Progress on bug#3521? mode-line updates break process-send-region

Progress on bug#3521? mode-line updates break process-send-region

[Tassilo Horn]

Hi all,

did anyone have a look at this bug?  IMO it should be fixed before 23.1
is released.

I can still reproduce it using emacs -Q (GNU Emacs 23.1.50.1
(x86_64-pc-linux-gnu, GTK+ Version 2.16.1) of 2009-06-28 on localhost)
updated and compiled 20 minutes ago with the recipe I posted in the bug
report.

Bye,
Tassilo

Re: Progress on bug#3521? mode-line updates break process-send-region

[Chong Yidong]

Tassilo Horn <tassilo@member.fsf.org> writes:

> did anyone have a look at this bug?  IMO it should be fixed before 23.1
> is released.
>
> I can still reproduce it using emacs -Q (GNU Emacs 23.1.50.1
> (x86_64-pc-linux-gnu, GTK+ Version 2.16.1) of 2009-06-28 on localhost)
> updated and compiled 20 minutes ago with the recipe I posted in the bug
> report.

Please do a search of prior revisions and find the change where this
problem first arose.  Thanks.

Re: Progress on bug#3521? mode-line updates break process-send-region

[Tassilo Horn]

Chong Yidong <cyd@stupidchicken.com> writes:

Hi Chong,

>> did anyone have a look at this bug?  IMO it should be fixed before
>> 23.1 is released.
>>
>> I can still reproduce it using emacs -Q (GNU Emacs 23.1.50.1
>> (x86_64-pc-linux-gnu, GTK+ Version 2.16.1) of 2009-06-28 on
>> localhost) updated and compiled 20 minutes ago with the recipe I
>> posted in the bug report.
>
> Please do a search of prior revisions and find the change where this
> problem first arose.  Thanks.

I'll see what I can do.  I've tried with emacs 22.3, and there I cannot
reproduce it.  So it seems to be introduced some times back in the 23
branch.  I remember, I've suffered from it at least about one year ago,
but I always suspected my local imap server (dovecot)...

Bye,
Tassilo

bug#3521: Progress on bug#3521? mode-line updates break process-send-region

[Chong Yidong]

Tassilo Horn <tassilo@member.fsf.org> writes:

>> Please do a search of prior revisions and find the change where this
>> problem first arose.  Thanks.
>
> I'll see what I can do.  I've tried with emacs 22.3, and there I cannot
> reproduce it.  So it seems to be introduced some times back in the 23
> branch.

Note, by the way, that your test case does something that you maybe did
not intend:

> (progn
>   (setq proctest
> 	(start-process "cat" (get-buffer-create "cat") "/bin/cat"))
>   (with-temp-buffer
>     (dotimes (dummy 999999)
>       (insert
>        (concat (number-to-string (random))
> 	       (number-to-string (random))
> 	       (number-to-string (random))
> 	       (number-to-string (random))
> 	       "\n")))
>     (process-send-region proctest (point-min) (point-max)))
>   (delete-process proctest))

It keeps growing the temporary buffer, and eventually you're sending
extremely large strings to the inferior process.

> Debugger entered--Lisp error: (file-error "writing to process" "bad address" #<process cat<2>>)
>   process-send-region(#<process cat<2>> 1 75145548)

On my 32-bit machine, the error occurs earlier.  But If I insert an
`erase-buffer' call into your loop, so that the size of the temporary
buffer is bounded, Emacs completes the 999999-iteration loop without any
error.

If there's any connection with mode-line updates, I have not been able
to reproduce it.

Re: Progress on bug#3521? mode-line updates break process-send-region

[Chong Yidong]

Tassilo Horn <tassilo@member.fsf.org> writes:

>> Please do a search of prior revisions and find the change where this
>> problem first arose.  Thanks.
>
> I'll see what I can do.  I've tried with emacs 22.3, and there I cannot
> reproduce it.  So it seems to be introduced some times back in the 23
> branch.

Note, by the way, that your test case does something that you maybe did
not intend:

> (progn
>   (setq proctest
> 	(start-process "cat" (get-buffer-create "cat") "/bin/cat"))
>   (with-temp-buffer
>     (dotimes (dummy 999999)
>       (insert
>        (concat (number-to-string (random))
> 	       (number-to-string (random))
> 	       (number-to-string (random))
> 	       (number-to-string (random))
> 	       "\n")))
>     (process-send-region proctest (point-min) (point-max)))
>   (delete-process proctest))

It keeps growing the temporary buffer, and eventually you're sending
extremely large strings to the inferior process.

> Debugger entered--Lisp error: (file-error "writing to process" "bad address" #<process cat<2>>)
>   process-send-region(#<process cat<2>> 1 75145548)

On my 32-bit machine, the error occurs earlier.  But If I insert an
`erase-buffer' call into your loop, so that the size of the temporary
buffer is bounded, Emacs completes the 999999-iteration loop without any
error.

