[Orgmode] Re: [Babel] Debugging shell code that misbehaves

[Dan Davison]

Sébastien Vauban <address@hidden>
writes:

> Hi Dan,
>
> Dan Davison wrote:
>> Sébastien Vauban writes:
>>
>>> #+TITLE:     Debugging shell code that misbehaves
>>> #+DATE:      2010-10-14
>>> #+LANGUAGE:  en_US
>>>
>>> * Use case
>>>
>>> Let's say that:
>>>
>>> - This file is in =~/Client/Spec=
>>> - I'm working on such a chunk of code:
>>>
>>> #+begin_src sh :results output :exports results
>>> grep indAllocType ../Ontology/champs.csv |\
>>> iconv -f LATIN1 -t UTF8 |\
>>> tr "[]" "|" | cut -d "|" -f 2 | cut -d " " -f 2- |\
>>> sed 's%, %\n%g'
>>> #+end_src
>>>
>>> and that, when executing it, I've got an error with not enough context for 
>>> me
>>> to understand what's the problem is.
>>>
>>> Sneak preview: under Cygwin, the =to= language (of the =iconv= command) must
>>> be =UTF-8= in the above case (with a dash).
>>>
>>> * Debug it
>>>
>>> So, in order to debug, I decide to add a =session= argument:
>>>
>>> #+begin_src sh :results output :exports results :session sva
>>> grep indAllocType ../Ontology/champs.csv |\
>>> iconv -f LATIN1 -t UTF8 |\
>>> tr "[]" "|" | cut -d "|" -f 2 | cut -d " " -f 2- |\
>>> sed 's%, %\n%g'
>>> #+end_src
>>>
>>> and, there, something totally different is occurring: =grep= is not finding
>>> the file anymore.
>>>
>>> Why? Because adding the session argument makes the code executed from my 
>>> *home
>>> directory*, while it was executed from the *document's directory* in the 
>>> first
>>> place.
>>
>> This would be a bug. For me (ubuntu), the session does start up in the
>> document's directory. Perhaps the emacs shell session behaves differently on
>> your operating system (cygwin?)?
>
> How could we proove who's doing that effect?  Do you have an idea for testing
> this, and identifying what portion of code (be it Org, Babel or Cygwin) would
> need to be fixed or customized?

Hi Seb,

What happens if you do:

1. Kill all current *shell* sessions/buffers
2. Use M-x cd to change the emacs default-directory
3. Issue M-x shell
4. Query the current directory of the shell session with pwd

Dan

>
>
>>> So, this is not the right way to debug... as *conditions do change*.
>>
>> but should not
>>
>>>
>>> * Solution?
>>>
>>> What's the right solution for such a case?
>>
>> Fix the bug. But failing that,
>
> Good...
>
>
>>> - Putting a full path to the file =champs.csv= (instead of the relative one)
>>>   is not OK for me, as all of this is under SVN, and I want this to be
>>>   executable on someone's else PC (even if placed somewhere else).
>>>
>>> - Add an explicit =cd= to the right place, before the commands execute. Not
>>>   possible, for the same reason as above.
>>
>> does it help to use a :dir header argument to specify default-directory
>> for the session?
>
> Such a dir spec does not seem to be respected:
>
> #+begin_src sh :results output :exports results :session sva :dir 
> /cygdrive/c/home/sva/Projects/Client
> grep indAllocType ../Ontology/champs.csv |\
> iconv -f LATIN1 -t UTF-8 |\
> tr "[]" "|" | cut -d "|" -f 2 | cut -d " " -f 2- |\
> sed 's%, %\n%g'
> #+end_src
>
> Doing pwd in the session buffer, and I see I'm still in ~ 
> (/cygdrive/c/home/sva).
>
> Moreover, no, that would not, as I cannot predict where the file will be
> located on my colleagues' machines, except if I could add there a relative
> spec such as ":dir ./".
>
> Best regards,
>   Seb