[compiler-rt] r230019 - [ASan] Make the argument of '__sanitizer_annotate_contiguous_container' is not aligned error message easier to understand

Renato Golin renato.golin at linaro.org
Mon Feb 23 12:27:26 PST 2015


On 23 February 2015 at 19:56, Kostya Serebryany <kcc at google.com> wrote:
> Do I understand correctly, that the test broke on ARM because "long t[100]"
> is not necessary aligned by 8 on ARM?

long on ARM is 32-bits and word aligned. But I agree with Timur, the
print seems completely off.


> I agree that having bots red is important, but in this case it would have
> been better to actually fix the test than to roll-back the patch  (which is
> correct, but just uncovers the incompatibility of the test with ARM).

I also agree, but this case is special. I sent an email to Timur over
the weekend and haven't thought much about it (since it was weekend),
but on Sunday I saw that another regression had crept up. I was more
uncertain as to what the other regression was from (I guessed the
Clang change, but...), so I decided to revert this change so I could
fix the second bug and get it green again.

I'm very lenient with regards to red bots, but I draw the line when
new regressions start creeping in because the bot is red for a long
time.


> Renato, may I ask you to fix the test for ARM and apply Timur's patch back?
> Not having access to ARM machine makes it harder for us to do.

Absolutely! I'll be working on this as of now. I'll probably need your
help, and if you can figure out what's wrong with the print function,
I could easily test it here before we commit again.

cheers,
--renato

PS: On a related note, I've set up a self-hosting Thumb2 compiler-rt
buildbot, so that we monitor it as well.
http://lab.llvm.org:8011/builders/clang-cmake-thumbv7-a15-full-sh



More information about the llvm-commits mailing list