[PATCH] git-svn: write memoized data explicitly to avoid Storable bug

[Sergey Vlasov <vsu@altlinux.ru>]

Apparently using the Storable module during global destruction is
unsafe - there is a bug which can cause segmentation faults:

  http://rt.cpan.org/Public/Bug/Display.html?id=36087
  http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=482355

The persistent memoization support introduced in commit 8bff7c538
relied on global destruction to write cached data, which was leading
to segfaults in some Perl configurations.  Calling Memoize::unmemoize
in the END block forces the cache writeout to be performed earlier,
thus avoiding the bug.

Signed-off-by: Sergey Vlasov <vsu@altlinux.ru>
---
 git-svn.perl |   16 ++++++++++++++++
 1 files changed, 16 insertions(+), 0 deletions(-)

 In my case segfaults happen only when the perl-IO-Compress module is
 installed - apparently the attempt to load Compress::Zlib (now
 provided by IO::Compress) changes something.

 This unaswered report is also suspicious:

   http://thread.gmane.org/gmane.comp.version-control.git/142161

 (when running ./t9151-svn-mergeinfo.sh -v, the segfault should be
 visible: "error: git-svn died of signal 11").

diff --git a/git-svn.perl b/git-svn.perl
index 19d6848..c416358 100755
--- a/git-svn.perl
+++ b/git-svn.perl
@@ -3169,6 +3169,22 @@ sub has_no_changes {
 			LIST_CACHE => 'FAULT',
 		;
 	}
+
+	sub unmemoize_svn_mergeinfo_functions {
+		return if not $memoized;
+		$memoized = 0;
+
+		Memoize::unmemoize 'lookup_svn_merge';
+		Memoize::unmemoize 'check_cherry_pick';
+		Memoize::unmemoize 'has_no_changes';
+	}
+}
+
+END {
+	# Force cache writeout explicitly instead of waiting for
+	# global destruction to avoid segfault in Storable:
+	# http://rt.cpan.org/Public/Bug/Display.html?id=36087
+	unmemoize_svn_mergeinfo_functions();
 }
 
 sub parents_exclude {
-- 
1.6.0.2.321.g8406