suppressed leaks show up in leak count for -no_check_leaks about drmemory HOT 8 OPEN

From [email protected] on July 23, 2010 07:26:48

Issue 16 has been merged into this issue.

Cc: [email protected]

from drmemory.

From [email protected] on August 29, 2010 13:58:38

fixed in r41 : turn "-check_leaks -possible_leaks" on by default

Status: Verified

from drmemory.

From [email protected] on August 30, 2010 04:38:04

Looks like only the default value of flags have been changed.

I think that's not enough - the leak checking is still performed when run with -no_check_leaks, they don't match suppressions and the number of leaks is still printed. We should at least skip printing of leak numbers in the error summary.

What do you think?

from drmemory.

From [email protected] on August 30, 2010 07:25:27

yes the issue is still there if "-no_check_leaks" is turned on.
I have another option "-no_count_leaks" which really turns off leak scanning.
is there a scenario where a user wants a count of leaks but does not want a list of the leaks with callstacks? if not, then perhaps "-no_check_leaks" should not be supported, and users would then either get the full leak count + list of leaks w/ callstacks + suppression power, or have no leak scan at all, and have nothing in between.

from drmemory.

From [email protected] on August 30, 2010 07:27:50

I can't imagine a use case when leak count is needed without the reports

from drmemory.

From [email protected] on August 30, 2010 07:34:56

isn't that what Memcheck does by default? gives a count of leaks by default, but only if you run with full leak info do you get info on individual leaks?

from drmemory.

From [email protected] on August 30, 2010 07:40:54

Oops yes, you're right.
I was confused because Chromium wrapper scripts implicitly set --leak-check=full.
Maybe there is such a use case...

Is it hard to account for suppressions when printing leak summary?
Should we open a new bug for that or reopen this one?

For now, I'll use -no_count_leaks instead of -no_check_leaks

Labels: Component-LeakCheck

from drmemory.

From [email protected] on August 30, 2010 08:24:20

re-opening this issue since changing the defaults is only a partial fix.

my -no_check_leaks (confusingly named I know) doesn't keep the callstacks around (it assumes you want to save memory if you don't want to see the callstacks) so it can't do the suppression. one option is to have a runtime option that does have all the info, including the callstacks, but just doesn't print them to the error list, only the summary, whether that's changing what -no_check_leaks does or adding a new (better-named) option and not supporting -no_check_leaks externally.

anyway I agree that the use case doesn't seem that compelling, so lowering priority. if no other change is made, the current docs should be updated to say that -no_check_leaks results in a count that ignore suppression.

Status: Accepted
Labels: -Priority-Medium Priority-Low

from drmemory.