From: Jeff King <peff@peff.net>
Date: Wed, 03 Aug 2011 09:32:52 GMT
Subject: Re: [PATCH 1bis/2] Diff patterns for POSIX shells
Message-ID: <20110803093252.GA16351@sigill.intra.peff.net>
In-Reply-To: <1312349176-20984-1-git-send-email-giuseppe.bilotta@gmail.com>

On Wed, Aug 03, 2011 at 07:26:16AM +0200, Giuseppe Bilotta wrote:

> All diffs following a function definition will have that function name
> as chunck header, but this is the best we can do with the current
> userdiff capabilities.

Curious as to how this would look in git.git, I tried "git log -p"
before and after your patches, and diffed the result. I noticed two
things:

  1. Given a block of shell code like this:

        foo() {
          ... do something ...
        }

        test_expect_success 'test foo' '
          ... the actual test ...
        '

     if we add new code after the test, the old regex would print:

        @@ -1,2 +3,4 @@ test_expect_success 'test foo' '

     and now we say:

        @@ -1,2 +3,4 @@ foo

     which seems more misleading. I know the function-matching code has
     no way to say "look for ^}, which signals end of function", so we
     can't be entirely accurate. But I wonder if the new heuristic
     (which seems to look for a name followed by parentheses) is
     actually any better than the old.

  2. What would have printed before:

       @@ -1,2 +3,4 @@ foo() {

     now prints

       @@ -1,2 +3,4 @@ foo

     without the parentheses or brace. It looks like the similar C one
     keeps the parentheses, at least. I find that a bit more readable,
     as it is more clear that the line indicates a function, and not
     simply some top-level command.

-Peff