bug#24124: emacs terminal escape command error

bug#24124: emacs terminal escape command error

[modem]

Hi,


I am facing a problem with emacs inside tmux.
The problem can be reproduced easily.

Emacs seems to have an terminal issue when being back from background
mode on shell.

Let me explain:
When I put emacs in background (^z) and then in foreground again ("fg",
bash) sometime emacs will not reset properly the terminal and I won't be
able to scroll up (using PageUp/PageDown).

After debugging the issue with a tmux developer, we tracked it down to
emacs and believe this is an emacs issue with terminal handling when
going back from background mode.

I am attaching a snippet of the technical detail about this issue :

>  When you ^Z, tmux gets 112 bytes in all cases:
>
>  input_parse: %0 ground, 40 bytes:
\n\n\033[K\033[?1l\033>\033[34h\033[?25h\033[?1049l\033[39;49m\r
>  input_parse: %0 ground, 72 bytes: \r\n[1]+  Stopped
emacs large_text_file.txt\r\nroot@milu:~#
>
>  \033[?1049l is rmcup which exits cursor positioning (alternate
screen) mode.
>
>  Then when you type fg, it gets:
>
>  input_parse: %0 ground, 694 bytes:
\033[?1049h\033[34l\033[?1h\033=\033[24;1H\033[K\033[H\033[?25l\033[3mFile
Edit Options Buffers ...
>
>  \033[?1049h is smcup which turns it on again.
>
>  But sometimes, it gets:
>
>  input_parse: %0 ground, 674 bytes:
\033[24;1H\033[K\033[H\033[?25l\033[3mFile Edit Options Buffers ...
>
>  Where emacs is not sending \033[1049h, or in fact any of the smcup,
>  cvvis, smkx sequence it normally sends.

This can be reproduced by hitting ^z and then "fg" in a loop for a
couples of minutes. If you run tmux with "tmux -vvvv" you will be able
to obtain the logs mentionned above. To observe the problem you can just
press PageUp each time you are back in emacs, when the bug will
reproduce you won't pageUp inside emacs but enter the tmux's copy-mode.
My ~/.tmux.conf only has 1 line:
bind-key -T root PPage if-shell -F "#{alternate_on}" "send-keys PPage"
"copy-mode -e -u"


If you need me to apply some patch and/or reproduce just let me know I
will happily helps.

Thanks a lot,


-- modem

bug#24124: emacs terminal escape command error

[Eli Zaretskii]

> From: modem <modem@member.fsf.org>
> Date: Mon, 1 Aug 2016 16:34:43 +0200
> 
> When I put emacs in background (^z) and then in foreground again ("fg",
> bash) sometime emacs will not reset properly the terminal and I won't be
> able to scroll up (using PageUp/PageDown).
> 
> After debugging the issue with a tmux developer, we tracked it down to
> emacs and believe this is an emacs issue with terminal handling when
> going back from background mode.
> 
> I am attaching a snippet of the technical detail about this issue :
> 
> >  When you ^Z, tmux gets 112 bytes in all cases:
> >
> >  input_parse: %0 ground, 40 bytes:
> \n\n\033[K\033[?1l\033>\033[34h\033[?25h\033[?1049l\033[39;49m\r
> >  input_parse: %0 ground, 72 bytes: \r\n[1]+  Stopped
> emacs large_text_file.txt\r\nroot@milu:~#
> >
> >  \033[?1049l is rmcup which exits cursor positioning (alternate
> screen) mode.
> >
> >  Then when you type fg, it gets:
> >
> >  input_parse: %0 ground, 694 bytes:
> \033[?1049h\033[34l\033[?1h\033=\033[24;1H\033[K\033[H\033[?25l\033[3mFile
> Edit Options Buffers ...
> >
> >  \033[?1049h is smcup which turns it on again.
> >
> >  But sometimes, it gets:
> >
> >  input_parse: %0 ground, 674 bytes:
> \033[24;1H\033[K\033[H\033[?25l\033[3mFile Edit Options Buffers ...
> >
> >  Where emacs is not sending \033[1049h, or in fact any of the smcup,
> >  cvvis, smkx sequence it normally sends.

Probably because Emacs tries to optimize the cursor motion commands
for minimal cost.  See the code of the cmgoto function (in cm.c in
Emacs sources).

But you didn't tell what terminfo entry is used by Emacs on your
system, and didn't tell what the above terminal commands do on your
terminal, so it's hard to proceed in investigating this issue.

