Re: commit-msg hook

[Paul Eggert <eggert@cs.ucla.edu>]

[-- Attachment #1: Type: text/plain, Size: 1602 bytes --]

Eli Zaretskii wrote:
> the problem with hanging is not due
> to Gawk, but instead due to 'printf': it doesn't produce the UTF-8
> encoding of the percent sign

OK, so we're dealing with a broken MS-Windows shell rather than a broken 
MS-Windows gawk.  That 'printf' command is there only so that the script can be 
pure ASCII, and it's simpler to avoid that nicety.  I installed the attached 
patch to try to work around the shell bug.

As regards the idea of using Emacs rather than the shell, I'm with Dmitry: let's 
not assume Emacs is already working when bootstrapping a new Emacs.  Instead, 
let's merely assume that a new developer has a working shell, since that's 
needed anyway for other reasons.

> Is it possible to install the hooks in GNUMakefile, so that running
> autogen.sh manually wouldn't be needed?  I normally almost never run
> autogen.sh, I always just run "make" from the top-level directory.

I'm leery of that.  GNUMakefile is run by everybody, not just by developers. 
And there are good reasons that the Git hooks are not themselves in the Git 
repository (if they were in the repository, and they were sufficiently messed 
up, then one couldn't 'git commit' fixes).  If we routinely updated the Git 
hooks automatically, we'd make it more likely that Git non-experts would put 
their painfully constructed repositories into a "broken" state, and we already 
have too much panic in that neighborhood.

That being said, it might make sense to separate the Git hook (re)initialization 
into an autogen.sh option, since it's cheap whereas autogen.sh by default is 
expensive.

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: 0001-Port-commit-msg-to-broken-MS-Windows-shell.patch --]
[-- Type: text/x-patch; name="0001-Port-commit-msg-to-broken-MS-Windows-shell.patch", Size: 1570 bytes --]

From 2cc9a32d361492f28faeced22e3800eb1a59c252 Mon Sep 17 00:00:00 2001
From: Paul Eggert <eggert@cs.ucla.edu>
Date: Sat, 11 Apr 2015 08:19:13 -0700
Subject: [PATCH] Port commit-msg to broken MS-Windows shell

* build-aux/git-hooks/commit-msg (cent_sign):
Just use UTF-8 here rather than ASCII + printf, as the latter fails
on a broken MS-Windows shell.  Reported by Eli Zaretskii in:
http://lists.gnu.org/archive/html/emacs-devel/2015-04/msg00592.html
---
 build-aux/git-hooks/commit-msg | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/build-aux/git-hooks/commit-msg b/build-aux/git-hooks/commit-msg
index ea8d909..3fc6e19 100755
--- a/build-aux/git-hooks/commit-msg
+++ b/build-aux/git-hooks/commit-msg
@@ -29,8 +29,7 @@ fi
 
 # Use a UTF-8 locale if available, so that the UTF-8 check works.
 # Use U+00A2 CENT SIGN to test whether the locale works.
-cent_sign_utf8_format='\302\242\n'
-cent_sign=`printf "$cent_sign_utf8_format"`
+cent_sign='¢'
 print_at_sign='{print substr("'$cent_sign'@", 2)}'
 at_sign=`$awk "$print_at_sign" 2>/dev/null`
 if test "$at_sign" != @; then
@@ -45,7 +44,7 @@ exec $awk -v at_sign="$at_sign" -v cent_sign="$cent_sign" '
   BEGIN {
     # These regular expressions assume traditional Unix unibyte behavior.
     # They are needed for old or broken versions of awk, e.g.,
-    # mawk 1.3.3 (1996), Gawk 3.0.4 (1999).
+    # mawk 1.3.3 (1996), or gawk on MSYS (2015).
     space = "[ \f\n\r\t\v]"
     non_space = "[^ \f\n\r\t\v]"
     non_print = "[\1-\37\177]"
-- 
2.1.0