looks like it hangs right before printing failures

[matthiaskrgr]

I was running ./run-test.sh core --all-targets with a couple of extra flags

diff --git a/run-test.sh b/run-test.sh
index a6a2841..8d4bc6d 100755
--- a/run-test.sh
+++ b/run-test.sh
@@ -32,6 +32,8 @@ export MIRI_LIB_SRC
 rm -f lib$CRATE
 ln -s "$MIRI_LIB_SRC"/$CRATE lib$CRATE

+export MIRIFLAGS="-Zmiri-check-number-validity -Zmiri-strict-provenance  -Zmiri-symbolic-alignment-check -Zmiri-tag-raw-pointers"
+
 # run test
 cd ./${CRATE}_miri_test
 cargo miri setup

and after printing failures: it looked like the script/test execution was stuck.
No further output was printed, I could watch the miri slowly eat more and more memory and I aborted it after reaching ~9-10 gigs.

perf top showed a lot going on miri::stacked_borrows::Stack, wondering if we hit some kind of loop somehow

  47,29%  miri                                 [.] <miri::stacked_borrows::Stack>::access
 38,71%  miri                                 [.] <miri::stacked_borrows::Stack>::grant
  0,59%  [kernel]                             [k] mem_cgroup_iter

miri 0.1.0 (3b8b6aa 2022-05-06)

[RalfJung]

This looks like the regular Stacked Borrows super-linear complexity to me. I just wonder what it is computing before printing the failures...

... why are there any failures anyway? Maybe because the symbolic alignment check is not compatible with all of the core tests? That's why we do --skip align on CI for that. Maybe try passing align to only run the failing lists, see if that makes a difference?

[RalfJung]

Closing in favor of rust-lang/miri#2273.