Comment by ~shakna on ~shakna/jstring
This has to do with AVX instruction handling, and is a false positive.
REPORTED RESOLVED NOT_OUR_BUG
Comment by ~shakna on ~shakna/jstring
It has to do with the way AVX instruction are handled.
This has an easy reproducer on any x86_64 with avx2 setup. Any program using wcsncmp will produce the issue.
Comment by ~shakna on ~shakna/jstring
This a false positive from valgrind.
REPORTED RESOLVED NOT_OUR_BUG
Ticket created by ~shakna on ~shakna/jstring
==28430== Conditional jump or move depends on uninitialised value(s) ==28430== at 0x49D9CC9: __wcsncmp_avx2 (in /usr/lib/libc-2.29.so) ==28430== by 0x1094D9: jstring_cmp (jstring.h:229) ==28430== by 0x109C2F: test_substr_valid (test.c:95) ==28430== by 0x10915D: main (test.c:123)Valgrind picks this up when run as:
valgrind --leak-check=full --track-origins=yes ./test
Ticket created by ~shakna on ~shakna/jstring
==28430== Invalid read of size 32 ==28430== at 0x49D9CA5: __wcsncmp_avx2 (in /usr/lib/libc-2.29.so) ==28430== by 0x1094D9: jstring_cmp (jstring.h:229) ==28430== by 0x1099C1: test_cmp_same_different_object (test.c:60) ==28430== by 0x109148: main (test.c:120) ==28430== Address 0x4a3d330 is 0 bytes inside a block of size 20 alloc'd ==28430== at 0x483877F: malloc (vg_replace_malloc.c:299) ==28430== by 0x109389: jstring_new (jstring.h:189) ==28430== by 0x109991: test_cmp_same_different_object (test.c:56) ==28430== by 0x109148: main (test.c:120)Valgrind picks up a possible bug when run as:
valgrind --leak-check=full --track-origins=yes ./test
Comment by ~shakna on ~shakna/jstring
v3 changes things a bit under the covers.
REPORTED RESOLVED WONT_FIX
Comment by ~shakna on ~shakna/jstring
Not really necessary anymore.
Using an alternative malloc like jemalloc is easy with LD_PRELOAD.
REPORTED RESOLVED WONT_FIX
Comment by ~shakna on ~shakna/evjl
Solved in v.0.2.0
REPORTED RESOLVED FIXED
Comment by ~shakna on ~shakna/evjl
This will be considered solved when the user can arbitrary link, and set library and include directories.