I will take a look a this issue tomorrow.
Thank you for filing it.

Looking...

It works for me locally:

$ gn gen //out/gn-vptr '--args=is_ubsan_vptr=true is_ubsan_no_recover=true is_debug=false is_component_build=false symbol_level=1 dcheck_always_on=true' --check
$ ninja -C out/gn-vptr/ all

For some reason, the bot has different gn flags:

Writing """\
clang_use_chrome_plugins = false
is_clang = true
is_component_build = false
is_debug = false
is_ubsan_no_recover = true
is_ubsan_vptr = true
llvm_force_head_revision = true
sanitizer_coverage_flags = "edge"
""" to /b/build/slave/ClangToTLinuxUBSanVptr/build/src/out/Release/args.gn.

I don't see a reason to have sanitizer_coverage_flags = "edge" on the bot.

The last related change was https://codereview.chromium.org/2000413002 (flipping the bot to GN), and the GYP bot supposedly had this flag set.

It does not fail on GYP, because -fsanitize-coverage= flags are filtered for mesa: https://cs.chromium.org/chromium/src/third_party/mesa/mesa.gyp?dr=C&q=file:mesa+file:gyp+sani&sq=package:chromium&l=85

In GN, it only filters out -fsanitize=vptr, as some mesa code is compiled w/o rtti.

I am trying to understand:

1. If we actually want to do that in GN
2. Why does it actually complain about the unused arg

... and I have lost a connection to my desktop. Great!

This is blocking switching the lkgr waterfall to gn.

The CL has been out for a while: https://codereview.chromium.org/2055323002/

+ClusterFuzz folks

Do you need -fsanitize-coverage=edge on the UBSan bots? If yes, how is it used? I am somewhat confused.

Sanitizer coverage is used on all sanitizer bots (including ubsan ones) to show coverage data in dashboard - https://cluster-fuzz.appspot.com/#coverage

Thank you, Abhishek!

It seems to me that -fsanitize-coverage=edge should not require to specify at least one sanitizer, like it does now.

Sorry, I didn't have time to work on this today, and it seems that I am out for 3 weeks.

There're a few other hanging ends for me, but they are mostly fixes under review, and I hope to push them. On this one, I declare a bankruptcy.

I guess I'll take this as this week's ToT sheriff.

http://reviews.llvm.org/rL272717 should fix this, we'll see the result on the ToT bots soon.

The bot has been green for a while now: https://build.chromium.org/p/chromium.fyi/builders/ClangToTLinuxUBSanVptr/builds/2311

Thank you, Evgenii for fixing it!

Guess we should roll clang soon. Eugenis should we wait for your send() interpose patch, or will that take a while? (Or maybe it already landed and I missed it?)

I have not worked on the send() yet. It should not take long, let my try.

Apparently this'll still break on the chromium.lkgr builders, since we haven't rolled clang yet ( bug 621581 ). Re-opening ...

This is fixed: https://build.chromium.org/p/chromium.fyi/builders/ClangToTLinuxUBSanVptr

Thx, Eugenii!