Eshell broken - piping?

Eshell broken - piping?

[Wojciech Meyer]

Hi,

I've noticed that examples from the wiki + sample command from comments
in Eshell source code don't not work anymore, e.g:

cat eshell-io.el | wc -l # taken from source code

just hangs (probably wc -l is waiting for input)

for i in 1 2 3 { grep -q a b && *echo has it } | wc -l #
http://www.gnu.org/software/emacs/manual/html_node/eshell/Known-problems.html

does not work either.

Is it something you are aware of, or maybe it used to work but broke on
the way? (I can do bisecting at some point, but before this I would like
to know if that worked for somebody)

TIA 

Wojciech

Re: Eshell broken - piping?

[Óscar Fuentes]

Wojciech Meyer <wojciech.meyer@googlemail.com> writes:

> I've noticed that examples from the wiki + sample command from comments
> in Eshell source code don't not work anymore, e.g:
>
> cat eshell-io.el | wc -l # taken from source code
>
> just hangs (probably wc -l is waiting for input)

It works fine here

GNU Emacs 24.0.50.1 (x86_64-unknown-linux-gnu, X toolkit) of 2010-07-30

BTW, from where is eshell-io.el ? I have lisp/eshell/esh-io.el.

[snip]

Re: Eshell broken - piping?

[Wojciech Meyer]

Óscar Fuentes <ofv@wanadoo.es> writes:

> The following message is a courtesy copy of an article
> that has been posted to gmane.emacs.devel as well.
>
> Wojciech Meyer <wojciech.meyer@googlemail.com> writes:
>
>> I've noticed that examples from the wiki + sample command from comments
>> in Eshell source code don't not work anymore, e.g:
>>
>> cat eshell-io.el | wc -l # taken from source code
>>
>> just hangs (probably wc -l is waiting for input)
>
> It works fine here
>
> GNU Emacs 24.0.50.1 (x86_64-unknown-linux-gnu, X toolkit) of 2010-07-30
>
> BTW, from where is eshell-io.el ? I have lisp/eshell/esh-io.el.
>

That was just a random example.

I don't really understand why it does not here, however I don't remember
if that was working. I use bzr HEAD. Problem is verified here with emacs
-Q. I will file a bug then. Is anybody else who got it working?

> [snip]

TIA
Wojciech

Re: Eshell broken - piping?

[Chad Brown]

On Jul 31, 2010, at 5:46 AM, Wojciech Meyer wrote:

> That was just a random example.
> 
> I don't really understand why it does not here, however I don't remember
> if that was working. I use bzr HEAD. Problem is verified here with emacs
> -Q. I will file a bug then. Is anybody else who got it working?

I can confirm that it is not working for me on bzr head, either.

I've got to run out the door, but when I just tested ``ls -al | tee /tmp/foo'' it 
seemed like it might be failing in an interesting way.

Hope that helps,
*Chad

Re: Eshell broken - piping?

[Jan Djärv]

It seems that the old "disable icanon" is responsible for this also.
If you enable icanon in terminal setup, things work as expected.

Can we please enable icanon?  Any possible data loss (still haven't heard of a 
real case where this happens) must be solved in another way.

	Jan D.


Chad Brown skrev 2010-07-31 18.03:
>
> On Jul 31, 2010, at 5:46 AM, Wojciech Meyer wrote:
>
>> That was just a random example.
>>
>> I don't really understand why it does not here, however I don't remember
>> if that was working. I use bzr HEAD. Problem is verified here with emacs
>> -Q. I will file a bug then. Is anybody else who got it working?
>
> I can confirm that it is not working for me on bzr head, either.
>
> I've got to run out the door, but when I just tested ``ls -al | tee /tmp/foo'' it
> seemed like it might be failing in an interesting way.
>
> Hope that helps,
> *Chad

Re: Eshell broken - piping?

[Óscar Fuentes]

Jan Djärv <jan.h.d@swipnet.se> writes:

> It seems that the old "disable icanon" is responsible for this also.
> If you enable icanon in terminal setup, things work as expected.
>
> Can we please enable icanon?  Any possible data loss (still haven't
> heard of a real case where this happens) must be solved in another
> way.

Blush...