Thanks.

bug#24124: emacs terminal escape command error

[Eli Zaretskii]

[Please keep the bug address on the CC list.]

> From: modem <modem@member.fsf.org>
> Date: Tue, 2 Aug 2016 11:47:55 +0200
> 
> Thanks you for your advises.
> 
> My terminfo: "echo $TERM" return "screen". Is that it ?

Probably.

> I recompiled emacs after commenting a few optimization in src/cm.c (cmgoto).
> Unfortunately I can still reproduce the bug with this binary.
> 
> I am attaching you the patch I did.

I think you disabled the wrong code.  What you need is force Emacs to
always do this:

      if (cost <= relcost)
	{	/* really is cheaper */
	  emacs_tputs (tty, p, 1, cmputc);
	  curY (tty) = row, curX (tty) = col;
	  return;
	}

bug#24124: emacs terminal escape command error

[modem]

[-- Attachment #1: Type: text/plain, Size: 1978 bytes --]

> [Please keep the bug address on the CC list.]
>
>> Thanks you for your advises.
>> 
>> My terminfo: "echo $TERM" return "screen". Is that it ?
> 
> Probably.
> 
>> I recompiled emacs after commenting a few optimization in src/cm.c (cmgoto).
>> Unfortunately I can still reproduce the bug with this binary.
>> 
>> I am attaching you the patch I did.
> 
> I think you disabled the wrong code.  What you need is force Emacs to
> always do this:
> 
>       if (cost <= relcost)œ
> 	{	/* really is cheaper */
> 	  emacs_tputs (tty, p, 1, cmputc);
> 	  curY (tty) = row, curX (tty) = col;
> 	  return;
> 	}

Hi Eli,

Thanks a lot. So I did another patch and another try. I can still
reproduce the bug.

Any more advices on stuff I should try are welcomed ;-)


Here is the patch:

--- emacs24-24.4+1.orig/src/cm.c        2014-10-21 23:45:42.000000000 +0200
+++ /dev/shm/cm.c       2016-08-03 15:10:55.803193217 +0200
@@ -327,10 +327,12 @@
     char *p;
     const char *dcm;

+#if 0
   /* First the degenerate case */
     if (row == curY (tty) && col == curX (tty)) /* already there */
     return;
-
+#endif // XXX MODEM
+
     if (curY (tty) >= 0 && curX (tty) >= 0)
     {
       /* We may have quick ways to go to the upper-left, bottom-left,
@@ -375,7 +377,7 @@
    * In the following comparison, the = in <= is because when the costs
    * are the same, it looks nicer (I think) to move directly there.
    */
-  if (directcost <= relcost)
+    if (directcost <= relcost || 1) // XXX MODEM
     {
       /* compute REAL direct cost */
       cost = 0;
@@ -383,14 +385,14 @@
            ? tgoto (dcm, row, col)
            : tgoto (dcm, col, row));
       emacs_tputs (tty, p, 1, evalcost);
-      if (cost <= relcost)
+      if (cost <= relcost || 1) // XXX MODEM
        {       /* really is cheaper */
          emacs_tputs (tty, p, 1, cmputc);
          curY (tty) = row, curX (tty) = col;
          return;
        }
     }
-
+
   switch (use)
     {
     case USEHOME:

[-- Attachment #2: second_try.patch --]
[-- Type: text/x-patch, Size: 1158 bytes --]

--- emacs24-24.4+1.orig/src/cm.c	2014-10-21 23:45:42.000000000 +0200
+++ /dev/shm/cm.c	2016-08-03 15:10:55.803193217 +0200
@@ -327,10 +327,12 @@
     char *p;
     const char *dcm;
 
+#if 0
   /* First the degenerate case */
     if (row == curY (tty) && col == curX (tty)) /* already there */
     return;
-
+#endif // XXX MODEM
+    
     if (curY (tty) >= 0 && curX (tty) >= 0)
     {
       /* We may have quick ways to go to the upper-left, bottom-left,
@@ -375,7 +377,7 @@
    * In the following comparison, the = in <= is because when the costs
    * are the same, it looks nicer (I think) to move directly there.
    */
-  if (directcost <= relcost)
+    if (directcost <= relcost || 1) // XXX MODEM
     {
       /* compute REAL direct cost */
       cost = 0;
@@ -383,14 +385,14 @@
            ? tgoto (dcm, row, col)
            : tgoto (dcm, col, row));
       emacs_tputs (tty, p, 1, evalcost);
-      if (cost <= relcost)
+      if (cost <= relcost || 1) // XXX MODEM
 	{	/* really is cheaper */
 	  emacs_tputs (tty, p, 1, cmputc);
 	  curY (tty) = row, curX (tty) = col;
 	  return;
 	}
     }
-
+  
   switch (use)
     {
     case USEHOME:

bug#24124: emacs terminal escape command error

[Eli Zaretskii]

