Re: [PATCH] Invoke git-repo-config directly.

[Junio C Hamano <junkio@cox.net>]

Qingning Huo <qhuo@mayhq.org> writes:

> So we recognize the full path approach is desired,

I said "if you can work problems your environment has _without_
doing the full path thing, then it would be ideal".

I never said full path is desired -- I despise full path, in
fact.  It makes certain things very inconvenient.

> ...but because of
> technique reasons (building and testing), it is not applied.

There were not even a patch.  I suspect people involved in the
discussion realized that approach was unworkably cumbersome.

In our Makefile, we have sed script mechanism to replace tokens
so we _could_ change our sources to do something like this:

        diff --git a/git-commit.sh b/git-commit.sh
        index 330a434..10835c6 100755
        --- a/git-commit.sh
        +++ b/git-commit.sh
	...
        @@ -115,7 +115,7 @@ run_status () {
                    echo '#
         # Initial commit
         #'
        -	    git-ls-files |
        +	    @@GIT_PATH@@git-ls-files |
                    sed -e '
                            s/\\/\\\\/g
                            s/ /\\ /g
        @@ -126,7 +126,7 @@ run_status () {
                    committable="$?"
                fi

        -	git-diff-files --name-status |
        +	@@GIT_PATH@@git-diff-files --name-status |
                sed -e '
                        s/\\/\\\\/g
                        s/ /\\ /g
	...

and sed it out with 's/@@GIT_PATH@@/$(gitexecdir_SQ)/g'.

However, you have to realize that I often want to try things out
*without* running "make", let alone installing.  The current way
things are set up lets me say:

	$ GIT_EXEC_PATH=$my_git_source \
          sh -x $my_git_source/git-commit.sh

to see where things break.  Changing things the way I quoted
above would make things _extremely_ inconvenient for me.

And it is ugly.  "Making things ugly and inconvenient for what
purpose?" is the question I have to ask myself at this point.

And if the answer is "to support unusual configuration which,
quite frankly, I think is broken", then...

We could probably define a shell function that looks like:

        git_exec () {
                cmd="$1"
                shift
                case "${GIT_EXEC_PATH+set}" in
                set) ;;
                *) GIT_EXEC_PATH='@@GIT_EXEC_PATH@@' ;;
                esac
                "$GIT_EXEC_PATH/git-$cmd" "$@"
        }

in git-sh-setup [*1*], and then rewrite the above to something
like this instead:

        diff --git a/git-commit.sh b/git-commit.sh
        index 330a434..8a73420 100755
        --- a/git-commit.sh
        +++ b/git-commit.sh
        ...
        @@ -115,7 +115,7 @@ run_status () {
                    echo '#
         # Initial commit
         #'
        -	    git-ls-files |
        +	    git_exec ls-files |
                    sed -e '
                            s/\\/\\\\/g
                            s/ /\\ /g
        @@ -126,7 +126,7 @@ run_status () {
                    committable="$?"
                fi

        -	git-diff-files  --name-status |
        +	git_exec diff-files  --name-status |
                sed -e '
                        s/\\/\\\\/g
                        s/ /\\ /g
        ...

But that does not cover Perl nor Python scripts, and does not
address the ugliness either.


[Footnote]

*1* BTW, I just noticed that git-sh-setup needs to be on user's
PATH, so we probably have to inline and duplicate the git_exec()
shell function definition at the beginning of each script after
all, when we make the initial ". git-sh-setup" inclusion to
honor GIT_EXEC_PATH without munging the user's PATH.

Which is not a big deal by itself, since we preprocess
*.{sh,perl,py} files anyway, but still it leaves a _big_
ugliness factor.