diff options
| author | joey <joey@0fa5a96a-9a0e-0410-b3b2-a0fd24251071> | 2007-10-10 18:40:54 +0000 |
|---|---|---|
| committer | joey <joey@0fa5a96a-9a0e-0410-b3b2-a0fd24251071> | 2007-10-10 18:40:54 +0000 |
| commit | 278b16c79ad98514e19dd3301e771516499f7a24 (patch) | |
| tree | 96cc329eef2585f5a5feaff6e246f35b10f965dc | |
| parent | 66a137848f0310b8721fbd276581890e26ecfbb4 (diff) | |
| download | ikiwiki-278b16c79ad98514e19dd3301e771516499f7a24.tar ikiwiki-278b16c79ad98514e19dd3301e771516499f7a24.tar.gz | |
* In the cgi edit path, reload the index file before rendering. A bug
showed up where a web edit that added a page caused a near-concurrent
web edit to fail in will_render. While it would be hard to reproduce this,
my analysis is that the failing cgi started first, loaded the index file
(prior to locking) then the other cgi created the new page and rendered
it, and then the failing cgi choked on the new file when _it_ tried to
render it. Ensuring that the index file is loaded after taking the lock
will avoid this bug.
| -rw-r--r-- | IkiWiki/CGI.pm | 4 | ||||
| -rw-r--r-- | debian/changelog | 12 |
2 files changed, 14 insertions, 2 deletions
diff --git a/IkiWiki/CGI.pm b/IkiWiki/CGI.pm index 155010a97..788d0487e 100644 --- a/IkiWiki/CGI.pm +++ b/IkiWiki/CGI.pm @@ -596,6 +596,10 @@ sub cgi_editpage ($$) { #{{{ # may have been committed while the post-commit hook was # disabled. require IkiWiki::Render; + # Reload index, since the first time it's loaded is before + # the wiki is locked, and things may have changed in the + # meantime. + loadindex(); refresh(); saveindex(); diff --git a/debian/changelog b/debian/changelog index ba24beee9..138412216 100644 --- a/debian/changelog +++ b/debian/changelog @@ -3,8 +3,16 @@ ikiwiki (2.10) UNRELEASED; urgency=low * Tidy ctime debug output for git. * French translation update. Closes: #445923 * Fix --get-ctime with git, needed to remove srcdir from filename. - - -- Joey Hess <joeyh@debian.org> Wed, 10 Oct 2007 14:14:18 -0400 + * In the cgi edit path, reload the index file before rendering. A bug + showed up where a web edit that added a page caused a near-concurrent + web edit to fail in will_render. While it would be hard to reproduce this, + my analysis is that the failing cgi started first, loaded the index file + (prior to locking) then the other cgi created the new page and rendered + it, and then the failing cgi choked on the new file when _it_ tried to + render it. Ensuring that the index file is loaded after taking the lock + will avoid this bug. + + -- Joey Hess <joeyh@debian.org> Wed, 10 Oct 2007 14:36:38 -0400 ikiwiki (2.9) unstable; urgency=low |