ASan ODR failure in ppapi::proxy::TCPSocketResourceConstants::kMaxReadSize

[Chris Palmer]

I keep getting this ASan ODR failure in ToT (for several days now). Anyone else, or is it just me?

$ ./out/Default/blink_heap_unittests 

=================================================================

==10984==ERROR: AddressSanitizer: odr-violation (0x7f63b125b5c0):

  [1] size=4 'ppapi::proxy::TCPSocketResourceConstants::kMaxReadSize' ../../ppapi/proxy/tcp_socket_resource_constants.cc:10:43

  [2] size=4 'ppapi::proxy::TCPSocketResourceConstants::kMaxReadSize' ../../ppapi/proxy/tcp_socket_resource_constants.cc:10:43

These globals were registered at these points:

  [1]:

    #0 0x49d0d7  (/ssd/chromium/src/out/Default/blink_heap_unittests+0x49d0d7)

    #1 0x7f63af15b2db  (/ssd/chromium/src/out/Default/./libcontent.so+0x2a9d2db)

  [2]:

    #0 0x49d0d7  (/ssd/chromium/src/out/Default/blink_heap_unittests+0x49d0d7)

    #1 0x7f639b34a0bb  (/ssd/chromium/src/out/Default/./libppapi_proxy.so+0x49c0bb)

==10984==HINT: if you don't care about these errors you may set ASAN_OPTIONS=detect_odr_violation=0

SUMMARY: AddressSanitizer: odr-violation: global 'ppapi::proxy::TCPSocketResourceConstants::kMaxReadSize' at ../../ppapi/proxy/tcp_socket_resource_constants.cc:10:43

==10984==ABORTING

[Lucas Gadani]

Not very helpful, but since no one else replied, I've been getting the same error for a while, and I've just been disabling odr violation detection on asan runs.

--
--
Chromium Developers mailing list: chromi...@chromium.org
View archives, change email options, or unsubscribe:
http://groups.google.com/a/chromium.org/group/chromium-dev

[Ken Rockot]

I would guess this is caused by a recent move[1] of these constant definitions to a separate source_set target which then gets separately linked into multiple link-time dependencies (e.g. component targets) of the test binary, hence multiple definitions. Probably the simplest thing to do here is change that "common" target[2] to a component.

[1] https://chromium.googlesource.com/chromium/src/+/3cab0ba1179fcb085af9efb0a2734c51b8d79be3

[2] https://cs.chromium.org/chromium/src/ppapi/proxy/BUILD.gn?rcl=df2469d2aacf348710be6f823bb4228696edef38&l=309

[Scott Graham]

Oh! I'll take a look.

(Was there a bot failing somewhere that didn't yell at me? 'cuz as always, 🎶 if you liked it you shoulda put a bot on it.)

[Ken Rockot]

I dunno if we enable ODR violation checks on ASAN bots.

In this case nothing is really broken given the usage of these symbols, and this only affects component builds, but I still think it's worth fixing ODR violations when we see them.

[Chris Palmer]

I don't know if it's on a bot, but the build that fails for me is also the only bot that is blocking a CL of mine from landing. But when I click to read its logs, I get a blank screen (...?), so I broke down and did a local build... and found this. :)

[Ken Rockot]

It seems pretty unlikely to me that this is the source of your breakage, so you'll probably want to look elsewhere in the meantime.

[Scott Graham]

Sorry, fix is here, but not landed yet https://codereview.chromium.org/2727093003/.

Agreed that the current code doesn't seem likely to be causing that sort of problem as it's just some integral constants being defined twice.