of location 38 - I use location 37 and bits 3-7 of location 38. R12 wasn't being passed to readlinev - RKL had just hacked around this... V flag is explicitly cleared on errorless exit from ReadLineV routine, as OS_CallAVector doesn't clear V on entry (RISC OS 2 PRMs, vol I, page 30). Module initialisation code would call OS_GenerateError upon failure - it should return with V set and R0 -> error block (RO3 PRMs, page 1-204). Other changes which I view as improvements: ------------------------------------------- You can now have separate history buffers for each TaskWindow (controlled by bit 0 of *Configure EdFlags). Ctrl-Delete wipes *current* history buffer; *EraseHistory and Shift-Ctrl-Delete wipe *all* history buffers. When in local buffer mode, *Recall and *EdStatus put a '*' by the last used buffer. If the command was typed in using OS_ReadLine, this will be the "current" buffer, but results may be less meaningful if the command is executed by a program. Added PageUp/PageDown to scroll through lines with start matching entered line (also on shift-cursor-up/down) [like Instigator and cursor up/down in 4DOS]. Added Emacs-like bindings [mostly from tcsh]: Duplicating existing keys: ^A ^E start/end of line ^B ^F cursor left/right ^N ^P next/prev line ^D delete right New actions: ^Q quote next key, ^T transpose, ^K delete to end of line, ^@ place mark, ^X exchange cursor with mark (Emacs ^X ^X), ^W delete region, Shift-^D delete to end of word, ^Y yank back last text deleted with ^K, ^W, or Shift-^D Added another style of filename completion (like 4DOS). Shift-Tab completes fully on first match, then Shift-Tab and Ctrl-Shift-Tab step through other matches. Removed nasty hacky "Supervisor awareness" as it didn't work for RO 3.x or at ShellCLI prompt (f12 from desktop). Replaced with better system (for Acc.Fred[Tab] it looks first for "Acc.Fred*"; if that fails it then looks for "Fred*" [on the assumption that Acc. is an abbreviated *-command]). There is no longer a limit on the number of lines in the history buffer, only on the size of the buffer. You can specify a minimum length of line to record in the history (currently you set bits 4-7 of the flags byte to one less than this length, so 4 records all lines of 5 characters or more). Doesn't display copyright message on start-up and reset. Syntax messages added for commands. Code doesn't bother trying to change cursor shape in TaskWindows (as VDU sequences are ignored and only slow things up more). The code is "lazier" about inserting and deleting characters during filename completion - this gives a slight speed increase in TaskWindows. My completion code doesn't insert `delete' characters into the keyboard buffer, which might have caused problems if you pressed Shift or Ctrl at just the wrong moment. Next/previous word treat '(' and ')' as word breaks (same as !Zap 1.10). Module help messages compacted using OS dictionary. Ctrl-L & HOME now redraw current line (bit like tcsh). ^J still behaves like RETURN but doesn't store line in history. Doesn't do *Unset LineEditor$Dir (was done incorrectly for *EraseHistory for example). Buffers for filename completion code are only allocated when they're needed. No swi-padding code as it is *not* needed (slightly faster and smaller). Added option to assemble with less verbose help text. You can configure completion to add a '.' when completing a directory name, (a bit like tcsh adds a '/'). However "ls foo/" works and "cat foo." doesn't, so this is configurable (set bit 1 of EdFlags to turn it on). It also is not currently done for 4DOS style completion, as the dot would stop the stepping through the current directory - I might recode to fix this later. You can configure overtype mode to be on by default if you prefer (though why you would is beyond me - still, DOS 6 does it...) Shortfalls compared to RKL's version: ===================================== Changes to *Configure EdSize don't affect existing history buffer(s), but are used for new buffers. I'm not sure if this is actually a shortfall or a useful feature - may be fixed. No wimp frontend (or is that an improvement?) - seriously, let me know if you think a Wimp frontend is needed - I might write one if it's wanted... Hasn't been tested with Arthur (the Archimedes OS before RISC OS) - the help text is compacted but otherwise I think all the new OS calls used fail-safe. Do let me know if you try it... Minor "feature" - *FX216,0 no longer stops current function key expansion during OS_ReadLine, but this is no big deal really - it could be solved by intercepting OS_Byte, but I can't see much point. Possible future additional features: ==================================== Getting there: Add completion of *-commands (I have a prototype working) Improve key redefining stuff Improve this documentation! Some actions can be pretty slow (eg type a very long line with no spaces in a taskwindow, go to the start and delete word (Ctrl-Shift-D). Then go and put the kettle on if you have an ARM 2. This