summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
-rw-r--r--doc/bugs/locking_fun.mdwn5
1 files changed, 4 insertions, 1 deletions
diff --git a/doc/bugs/locking_fun.mdwn b/doc/bugs/locking_fun.mdwn
index 6d5f79ce5..64d06917f 100644
--- a/doc/bugs/locking_fun.mdwn
+++ b/doc/bugs/locking_fun.mdwn
@@ -146,7 +146,9 @@ makes the commit hook a NOOP.
(it may still need to send commit mails)
* CGI removes wclock, thus re-enabling the commit hook
* CGI updates the WC (since the commit hook didn't)
-* CGI renders the wiki
+* CGI renders the wiki (always. commits may have came in and not been
+ rendered)
+* CGI checks for conflicts, and if any are found does its normal dance
> It seems like there are two things to be concerned with: RCS commit between
> disable of hook and CGI commit, or RCS commit between CGI commit and re-enable
@@ -180,6 +182,7 @@ NOOP:
* svn commit -m "conflict" (this makes a change to repo immediately, then
runs the post-commit hook, which becomes a NOOP)
* cgi calls rcs_commit, which fails due to the conflict just introduced
+* cgi renders the wiki
Actually, the only thing that scares me about this apprach a little is that
we have two locks. The CGI takes them in the order (wikilock, wclock).