Open CaseyCarter opened 5 months ago
Added a new failure from March 23 to the table. Given that these failures are all x86-specific, I now suspect the root cause is the ASan-x86-specific OPARGPLACE
backend bug VSO-1974914. This issue causes thousands of checked-compiler assertion failures in internal STL ASan-x86 testing that would likely manifest as Silent Bad Codegen with a release compiler.
The fix for VSO-1974914 will likely ship in 17.11.p1. We should close this issue after updating our CI compiler to 17.11p1, and we'll reopen if we continue to see sporadic failures.
I think I'm still seeing this with the 17.11 Preview 1 toolset update.
In STL-ASan-CI runs, seemingly random x86 test cases fail with exit code 3221225477 (0xC0000005) indicating an Access Violation. Let's track sporadic AVs here and see if there's a pattern.
P0896R4_ranges_alg_remove_copy
GH_001123_random_cast_out_of_range
P0768R1_spaceship_operator
VSO_0180466_algorithm_overhauls
tr1/tests/new
P1023R0_constexpr_for_array_comparisons
P0448R4_spanstream
Dev10_860421_deque_push_back_pop_front
range.iota.view/iterator/ctor.default.pass.cpp
P1135R6_atomic_wait_vista
P2374R4_checked_arithmetic_operations
+ 2 moreGH_001010_filesystem_error_encoding
P2278R4_basic_const_iterator
+ 1 moreNote that these invalid accesses are not caught and reported by the ASan runtime, so they are presumably uninstrumented. Together with the fact that this is ASan-only and x86-only, I suspect a race in ASan setup or teardown.