could be improved if the code was lazier and redrew only the finished result - although as it stands it does look quite slick on the command line with an ARM 3 or above. Doesn't work wonderfully in VDU5 mode [neither does RKL's] - maybe fix. Maybe add Ctrl-Page-Up/Down to scroll case sensitive. Maybe change the name (?) If we allow the size of existing buffers to change, we should preserve as much of the history as will fit in new size buffer. As with RKL's, you can recall lines which have characters in which are outside the range specified by R2 and R3 on entry to OS_ReadLine - the only solutions I can see are to strip them out (not wonderful) or to only allow recall if all chars in the range &20-&FF are allowed (this applies to ShellCLI, BASIC, and GOS (TaskWindows)). The problem also applies to filename completion. Ideas on this one are especially welcome. Other (please specify): .................................................... Revision History: ================= 2.63 Shift-SPACE now produces a SPACE in the Default map (and Olly's) Keypad /*-+. and Enter now work in the Default map (and Olly's) Fixed bug in code which works out !Zap key numbers - discovered via dropped characters in combination with Brian Brunswick's Bits module Fixed non-freeing of a small amount of RMA if allocation of a new buffer fails part-way through Fixed problem of "LineEditor in use" when it had never been used 2.62 With global history, if you hadn't used a non-TaskWindow command line, then the history was lost if all TaskWindows were closed - fixed 2.61 Fixed bug with reading value of EdFlags Fixed off-by-one error in interpretation of minimum line length to store 2.60 Cursor probably didn't get redefined for RISC OS 2 - should be now Fixed bug which recalled junk on pressing cursor-up on a non-empty line when history was empty Streamlined service call handler for "nothing of interest" case Should cope better if any SWI OS_Write* returns an error (and hence corrupts R0) - eg if *Spool-ing output and the disk becomes full All case-insensitive matching code rewritten to be smaller and faster Next/previous word now works just like in Zap 1.10 - ie ' ', '(' and ')' are treated as word breaks Now allocates leafname buffers for filename completion on-the-fly, so we can now cope with arbitrarily long leafnames. Tested with Jason Tribbeck's LongFilenames module (version 0.08) Added ^Y to yank back text deleted with ^K, ^W (like emacs) Filename completion now works properly in overtype mode *Help Editor output tweaked to fit better on 80x32 screen Ctrl-Shift-D deletes to end of word (Emacs Meta-D) Keypresses now decoded using !Zap key codes and a 512 byte table to enable easy re-mapping Compacted help messages using own dictionary You can specify a minimum length of line to record in the history - currently you set bits 4-7 of EdFlags to one less (eg 4 to store lines of 5 chars or more), so 1 <= min. length <= 16 Doesn't output VDU sequences to change cursor on start-up now (stops empty window appearing when LineEditor module run from Filer) Ctrl-Down on typed line with empty history gave garbage - fixed Delete word can be yanked back 2.53 Path buffer for filename completion now allocated on the fly 2.52 Routine `free' now returns R0=0 Explicitly clear V on errorless exit from ReadLineV (bug-fix over RKL) Fettled help messages and compacted a little more R12 is mostly derefenced on entry (smaller and faster) OS_GBPB buffer allocated on the fly Print padded decimal routine uses 12 bytes of stack for workspace 2.51 Dom suggested adding a '.' when filename completion completed to a directory - controlled by bit 1 of EdFlags Set bit 2 of EdFlags to turn overtype mode on by default 2.50 [Bundled in !Zap 1.10 release] Pressing Tab/^I on an empty line incorrectly moved the cursor left one character if a completion occured Deleting left at the end of a recalled line didn't set the modified flag With empty history, typing something and pressing PageUp recalled crap Added 4DOS style filename completion: Shift-Tab completes fully, then steps thru' completions - Shift-Ctrl-Tab steps backwards Shift-Copy does the same as Copy (to match !Zap defaults) Updated and tidied output from *Help Editor 2.40 Tidied up a little for release with !Zap 1.10 ('cept it's been delayed) 2.35 TaskWindow tracking workspace if now freed on soft-reset If a claim for RMA fails, a beep is emitted and something sensible done (start-up gives an error, a function key is ignored, starting a new TaskWindow reverts to the default OS_ReadLine routine) Fixed minor bug in filename expansion code (introduced in 2.34) PageUp and PageDown now don't insert start line into the history 2.34 Fixed minor bug in filename completion code Added code to find task handle for RISC OS 2 (easier than I thought) Now marks "current" buffer with a '*' in *Recall and *EdStatus if local history is in operation Recoded filename completion to enter characters directly, rather than via keyboard buffer 2.33 First beta release