On Wed, Jan 04, 2017 at 08:55:45AM +0000, Vadim Yanitskiy wrote:
I think it's due to --enable-sanitize, because when I am trying to compile libosmocore this way:
./configure --enable-static --enable-sanitize make check
Some other tests fail (current master version, clang-3.6):
Regression tests. 8: gea FAILED (testsuite.at:51) 9: msgfile FAILED (testsuite.at:59) 15: lapd FAILED (testsuite.at:96) 16: gsm0808 FAILED (testsuite.at:102) 17: gsm0408 FAILED (testsuite.at:108) 30: bssgp-fc FAILED (testsuite.at:190)
I pushed some fixes,see https://gerrit.osmocom.org/1530 and https://gerrit.osmocom.org/1531 - the remaining failures are all LeakSanitizer warnings, and I suspect that while we release all memory back to talloc, talloc itself doesn't release it back to glibc malloc/free, and hence LeakSanitizer is kicking in.
A quick online search didn't seem like anyone else has yet come up with a solution for this, so I guess we should turn it off. I remember some discussion on the list about a year ago where the concensus was to disable LeakSanitizer by environment variables on the buildhost anyway?
Hi,
I pushed some fixes,see https://gerrit.osmocom.org/1530 and https://gerrit.osmocom.org/1531 - the remaining failures are all LeakSanitizer warnings, and I suspect that while we release all memory back to talloc, talloc itself doesn't release it back to glibc malloc/free, and hence LeakSanitizer is kicking in.
Just rebased to a new version with both changes applied, and I can confirm, that the 'gea' test is ok now. Other still fail.
Max wrote:
I wonder why this has not been triggering jenkins build failures? Are we using it wrong?
Interesting questions for me too. Which GCC version is used on buildhost? Was one compiled manually?
With best regards, Vadim Yanitskiy.
2017-01-04 17:28 GMT+07:00 Harald Welte laforge@gnumonks.org:
On Wed, Jan 04, 2017 at 08:55:45AM +0000, Vadim Yanitskiy wrote:
I think it's due to --enable-sanitize, because when I am trying to compile libosmocore this way:
./configure --enable-static --enable-sanitize make check
Some other tests fail (current master version, clang-3.6):
Regression tests. 8: gea FAILED (
testsuite.at:51)
9: msgfile FAILED (
testsuite.at:59)
15: lapd FAILED (
testsuite.at:96)
16: gsm0808 FAILED (
testsuite.at:102)
17: gsm0408 FAILED (
testsuite.at:108)
30: bssgp-fc FAILED (
testsuite.at:190)
I pushed some fixes,see https://gerrit.osmocom.org/1530 and https://gerrit.osmocom.org/1531 - the remaining failures are all LeakSanitizer warnings, and I suspect that while we release all memory back to talloc, talloc itself doesn't release it back to glibc malloc/free, and hence LeakSanitizer is kicking in.
A quick online search didn't seem like anyone else has yet come up with a solution for this, so I guess we should turn it off. I remember some discussion on the list about a year ago where the concensus was to disable LeakSanitizer by environment variables on the buildhost anyway? --
- Harald Welte laforge@gnumonks.org
http://laforge.gnumonks.org/
================ "Privacy in residential applications is a desirable marketing option." (ETSI EN 300 175-7 Ch. A6)
On Sat, Jan 07, 2017 at 03:14:15AM +0700, Vadim Yanitskiy wrote:
Which GCC version is used on buildhost?
# gcc --version gcc (Debian 4.9.2-10) 4.9.2
Let me know whether I should do anything there, preferably sticking to Debian packages.
~N
I'd suggest following:
1) disable leak sanitizer via env variable as suggested by Harald (until we find a way to make it compatible with talloc)
2) add latest ubuntu as test builder in addition to debian - it has much newer toolchain and hence can help us catch more bugs
On 07.01.2017 02:12, Neels Hofmeyr wrote:
On Sat, Jan 07, 2017 at 03:14:15AM +0700, Vadim Yanitskiy wrote:
Which GCC version is used on buildhost?
# gcc --version gcc (Debian 4.9.2-10) 4.9.2
Let me know whether I should do anything there, preferably sticking to Debian packages.
~N
-
Harald Welte -
Max -
Neels Hofmeyr -
Vadim Yanitskiy