Junio C Hamano <gits...@pobox.com> writes:

> Adam Tkac <at...@redhat.com> writes:
>> Subject: Re: [PATCH v2] If `egrep` is aliased, temporary disable it in 
>> bash.completion
> The code does not seem to do anything special if it is not aliased,
> though, so "If ..." part does not sound correct; perhaps you meant
> "just in case egrep is aliased to something totally wacky" or
> something?
> The script seems to use commands other than 'egrep' that too can be
> aliased to do whatever unexpected things.  How does this patch get
> away without backslashing them all, like
>       \echo ...
>         \sed ...
>         \test ...
>         \: comment ...
>       \git args ...
> and still fix problems for users?  Can't the same solution you would
> give to users who alias one of the above to do something undesirable
> be applied to those who alias egrep?
> Puzzled...

Sorry for having been more snarky than necessary (blame it to lack
of caffeine).  What I was trying to get at were:

 * I have this suspicion that this patch exists only because you saw
   somebody who aliases egrep to something unexpected by the use of
   it in this script, and egrep *happened* to be the only such
   "unreasonable" alias.  The reporter may not have aliased echo or
   sed away, or the aliases to these command *happened* to produce
   "acceptable" output (even though it might have been slightly
   different from unaliased one, the difference *happened* not to
   matter for the purpose of this script).

 * To the person who observes the same aliasing breakage due to his
   aliasing sed to something else, you would solve his problem by
   telling him "don't do that, then".  If that is the solution, why
   wouldn't it work for egrep?

 * The next person who aliased other commands this script uses in
   such a way that the behaviour of the alias differs sufficiently
   from the unaliased version, you will have to patch the file
   again, with the same backslashing.  This patch is not a solution,
   but a band-aid that only works for a particular case you
   *happened* to have seen.

 * A complete solution that follows the direction this patch
   suggests would involve backslashing *all* commands that can
   potentially aliased away.  Is that really the direction we would
   want to go in (answer: I doubt it)?  Is that the only approach to
   solve this aliasing issue (answer: I don't know, but we should
   try to pursue it before applying a band-aid that is not a

Is there a way to tell bash "do not alias-expand from here up to
there"?  Perhaps "shopt -u expand_aliases" upon entry and restore
its original value when we exit, or something?

IOW, something along this line?

 contrib/completion/git-completion.bash | 13 +++++++++++++
 1 file changed, 13 insertions(+)

diff --git i/contrib/completion/git-completion.bash 
index 0b77eb1..193f53c 100644
--- i/contrib/completion/git-completion.bash
+++ w/contrib/completion/git-completion.bash
@@ -23,6 +23,14 @@
 #    3) Consider changing your PS1 to also show the current branch,
 #       see git-prompt.sh for details.
+if shopt -q expand_aliases
+       _git__aliases_were_enabled=yes
+       _git__aliases_were_enabled=
+shopt -u expand_aliases
 *:*) : great ;;
@@ -2504,3 +2512,8 @@ __git_complete gitk __gitk_main
 if [ Cygwin = "$(uname -o 2>/dev/null)" ]; then
 __git_complete git.exe __git_main
+if test -n "$_git__aliases_were_enabled"
+       shopt -s expand_aliases

To unsubscribe from this list: send the line "unsubscribe git" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Reply via email to