[Babel] No output returned if just one command is failing

[Babel] No output returned if just one command is failing

[Sébastien Vauban]

#+TITLE:     Babel doesn't return output if one command fails
#+DATE:      2010-12-01
#+LANGUAGE:  en_US

* Abstract

When just one command fails in Babel, then there is no shell output at all.

* One command fails

For example, the last command (=datee=) is not found:

#+begin_src sh :var file=(buffer-file-name) :results output :exports both
echo $(basename $file)
datee
#+end_src

#+results:

Result set is empty, while the =echo= command did work, and did produce
output.

* All commands are successful

Only if all commands are successful, I see all results:

#+begin_src sh :var file=(buffer-file-name) :results output :exports both
echo $(basename $file)
date
#+end_src

#+results:
#+begin_example
ecm-babel-one-error-no-results.txt
Wed, Dec 01, 2010 10:21:09 AM
#+end_example

Best regards,
  Seb

-- 
Sébastien Vauban


_______________________________________________
Emacs-orgmode mailing list
Please use `Reply All' to send replies to the list.
Emacs-orgmode-mXXj517/zsQ@public.gmane.org
http://lists.gnu.org/mailman/listinfo/emacs-orgmode

Re: [Babel] No output returned if just one command is failing

[Eric Schulte]

Hi Seb,

I don't forsee adding partial results insertion both because

- it would add a good deal of complexity to the code to insert results
  part-way through a run

- the current behavior of only inserting results on a fully successful
  run is reasonable and is probably more obvious (at least to me) than
  inserting partial results

Best -- Eric

Sébastien Vauban <wxhgmqzgwmuf@spammotel.com> writes:

> #+TITLE:     Babel doesn't return output if one command fails
> #+DATE:      2010-12-01
> #+LANGUAGE:  en_US
>
> * Abstract
>
> When just one command fails in Babel, then there is no shell output at all.
>
> * One command fails
>
> For example, the last command (=datee=) is not found:
>
> #+begin_src sh :var file=(buffer-file-name) :results output :exports both
> echo $(basename $file)
> datee
> #+end_src
>
> #+results:
>
> Result set is empty, while the =echo= command did work, and did produce
> output.
>
> * All commands are successful
>
> Only if all commands are successful, I see all results:
>
> #+begin_src sh :var file=(buffer-file-name) :results output :exports both
> echo $(basename $file)
> date
> #+end_src
>
> #+results:
> #+begin_example
> ecm-babel-one-error-no-results.txt
> Wed, Dec 01, 2010 10:21:09 AM
> #+end_example
>
>
> Best regards,
>   Seb

Re: [Babel] No output returned if just one command is failing

[Sébastien Vauban]

Hi Eric,

"Eric Schulte" wrote:
> I don't forsee adding partial results insertion both because
>
> - it would add a good deal of complexity to the code to insert results
>   part-way through a run

I can't comment on this, of course.

> - the current behavior of only inserting results on a fully successful
>   run is reasonable and is probably more obvious (at least to me) than
>   inserting partial results

Being fond of Babel, I'm using it always, everywhere. I prefer:

1. typing my shell commands in an Org buffer,
2. evaluate the block,
3. get the results automagically inserted in the buffer,
4. (eventually, version the whole file for later comparisons when updating the
   code),
5. export the whole to HTML and/or PDF.

The current behavior, even if totally respectable and defendable, inhibits
such a way of working: if you write (or update) a shell code, and don't see
(more or less) the same things as the ones you would see in a shell buffer,
then you can't use such an Org buffer -- as long as one command fails.

I don't especially want you to change your position, but I'm explaining the
"negative" consequences for me.

Thanks!

Best regards,
  Seb

-- 
Sébastien Vauban


_______________________________________________
Emacs-orgmode mailing list
Please use `Reply All' to send replies to the list.
Emacs-orgmode-mXXj517/zsQ@public.gmane.org
http://lists.gnu.org/mailman/listinfo/emacs-orgmode

Re: [Babel] No output returned if just one command is failing

[Sébastien Vauban]

Hi Eric,

Sébastien Vauban wrote:
> "Eric Schulte" wrote:
>> I don't forsee adding partial results insertion both because
>>
>> - it would add a good deal of complexity to the code to insert results
>>   part-way through a run
>
> I can't comment on this, of course.
>
>> - the current behavior of only inserting results on a fully successful run
>>   is reasonable and is probably more obvious (at least to me) than
>>   inserting partial results
>
> Being fond of Babel, I'm using it always, everywhere. I prefer:
>
> 1. typing my shell commands in an Org buffer,
> 2. evaluate the block,
> 3. get the results automagically inserted in the buffer,
> 4. (eventually, version the whole file for later comparisons when updating
>    the code),
> 5. export the whole to HTML and/or PDF.
>
> The current behavior, even if totally respectable and defendable, inhibits
> such a way of working: if you write (or update) a shell code, and don't see
> (more or less) the same things as the ones you would see in a shell buffer,
> then you can't use such an Org buffer -- as long as one command fails.
>
> I don't especially want you to change your position, but I'm explaining the
> "negative" consequences for me.

