Fixes for several memory leaks

[Lars Hjemli]

On Tue, Apr 5, 2011 at 10:51, Lukas Fleischer <cgit at cryptocrack.de> wrote:
> On Sat, Apr 02, 2011 at 10:18:44PM +0200, Lukas Fleischer wrote:
>> I created patches for a whole bunch of memory leaks (probably around 100
>> leaks altogether) which mostly came up due to assigning newly allocated
>> strings (using xstrdupn() in most cases) to configuration strings
>> without free()'ing the previous pointer at all.

Thanks. Even though memleaks have been consciously ignored in cgit
(since it's a cgi app), I appreciate the cleanup which these paches
introduce.

>> Note that this doesn't
>> cover everything by far. There's a lot of more leaks (mainly with repo,
>> ref and commit handling) which I didn't have time to fix yet tho.

Well, it's not certain that we want to fix them...


>> Clone URL of my cgit working tree is [1], web interface of the "wip"
>> branch can be found on [2].
>>
>> [1] git://git.cryptocrack.de/cgit.git
>> [2] http://git.cryptocrack.de/cgit.git/?h=wip

The fix to trim_end() has a couple of alternate incarnations on the
list at the moment, we'll have to see which one ends up as preferred.

>
> Added some null pointer dereference fixes.

The fix to reencode() is probably a good safety measure for the future
(the bug cannot be triggered by current callers), but the right fix to
the issue found by clang-analyzer in cgit_print_diff() probably looks
like this:

+++ b/ui-diff.c
@@ -368,8 +368,10 @@ void cgit_print_diff(const char *new_rev, const char *old_r
                return;
        }
        commit = lookup_commit_reference(new_rev_sha1);
-       if (!commit || parse_commit(commit))
+       if (!commit || parse_commit(commit)) {
                cgit_print_error(fmt("Bad commit: %s", sha1_to_hex(new_rev_sha1)
+               return;
+       }

        if (old_rev)
                get_sha1(old_rev, old_rev_sha1);

--
larsh