> From: modem <modem@member.fsf.org>
> Date: Wed, 3 Aug 2016 15:21:24 +0200
> 
> > I think you disabled the wrong code.  What you need is force Emacs to
> > always do this:
> > 
> >       if (cost <= relcost)œ
> > 	{	/* really is cheaper */
> > 	  emacs_tputs (tty, p, 1, cmputc);
> > 	  curY (tty) = row, curX (tty) = col;
> > 	  return;
> > 	}
> 
> Hi Eli,
> 
> Thanks a lot. So I did another patch and another try. I can still
> reproduce the bug.
> 
> Any more advices on stuff I should try are welcomed ;-)

I'm stumped.  It could be some problem with some other piece of
software, not with Emacs.

Is there someone you could ask to produce a human-readable description
of the command Emacs sent to the terminal driver, both in the "good"
and in the "bad" cases?  I mean this citation which you brought
originally, but without any details:

> I am attaching a snippet of the technical detail about this issue :
> 
> >  When you ^Z, tmux gets 112 bytes in all cases:
> >
> >  input_parse: %0 ground, 40 bytes:
> \n\n\033[K\033[?1l\033>\033[34h\033[?25h\033[?1049l\033[39;49m\r
> >  input_parse: %0 ground, 72 bytes: \r\n[1]+  Stopped
> emacs large_text_file.txt\r\nroot@milu:~#
> >
> >  \033[?1049l is rmcup which exits cursor positioning (alternate
> screen) mode.
> >
> >  Then when you type fg, it gets:
> >
> >  input_parse: %0 ground, 694 bytes:
> \033[?1049h\033[34l\033[?1h\033=\033[24;1H\033[K\033[H\033[?25l\033[3mFile
> Edit Options Buffers ...
> >
> >  \033[?1049h is smcup which turns it on again.
> >
> >  But sometimes, it gets:
> >
> >  input_parse: %0 ground, 674 bytes:
> \033[24;1H\033[K\033[H\033[?25l\033[3mFile Edit Options Buffers ...
> >
> >  Where emacs is not sending \033[1049h, or in fact any of the smcup,
> >  cvvis, smkx sequence it normally sends.

It would be nice to be able to understand what was meant by whoever
wrote this, as smcup, cvvis, etc. don't tell me much (I'm not a
terminfo expert).

Thanks.

bug#24124: emacs terminal escape command error

[Stefan Kangas]

Eli Zaretskii <eliz@gnu.org> writes:

> [Please keep the bug address on the CC list.]
>
>> From: modem <modem@member.fsf.org>
>> Date: Tue, 2 Aug 2016 11:47:55 +0200
>> 
>> Thanks you for your advises.
>> 
>> My terminfo: "echo $TERM" return "screen". Is that it ?
>
> Probably.
>
>> I recompiled emacs after commenting a few optimization in src/cm.c (cmgoto).
>> Unfortunately I can still reproduce the bug with this binary.
>> 
>> I am attaching you the patch I did.
>
> I think you disabled the wrong code.  What you need is force Emacs to
> always do this:
>
>       if (cost <= relcost)
> 	{	/* really is cheaper */
> 	  emacs_tputs (tty, p, 1, cmputc);
> 	  curY (tty) = row, curX (tty) = col;
> 	  return;
> 	}

That was 3 years ago.  Did you ever get a chance to try the suggested
fix by Eli above?

Best regards,
Stefan Kangas

bug#24124: emacs terminal escape command error

[Stefan Kangas]

Stefan Kangas <stefan@marxist.se> writes:

> That was 3 years ago.  Did you ever get a chance to try the suggested
> fix by Eli above?

More information was requested, but none was given within 10 weeks, so
I'm closing this bug.  If this is still an issue, please reply to this
email (use "Reply to all" in your email client) and we can reopen the
bug report.

Best regards,
Stefan Kangas