BTW, what's the internal "definition" (in Org Babel) of "unsuccessful run" of
a (sh) command block?

Is it "return code != 0"?

If yes, such a sh block would never produce any results when the machine is
not a known name (as first command would "fail"):

#+begin_src sh :var machine :results output
ping $machine
rc=$?
if [ $rc == 0 ]; then
    echo "Machine pinged successfully with its name...";
else
    echo "Trying to ping machine by its IP...";
    ipmachine=$(grep whichever-list.txt | cut -f 2);
    ping $ipmachine
fi
#+end_src

(the above is just a simply sample of code which includes tests of rc)

Though, it does currently work with both known and unknown machines. So I'm
clearly missing something here.

Best regards,
  Seb

-- 
Sébastien Vauban


_______________________________________________
Emacs-orgmode mailing list
Please use `Reply All' to send replies to the list.
Emacs-orgmode-mXXj517/zsQ@public.gmane.org
http://lists.gnu.org/mailman/listinfo/emacs-orgmode

Re: [Babel] No output returned if just one command is failing

[Dan Davison]

Hi Seb, I definitely have some sympathy with your request. On two
occasions I've had to manually make this change just to carry on
working. The change I made is straightforward if you need it as a
temporary hack:

--8<---------------cut here---------------start------------->8---
diff --git a/lisp/ob-eval.el b/lisp/ob-eval.el
index 8832d91..1cce003 100644
--- a/lisp/ob-eval.el
+++ b/lisp/ob-eval.el
@@ -57,7 +57,7 @@ STDERR with `org-babel-eval-error-notify'."
          (progn
            (with-current-buffer err-buff
              (org-babel-eval-error-notify exit-code (buffer-string)))
-           nil)
+           (buffer-string))
        (buffer-string)))))
 
 (defun org-babel-eval-read-file (file)
--8<---------------cut here---------------end--------------->8---

But do we actually change babel in this direction? It's important to
distinguish between :results output and :results value here. The change
that seems superficially attractive is to allow :results output to
output the contents of standard output, even if there has been an
error. After all, stdout might contain some useful data, and currently
babel refuses point blank to give it to you if there's been an
error. And as you say, this is the behavior we are used to in the
shell. This hypothetical change would leave the default :results value
alone (so the above patch would need modification).

The thing is that babel currently has a clear, simple, rule which says:
if there's an error, the result is the elisp value nil.

Eric and I have discussed in the past whether there should be any change
in this direction. The idea of a :debug header arg has been floated,
that would allow this behavior. Or tacking stdout on to the error
output. I tend to think that the behavior you request does need to be
made available, somehow, whether by default or not.

Dan

Sébastien Vauban <wxhgmqzgwmuf-geNee64TY+gS+FvcfC7Uqw@public.gmane.org>
writes:

> Hi Eric,
>
> "Eric Schulte" wrote:
>> I don't forsee adding partial results insertion both because
>>
>> - it would add a good deal of complexity to the code to insert results
>>   part-way through a run
>
> I can't comment on this, of course.
>
>> - the current behavior of only inserting results on a fully successful
>>   run is reasonable and is probably more obvious (at least to me) than
>>   inserting partial results
>
> Being fond of Babel, I'm using it always, everywhere. I prefer:
>
> 1. typing my shell commands in an Org buffer,
> 2. evaluate the block,
> 3. get the results automagically inserted in the buffer,
> 4. (eventually, version the whole file for later comparisons when updating the
>    code),
> 5. export the whole to HTML and/or PDF.
>
> The current behavior, even if totally respectable and defendable, inhibits
> such a way of working: if you write (or update) a shell code, and don't see
> (more or less) the same things as the ones you would see in a shell buffer,
> then you can't use such an Org buffer -- as long as one command fails.
>
> I don't especially want you to change your position, but I'm explaining the
> "negative" consequences for me.
>
> Thanks!
>
> Best regards,
>   Seb

Re: [Babel] No output returned if just one command is failing

[Eric Schulte]

Hi,

You both make good points in favor of this behavior, and I had no idea
this would be as easy as Dan's patch below, I was thinking that some
sort of intermediate results return would be required.

