Re: [PATCH] Corrects an Abort Trap: 6 error with completions.

From: Steven Klass <sklass@7stalks.com>
Date: Jul 8, 2013, 17:45 UTC
Message-ID: <9699781D-9A21-4C65-86E7-50156F934C08@7stalks.com>
In-Reply-To: <7vwqp1yn1w.fsf@alter.siamese.dyndns.org>

Hi Junio,

First - Thanks so much for your reply!

	The original cause was simply running a non-related command.  Specifically I was trying to build some internal software.  Our internal software uses a build build tool from perforce called jam (http://www.perforce.com/resources/documentation/jam).  Similar to the make build tool this looks up a Jamfile (ergo Makefile) and does the recipe.  The issue is that running `jam` fatally aborts right from the start with an unintuitive `abort trap: 6` error.

	As you can well imagine figuring out what was causing this was pretty difficult - because both the error was cryptic and so unrelated.  I started backing out my environment until I was able to narrow this down to the bash_completions file.  Further investigation finally nailed it on the concatenation of the  `branch. remote.` line 2165.    In summary when the lines were joined it showed the error, when they were on separate lines everything processed as expected.

	Again - my only change was to shift them (  branch. remote. ) to two separate lines.

HTH

---

Steven Klass

(480) 225-1112 sklass@7stalks.com http://www.7stalks.com

On Jul 8, 2013, at 10:32 AM, Junio C Hamano <gitster@pobox.com> wrote:

Show 72 quoted lines

> Steven Klass <sklass@7stalks.com> writes:
> 
>> Hi Folks,
>> 
>> 	Corrects an Abort Trap: 6 error with completions.
>> 	Fixed an issue where the two commands on a single line  would cause a strange unrelated 'Abort trap: 6' error on  non-git commands on Mac OSX 10.8.
>> 
>> Signed-off-by: Steven Klass <sklass@7stalks.com>
> 
> Can you explain how/why the original causes "abort trap: 6"
> (whatever it is) and how/why the updated one avoids it in the log
> message?
> 
> It also is not quite clear when the error happens.  Do you mean, by
> "non-git commands", something like:
> 
>    $ ca<TAB>
> 
> does not complete to "cal", "case", "cat", etc. and instead breaks
> the shell?
> 
> I am confused.  The only change I can see in the patch is that it
> makes the argument to this call to the __gitcomp shell function be a
> string with tokens separated by LF and HT and no SP (the original
> assumes that the tokens will be split by LF, HT or SP, and the shell
> function locally sets $IFS to make sure that the change in this
> patch does not make any difference).  And in many other places in
> the same script, the __gitcomp shell function is called with an
> argument with LF, HT or SP spearated tokens, e.g.
> 
>    _git_add ()
>    {
>            case "$cur" in
>            --*)
>                    __gitcomp "
>                            --interactive --refresh --patch --update --dry-run
>                            --ignore-errors --intent-to-add
>                            "
>                    return
>            esac
> 
> 
> 
>> 
>> ---
>> contrib/completion/git-completion.bash | 3 ++-
>> 1 file changed, 2 insertions(+), 1 deletion(-)
>> mode change 100644 => 100755 contrib/completion/git-completion.bash
>> 
>> diff --git a/contrib/completion/git-completion.bash b/contrib/completion/git-completion.bash
>> old mode 100644
>> new mode 100755
>> index 6c3bafe..d63b1ba
>> --- a/contrib/completion/git-completion.bash
>> +++ b/contrib/completion/git-completion.bash
>> @@ -2165,7 +2165,8 @@ _git_config ()
>>           user.name
>>           user.signingkey
>>           web.browser
>> -          branch. remote.
>> +          branch.
>> +          remote.
>>      "
>> }
>> 
>> ---
>> 
>> Steven Klass
>> 
>> (480) 225-1112
>> sklass@7stalks.com
>> http://www.7stalks.com