I'm using a build with one commit reversed, because otherwise magit.el
hangs often while invoking git. I just tried the sources from bzr and,
indeed, cat foo | wc hangs. So they seem the same problem.

The bug was documented on magit's tracker here:

http://github.com/philjackson/magit/issues#issue/18

This is the revision that caused havoc on magit (and on eshell as well):

http://repo.or.cz/w/emacs.git/commitdiff/ee1b6e21aa2b831499659b7f58ec94b95016cec7

+2010-06-01  Stefan Monnier  <monnier@iro.umontreal.ca>
+
+       * sysdep.c (child_setup_tty): Move the non-canonical initialization to
+       the HAVE_TERMIO where it belongs (bug#6149).

Re: Eshell broken - piping?

[Wojciech Meyer]

Óscar Fuentes <ofv@wanadoo.es> writes:

> Jan Djärv <jan.h.d@swipnet.se> writes:
>
>> It seems that the old "disable icanon" is responsible for this also.
>> If you enable icanon in terminal setup, things work as expected.
>>
>> Can we please enable icanon?  Any possible data loss (still haven't
>> heard of a real case where this happens) must be solved in another
>> way.
>
> Blush...
>
> I'm using a build with one commit reversed, because otherwise magit.el
> hangs often while invoking git. I just tried the sources from bzr and,
> indeed, cat foo | wc hangs. So they seem the same problem.
>
> The bug was documented on magit's tracker here:
>
> http://github.com/philjackson/magit/issues#issue/18
>
> This is the revision that caused havoc on magit (and on eshell as well):
>
> http://repo.or.cz/w/emacs.git/commitdiff/ee1b6e21aa2b831499659b7f58ec94b95016cec7
>
> +2010-06-01  Stefan Monnier  <monnier@iro.umontreal.ca>
> +
> +       * sysdep.c (child_setup_tty): Move the non-canonical initialization to
> +       the HAVE_TERMIO where it belongs (bug#6149).

Thanks for this, it will make life easier. (Yes, I tried to attack the
problem, from the lisp side, however my search was not fruit-full).

BTW: Looking at the commit, it looks like it is activated only in TTY -
and I found the problem in GTK+ emacs?

Cheers;
Wojciech

Re: Eshell broken - piping?

[Jan Djärv]

Wojciech Meyer skrev 2010-07-31 20.52:

> BTW: Looking at the commit, it looks like it is activated only in TTY -
> and I found the problem in GTK+ emacs?

Emacs communicates with subprocess with pty:s.

	Jan D.

Re: Eshell broken - piping?

[Óscar Fuentes]

Wojciech Meyer <wojciech.meyer@googlemail.com> writes:

[snip]

> BTW: Looking at the commit, it looks like it is activated only in TTY -
> and I found the problem in GTK+ emacs?

AFAIK it affects how Emacs interacts with running applications,
including controlling pipes in eshell's case.

Re: Eshell broken - piping?

[Wojciech Meyer]

Óscar Fuentes <ofv@wanadoo.es> writes:

> Jan Djärv <jan.h.d@swipnet.se> writes:
>
>> It seems that the old "disable icanon" is responsible for this also.
>> If you enable icanon in terminal setup, things work as expected.
>>
>> Can we please enable icanon?  Any possible data loss (still haven't
>> heard of a real case where this happens) must be solved in another
>> way.
>
> Blush...
>
> I'm using a build with one commit reversed, because otherwise magit.el
> hangs often while invoking git. I just tried the sources from bzr and,
> indeed, cat foo | wc hangs. So they seem the same problem.
>
> The bug was documented on magit's tracker here:
>
> http://github.com/philjackson/magit/issues#issue/18
>
> This is the revision that caused havoc on magit (and on eshell as well):
>
> http://repo.or.cz/w/emacs.git/commitdiff/ee1b6e21aa2b831499659b7f58ec94b95016cec7
>
> +2010-06-01  Stefan Monnier  <monnier@iro.umontreal.ca>
> +
> +       * sysdep.c (child_setup_tty): Move the non-canonical initialization to
> +       the HAVE_TERMIO where it belongs (bug#6149).

FYI: I've reverted this change and it works now.
(and I more fond to use Eshell now)

Cheers;
Wojciech