I would be open to either of Dan's suggested changes
1. return partial results for ":results output"
2. return partial results controlled by some new header argument

I am worried by the prospect of partial results being used by another
code block in chained code block execution.  Perhaps some special care
should be taken to ensure that this will not happen, or perhaps this
should happen only in case two above where the user has explicitly
specified that partial results are OK.

Thoughts? -- Eric

Dan Davison <dandavison7@gmail.com> writes:

> Hi Seb, I definitely have some sympathy with your request. On two
> occasions I've had to manually make this change just to carry on
> working. The change I made is straightforward if you need it as a
> temporary hack:
>
> diff --git a/lisp/ob-eval.el b/lisp/ob-eval.el
> index 8832d91..1cce003 100644
> --- a/lisp/ob-eval.el
> +++ b/lisp/ob-eval.el
> @@ -57,7 +57,7 @@ STDERR with `org-babel-eval-error-notify'."
>           (progn
>             (with-current-buffer err-buff
>               (org-babel-eval-error-notify exit-code (buffer-string)))
> -           nil)
> +           (buffer-string))
>         (buffer-string)))))
>  
>  (defun org-babel-eval-read-file (file)
>
> But do we actually change babel in this direction? It's important to
> distinguish between :results output and :results value here. The change
> that seems superficially attractive is to allow :results output to
> output the contents of standard output, even if there has been an
> error. After all, stdout might contain some useful data, and currently
> babel refuses point blank to give it to you if there's been an
> error. And as you say, this is the behavior we are used to in the
> shell. This hypothetical change would leave the default :results value
> alone (so the above patch would need modification).
>
> The thing is that babel currently has a clear, simple, rule which says:
> if there's an error, the result is the elisp value nil.
>
> Eric and I have discussed in the past whether there should be any change
> in this direction. The idea of a :debug header arg has been floated,
> that would allow this behavior. Or tacking stdout on to the error
> output. I tend to think that the behavior you request does need to be
> made available, somehow, whether by default or not.
>
> Dan
>
> Sébastien Vauban <wxhgmqzgwmuf-geNee64TY+gS+FvcfC7Uqw@public.gmane.org>
> writes:
>
>> Hi Eric,
>>
>> "Eric Schulte" wrote:
>>> I don't forsee adding partial results insertion both because
>>>
>>> - it would add a good deal of complexity to the code to insert results
>>>   part-way through a run
>>
>> I can't comment on this, of course.
>>
>>> - the current behavior of only inserting results on a fully successful
>>>   run is reasonable and is probably more obvious (at least to me) than
>>>   inserting partial results
>>
>> Being fond of Babel, I'm using it always, everywhere. I prefer:
>>
>> 1. typing my shell commands in an Org buffer,
>> 2. evaluate the block,
>> 3. get the results automagically inserted in the buffer,
>> 4. (eventually, version the whole file for later comparisons when updating the
>>    code),
>> 5. export the whole to HTML and/or PDF.
>>
>> The current behavior, even if totally respectable and defendable, inhibits
>> such a way of working: if you write (or update) a shell code, and don't see
>> (more or less) the same things as the ones you would see in a shell buffer,
>> then you can't use such an Org buffer -- as long as one command fails.
>>
>> I don't especially want you to change your position, but I'm explaining the
>> "negative" consequences for me.
>>
>> Thanks!
>>
>> Best regards,
>>   Seb

Re: [Babel] No output returned if just one command is failing

[Sébastien Vauban]

Hi Eric and Dan,

Dan Davison wrote:
> Sébastien Vauban wrote:
>> "Eric Schulte" wrote:
>>> I don't forsee adding partial results insertion both because
>>>
>>> - it would add a good deal of complexity to the code to insert results
>>>   part-way through a run
>>
>> I can't comment on this, of course.
>>
>>> - the current behavior of only inserting results on a fully successful run
>>>   is reasonable and is probably more obvious (at least to me) than
>>>   inserting partial results
>>
>> Being fond of Babel, I'm using it always, everywhere. I prefer:
>>
>> 1. typing my shell commands in an Org buffer,
>> 2. evaluate the block,
>> 3. get the results automagically inserted in the buffer,
>> 4. (eventually, version the whole file for later comparisons when updating
>>    the code),
>> 5. export the whole to HTML and/or PDF.
>>
>> The current behavior, even if totally respectable and defendable, inhibits
>> such a way of working: if you write (or update) a shell code, and don't see
>> (more or less) the same things as the ones you would see in a shell buffer,
>> then you can't use such an Org buffer -- as long as one command fails.
>>
>> I don't especially want you to change your position, but I'm explaining the
>> "negative" consequences for me.
>
> I definitely have some sympathy with your request. On two occasions I've had
> to manually make this change just to carry on working.
>
> But do we actually change babel in this direction? [...]
>
> The thing is that babel currently has a clear, simple, rule which says: if
> there's an error, the result is the elisp value nil.
>
> Eric and I have discussed in the past whether there should be any change
> in this direction. The idea of a :debug header arg has been floated,
> that would allow this behavior. Or tacking stdout on to the error
> output. I tend to think that the behavior you request does need to be
> made available, somehow, whether by default or not.

