Re: [PATCH 0/8] git-prompt: support more shells

From: avih <avihpit@yahoo.com>
Date: Jul 24, 2024, 17:08 UTC
Message-ID: <2104161937.2499137.1721840883913@mail.yahoo.com>
In-Reply-To: <xmqq7cdazu4a.fsf@gitster.g>

 On Wednesday, July 24, 2024 at 06:29:12 PM GMT+3, Junio C Hamano <gitster@pobox.com> wrote:

Show 12 quoted lines

>
> It is a bit more nuanced than that, though.  Here is what we say:
>
> - Even though "local" is not part of POSIX, we make heavy use of it
>   in our test suite.  We do not use it in scripted Porcelains, and
>   hopefully nobody starts using "local" before all shells that matter
>   support it (notably, ksh from AT&T Research does not support it yet).
>
> For the purpose of git-prompt, I think it should be OK ...
> to declare that we now support shells
> other than bash and zsh as long as they are reasonably POSIX and
> support "local" that is dynamic.

I have to admit I missed "in our test suite". Right, so no "local" in Porcelains, but yes in the test suite.

But yes, agreed, because it supports so many more shells. The commit "git-prompt: ta-da! document.." does document it.

> (without
> "local", it is harder, if not impossible, to clobber end-user's
> shell variable namespace with various temporaries we need to use
> during prompt computation)

It actually is technically possible with git-prompt.sh, with 1 LOC. See the "anecdote" at end of the same "ta-da!" commit message which does exactly that. Though for obvious reason we can't really do that.

> Do we know what kind of "local" is ksh93 adding?  The same as their
> "typeset" that is not dynamic?  That is so different from what others
> do and scripts expect to be all that useful, I am afraid.

I would think it has to be similar enough to other shells, or else it creates a compatibility nightmare IMO. But that's a guess.

Somewhat off topic, so apologies if this shouldn't be here:

As for the Porcelains, I have to assume that it can be unpleasant to maintain big scripts without "local". Would there be an interest in adding a facility with the same semantics as "local", but posix compliant (and also posix-ish shells), for use in Porcelains?

It's not a drop-in replacement, but the syntax is reasonable IMO:

    locals myfunc x y
    _myfunc () {
        echo "$? $1 $2"
        x=1 y=2
        return 33
    }

    x=x; unset y
    (exit 42) || myfunc foo bar
    echo "$? $x ${y-unset}"

Prints:
    42 foo bar
    33 x unset

"locals myfunc x y" creates a wrapper function "myfunc" which saves the state of $x and $y, calls _myfunc "$@", then restores the state (and propagates the initial and final $? appropriately). Recursion is supported, the wrapper doesn't create additional variables, and no subshells are used at the wrapper (also not at "locals").

The implementation of "locals" is small (10-20 LOC), but we can't expect scripts to embed it, so it would need to be sourced (dot).

If there is interest in such thing, let me know, and I can submit a patch (independent of this patchset) to adds such file which can then be sourced by other scripts in order to use "locals".

Unrelated, and it might not mean much, but I did want to thank you for maintaining git all those years.