Re: [Evolution-hackers] EWS MessageInfo leak wtf

[Chenthill Palanisamy <pchenthill gmail com>]

On Sat, Apr 9, 2011 at 4:08 AM, David Woodhouse <dwmw2 infradead org> wrote:
> ==5510== 41 bytes in 1 blocks are possibly lost in loss record 8,473 of 17,762
> ==5510==    at 0x4A05E46: malloc (vg_replace_malloc.c:195)
> ==5510==    by 0x3B6A048B52: g_malloc (gmem.c:164)
> ==5510==    by 0x3B6A06101D: g_strdup (gstrfuncs.c:102)
> ==5510==    by 0x1693D7B1: ews_message_info_clone (camel-ews-summary.c:75)
> ==5510==    by 0x1693E4FD: camel_ews_summary_add_message_info (camel-ews-summary.c:417)
> ==5510==    by 0x16940ADD: camel_ews_utils_sync_created_items (camel-ews-utils.c:973)
> ==5510==    by 0x16939819: sync_created_items (camel-ews-folder.c:660)
> ==5510==    by 0x16939B41: ews_refresh_info_sync (camel-ews-folder.c:750)
> ==5510==    by 0x4C78C99: camel_folder_refresh_info (camel-folder.c:1156)
> ==5510==    by 0x33B6E7F0F7: mail_msg_proxy (mail-mt.c:469)
> ==5510==    by 0x3B6A06BBC3: g_thread_pool_thread_proxy (gthreadpool.c:319)
> ==5510==    by 0x3B6A069445: g_thread_create_proxy (gthread.c:1897)
> ==5510==
>
> Can't repeat it; don't know how it happened... except of course that it
> obviously happened when SyncFolderItems returned some created items.
>
> The frame in ews_message_info_clone() at camel_ews_summary:75 is setting
> mi->change_key on the newly cloned MessageInfo:
> http://git.infradead.org/evolution-ews.git/blob/cd1face:/src/camel/camel-ews-summary.c#l75
>
> (I'm also *sure* it happened after commit 17bd5273 in which I fixed a
> consistent leak of our mi->change_key by adding a message_info_free()
> method to our class. Not only do I remember it that way, but the
> timestamp of my valgrind log file is about four hours later than that
> commit, and the *other* valgrind warnings that the ->change_key leak
> caused are noticeable in their absence.)
>
> I'm confused by the use of camel_message_info_clone() in the first line
> of camel_ews_summary_add_message_info():
> http://git.infradead.org/evolution-ews.git/blob/cd1face:/src/camel/camel-ews-summary.c#l417
> There is *one* caller of this function, and it's the one in the above
> backtrace — camel_ews_utils_sync_create_items():
> http://git.infradead.org/evolution-ews.git/blob/cd1face:/src/camel/camel-ews-utils.c#l973
>
> And that *creates* a new MessageInfo of its own... which AFAICT never
> gets freed, although valgrind doesn't seem to be complaining about that,
> and I don't know why.
>
> Removing the clone in camel_ews_summary_add_message_info() and just
> using '(void *)info' seems to work just as well; I'm not really sure
> what's going on here. And either way, I've never been able to repeat the
> above warning.
>
> Chen, was there a reason for the clone? And why isn't valgrind
> complaining about it? Can anyone see something that would cause the
> above complaint that it *did* make?
The ews_summary_clone seems un-necessary there. I picked up some code from the
imapx provider and certainly did a mistake there. The const identifier
for the CamelMessageInfo in the
signature of the function and message_info_clone can be removed.

- Chenthill.
>
> --
> dwmw2
>
>