If find this conclusion a bit contradictory with the fact that you even needed
it yourself. I'm not talking of having this by default, but somehow possible.
But, I know, we (you) have to ensure everything stays coherent, and somehow
simple...

BTW, what's the internal "definition" (in Org Babel) of "unsuccessful run" of
a (sh) command block?

Is it "return code != 0"?

If yes, such a sh block would never produce any results when the machine is
not a known name (as first command would "fail"):

#+begin_src sh :var machine :results output
ping $machine
rc=$?
if [ $rc == 0 ]; then
    echo "Machine pinged successfully with its name...";
else
    echo "Trying to ping machine by its IP...";
    ipmachine=$(grep whichever-list.txt | cut -f 2);
    ping $ipmachine
fi
#+end_src

(the above is just a simply sample of code which includes tests of rc)

Though, it does currently work with both known and unknown machines. So I'm
clearly missing something here.

Best regards,
  Seb

PS- I already sent this (the last paragraphs) hours ago. It did not reach the
    ML. Is it a Gmane problem?  Or something else (Gnus)?

-- 
Sébastien Vauban


_______________________________________________
Emacs-orgmode mailing list
Please use `Reply All' to send replies to the list.
Emacs-orgmode-mXXj517/zsQ@public.gmane.org
http://lists.gnu.org/mailman/listinfo/emacs-orgmode

Re: [Babel] No output returned if just one command is failing

[Dan Davison]

Sébastien Vauban <wxhgmqzgwmuf-geNee64TY+gS+FvcfC7Uqw@public.gmane.org>
writes:

> Hi Eric and Dan,
>
> Dan Davison wrote:
>> Sébastien Vauban wrote:
>>> "Eric Schulte" wrote:
>>>> I don't forsee adding partial results insertion both because
>>>>
>>>> - it would add a good deal of complexity to the code to insert results
>>>>   part-way through a run
>>>
>>> I can't comment on this, of course.
>>>
>>>> - the current behavior of only inserting results on a fully successful run
>>>>   is reasonable and is probably more obvious (at least to me) than
>>>>   inserting partial results
>>>
>>> Being fond of Babel, I'm using it always, everywhere. I prefer:
>>>
>>> 1. typing my shell commands in an Org buffer,
>>> 2. evaluate the block,
>>> 3. get the results automagically inserted in the buffer,
>>> 4. (eventually, version the whole file for later comparisons when updating
>>>    the code),
>>> 5. export the whole to HTML and/or PDF.
>>>
>>> The current behavior, even if totally respectable and defendable, inhibits
>>> such a way of working: if you write (or update) a shell code, and don't see
>>> (more or less) the same things as the ones you would see in a shell buffer,
>>> then you can't use such an Org buffer -- as long as one command fails.
>>>
>>> I don't especially want you to change your position, but I'm explaining the
>>> "negative" consequences for me.
>>
>> I definitely have some sympathy with your request. On two occasions I've had
>> to manually make this change just to carry on working.
>>
>> But do we actually change babel in this direction? [...]
>>
>> The thing is that babel currently has a clear, simple, rule which says: if
>> there's an error, the result is the elisp value nil.
>>
>> Eric and I have discussed in the past whether there should be any change
>> in this direction. The idea of a :debug header arg has been floated,
>> that would allow this behavior. Or tacking stdout on to the error
>> output. I tend to think that the behavior you request does need to be
>> made available, somehow, whether by default or not.
>
> If find this conclusion a bit contradictory with the fact that you even needed
> it yourself.

Hey Seb -- you mis-read me there. I was agreeing with you that the
behavior should be made available.

