Date:	Wed, 25 Oct 1995 16:23:01 +0200 (IST)
From:	Eli Zaretskii <eliz AT is DOT elta DOT co DOT il>
To:	Richard Stallman <rms AT gnu DOT ai DOT mit DOT edu>
Cc:	jlund AT math DOT ku DOT dk, kfstorm AT gnu DOT ai DOT mit DOT edu, hniss AT diku DOT dk,
	Morten Welinder <terra AT diku DOT dk>, djgpp AT sun DOT soe DOT clarkson DOT edu
Subject:	Re: Error, Emacs for DOS, Random upper case letters

On Sun, 22 Oct 1995, Richard Stallman wrote:

> I've written a patch that ought to record the data that dos_rawgetc
> gets from the system.  I cannot test it or even compile it; I have to
> depend on you to do that and to fix any small bugs it may have.

(For those who see this for the first time: the problem was that 
extraneous characters would appear in the buffer when a key was pressed 
during auto-save in the DOS port of Emacs (like if you press PageDown to 
scroll the buffer which shows a modified file).

I used this patch to investigate the problem.  I'll post a full report 
tomorrow (got to run now), but one thing is certain: this is somehow 
connected to the way the terminal bell is rung on a DOS machine.  The 
problem never happens if you use visible bell; even under non-visible 
bell, it never happens until the bell is rung for the first time.  I 
really don't understand how these two are connected, but DJGPP reprograms 
the keyboard interrupt to catch the keystrokes, so it might be something 
special to DJGPP.  The way the bell is rung in DOS version of Emacs is by 
writing the bell character ('\07') to the standard output with a call to 
write(); this boils down to calling DOS function 40h to write a single 
character to stdout.  I managed to solve the problem without 
understanding it by using different DOS functions (2 or 6) to write the 
same character to stdout.

If anybody can shed any light on this strange phenomenon, I'd appreciate 
it, because I cannot be 100% sure I solved the bug without understanding 
it.  Otherwise, I'd say let's wait for DJGPP v2.0 and see if the problem 
reappears then, because DJGPP v1.x is a dead product, and there is no 
sense in debugging its library (which was extensively rewritten for 
v2.0).

- Raw text -