subversion-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From rhuij...@apache.org
Subject svn propchange: r1659867 - svn:log
Date Sat, 14 Feb 2015 23:59:04 GMT
Author: rhuijben
Revision: 1659867
Modified property: svn:log

Modified: svn:log at Sat Feb 14 23:59:04 2015
------------------------------------------------------------------------------
--- svn:log (original)
+++ svn:log Sat Feb 14 23:59:04 2015
@@ -3,10 +3,18 @@ handling and abort a commit that just su
 callback failed.
 
 Daniel Shahaf reproduced this edge case by somehow causing a SIGPIPE in
-the commit handler by producing a ^C at the right time. See todays irc
-log for details.
+the commit (info) callback by producing a ^C at the right time. See
+todays irc log for details. 
 
-ra-svn already had a similar safety net here.
+The editor->close_edit() almost succeeded, but right before returning
+SVN_NO_ERROR we call into the callback, and then svn tries to write out
+"Committed revision 64.\n", but fails with SIGPIPE. This makes close_edit()
+return this SIGPIPE and import/commit and probably other callers use this
+to trigger an abort_edit() call.
+
+This scenario is added as testcase to the ra tests.
+
+Note: ra-svn already had a similar safety net here.
 
 * subversion/libsvn_ra_serf/commit.c
   (close_edit): Make sure we don't try to abort a succeeded txn, by


Mime
View raw message