Re: [Babel] No output returned if just one command is failing

[Sébastien Vauban]

Hi Dan and Eric,

Dan Davison wrote:
> Sébastien Vauban writes:
>> Dan Davison wrote:
>>> Sébastien Vauban wrote:
>>>> "Eric Schulte" wrote:
>>>>> I don't forsee adding partial results insertion both because
>>>>>
>>>>> - it would add a good deal of complexity to the code to insert results
>>>>>   part-way through a run
>>>>
>>>> I can't comment on this, of course.
>>>>
>>>>> - the current behavior of only inserting results on a fully successful run
>>>>>   is reasonable and is probably more obvious (at least to me) than
>>>>>   inserting partial results
>>>>
>>>> Being fond of Babel, I'm using it always, everywhere. I prefer:
>>>>
>>>> 1. typing my shell commands in an Org buffer,
>>>> 2. evaluate the block,
>>>> 3. get the results automagically inserted in the buffer,
>>>> 4. (eventually, version the whole file for later comparisons when updating
>>>>    the code),
>>>> 5. export the whole to HTML and/or PDF.
>>>>
>>>> The current behavior, even if totally respectable and defendable, inhibits
>>>> such a way of working: if you write (or update) a shell code, and don't see
>>>> (more or less) the same things as the ones you would see in a shell buffer,
>>>> then you can't use such an Org buffer -- as long as one command fails.
>>>>
>>>> I don't especially want you to change your position, but I'm explaining the
>>>> "negative" consequences for me.
>>>
>>> I definitely have some sympathy with your request. On two occasions I've had
>>> to manually make this change just to carry on working.
>>>
>>> But do we actually change babel in this direction? [...]
>>>
>>> The thing is that babel currently has a clear, simple, rule which says: if
>>> there's an error, the result is the elisp value nil.
>>>
>>> Eric and I have discussed in the past whether there should be any change
>>> in this direction. The idea of a :debug header arg has been floated,
>>> that would allow this behavior. Or tacking stdout on to the error
>>> output. I tend to think that the behavior you request does need to be
>>> made available, somehow, whether by default or not.
>>
>> If find this conclusion a bit contradictory with the fact that you even needed
>> it yourself.
>
> Hey Seb -- you mis-read me there. I was agreeing with you that the
> behavior should be made available.

Yes, it seems I did add a implicit *not* in your last sentence...

My impressions are that shell blocks may be a totally different beast from the
other source blocks. If so, it could legitimate a slightly different treatment
(or default set of options).

All of this discussion once again turns around the handling of errors, and the
two streams (stdout and stderr). Aaaah, here's well a proof it's different
from the rest (languages such as emacs-lisp), at least in their classic usage?

My dream would be to be able to write, debug and use code blocks from Org, and
only Org. Not be forced to use the shell because some unknown condition make
them return an error.

    Note: Can you explain me what's an error in the case of a sh block?  See
    my previous post(s) of today. Thanks...

Maybe once acceptable solution would be to output everything as it is in the
terminal, mixing both streams.

A possible dream of mine could maybe even become true here: the possibility to
clearly distinguish both streams by coloring them differently in the Org
results: for example, all lines from stdout in black, all lines from stderr in
red...

Though, I understand Eric's fears of problems for chained blocks execution.
But, anyway, what else would you expect than what occurs in a terminal?  Why
should it be different?

Best regards,
  Seb

-- 
Sébastien Vauban


_______________________________________________
Emacs-orgmode mailing list
Please use `Reply All' to send replies to the list.
Emacs-orgmode-mXXj517/zsQ@public.gmane.org
http://lists.gnu.org/mailman/listinfo/emacs-orgmode

Re: [Babel] No output returned if just one command is failing

[Sébastien Vauban]

Hi Eric,

"Eric Schulte" wrote:
> You both make good points in favor of this behavior, and I had no idea
> this would be as easy as Dan's patch below, I was thinking that some
> sort of intermediate results return would be required.
>
> I would be open to either of Dan's suggested changes
> 1. return partial results for ":results output"
> 2. return partial results controlled by some new header argument
>
> I am worried by the prospect of partial results being used by another
> code block in chained code block execution.  Perhaps some special care
> should be taken to ensure that this will not happen, or perhaps this
> should happen only in case two above where the user has explicitly
> specified that partial results are OK.

As I just expressed it, I would vote for "show the results as they are", and
"as they would be in a terminal". No special header needed, in my current view
of the Babel world.

In fact, I find the words "partial results" to be rather negative. They just
are the results a terminal would display: they're *all the available results*
that the block can generate in its execution context.

Best regards,
  Seb

PS- I wanna repeat I may be proven wrong. Maybe I don't think (yet?) at all
    impacts such a decision may have!

-- 
Sébastien Vauban


_______________________________________________
Emacs-orgmode mailing list
Please use `Reply All' to send replies to the list.
Emacs-orgmode-mXXj517/zsQ@public.gmane.org
http://lists.gnu.org/mailman/listinfo/emacs-orgmode