If there's any connection with mode-line updates, I have not been able
to reproduce it.

bug#3521: Progress on bug#3521? mode-line updates break process-send-region

[Andreas Schwab]

Chong Yidong <cyd@stupidchicken.com> writes:

> On my 32-bit machine, the error occurs earlier.  But If I insert an
> `erase-buffer' call into your loop, so that the size of the temporary
> buffer is bounded, Emacs completes the 999999-iteration loop without any
> error.

The whole point of the test case is the big size of the buffer string.
This is a typical symptom of a GC bug.  I have checked in a fix.

Andreas.

-- 
Andreas Schwab, schwab@linux-m68k.org
GPG Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: Progress on bug#3521? mode-line updates break process-send-region

[Andreas Schwab]

Chong Yidong <cyd@stupidchicken.com> writes:

> On my 32-bit machine, the error occurs earlier.  But If I insert an
> `erase-buffer' call into your loop, so that the size of the temporary
> buffer is bounded, Emacs completes the 999999-iteration loop without any
> error.

The whole point of the test case is the big size of the buffer string.
This is a typical symptom of a GC bug.  I have checked in a fix.

Andreas.

-- 
Andreas Schwab, schwab@linux-m68k.org
GPG Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: Progress on bug#3521? mode-line updates break process-send-region

[Tassilo Horn]

Andreas Schwab <schwab@linux-m68k.org> writes:

Hi Andreas,

>> On my 32-bit machine, the error occurs earlier.  But If I insert an
>> `erase-buffer' call into your loop, so that the size of the temporary
>> buffer is bounded, Emacs completes the 999999-iteration loop without
>> any error.
>
> The whole point of the test case is the big size of the buffer string.
> This is a typical symptom of a GC bug.  I have checked in a fix.

Wow, this was quick.  Tomorrow, I'll test it and report back if the bug
is gone.

Chong wrote:
> If there's any connection with mode-line updates, I have not been able
> to reproduce it.

I already reported the bug as #2272, and Andreas checked in a fix for
it.  At that time, I couldn't reproduce it, but out of sudden, it hit me
again, and David Engster found out that it seems to be triggered when
the mode-line is updated.  He got the error because of `display-time'
and I got it when `rcirc-track-minor-mode' updated the mode-line.  So
since Andreas last fix, the update of the mode-line was needed for me to
trigger the bug again.

Bye,
Tassilo

bug#3521: Progress on bug#3521? mode-line updates break process-send-region

[Tassilo Horn]

Andreas Schwab <schwab@linux-m68k.org> writes:

Hi Andreas,

>> On my 32-bit machine, the error occurs earlier.  But If I insert an
>> `erase-buffer' call into your loop, so that the size of the temporary
>> buffer is bounded, Emacs completes the 999999-iteration loop without
>> any error.
>
> The whole point of the test case is the big size of the buffer string.
> This is a typical symptom of a GC bug.  I have checked in a fix.

Wow, this was quick.  Tomorrow, I'll test it and report back if the bug
is gone.

Chong wrote:
> If there's any connection with mode-line updates, I have not been able
> to reproduce it.

I already reported the bug as #2272, and Andreas checked in a fix for
it.  At that time, I couldn't reproduce it, but out of sudden, it hit me
again, and David Engster found out that it seems to be triggered when
the mode-line is updated.  He got the error because of `display-time'
and I got it when `rcirc-track-minor-mode' updated the mode-line.  So
since Andreas last fix, the update of the mode-line was needed for me to
trigger the bug again.

Bye,
Tassilo

bug#3521: Progress on bug#3521? mode-line updates break process-send-region

[David Engster]

Andreas Schwab <schwab@linux-m68k.org> writes:
> Chong Yidong <cyd@stupidchicken.com> writes:
>> On my 32-bit machine, the error occurs earlier.  But If I insert an
>> `erase-buffer' call into your loop, so that the size of the temporary
>> buffer is bounded, Emacs completes the 999999-iteration loop without any
>> error.
>
> The whole point of the test case is the big size of the buffer string.
> This is a typical symptom of a GC bug.  I have checked in a fix.

Thank you! The bug is fixed for me.

It doesn't really matter anymore, but for what it's worth, I think this
bug was introduced with the merge of the unicode branch on
2008-02-01. But just when I wanted to dig further, Andreas checked in
the fix. :-)

-David

Re: Progress on bug#3521? mode-line updates break process-send-region

[Tassilo Horn]

David Engster <deng@randomsample.de> writes:

Hi!

>> The whole point of the test case is the big size of the buffer
>> string.  This is a typical symptom of a GC bug.  I have checked in a
>> fix.
>
> Thank you! The bug is fixed for me.

I cannot reproduce it anymore, too.  Thanks Andreas!

Bye,
Tassilo

bug#3521: Progress on bug#3521? mode-line updates break process-send-region

[Tassilo Horn]

David Engster <deng@randomsample.de> writes:

Hi!

>> The whole point of the test case is the big size of the buffer
>> string.  This is a typical symptom of a GC bug.  I have checked in a
>> fix.
>
> Thank you! The bug is fixed for me.

I cannot reproduce it anymore, too.  Thanks Andreas!

Bye,
Tassilo