So, this does not work as it introduces an undefined reference to the operator delete. And after a second thought, it might be that it's the warning that got enabled by default is to blame.

I see the following failure in the bootstrap Clang build:

I am preparing a fix that only makes this typedef under Android.

FYI: this breaks some bots, like http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux/builds/1239

Sorry, I had to revert this revision as asan/msan detected issues with the change:
http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux-fast/builds/3988
http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux-bootstrap/builds/1173

For the record, this makes LLD to crash while linking Chromium with ThinLTO: https://bugs.llvm.org//show_bug.cgi?id=32413

Thank you, Peter!

Generally, looks good, modulo the comment.

I believe this approach is too fragile, in particular because if llvmcache.timestamp is deleted, the cache no longer works, but the directory exists. That creates a broken state and user will have to learn about how the cache works. That breaks the principle of least surprise.

Do we wait for anything or anybody? If not, let's submit the CL. That should fix ThinLTO bot in Chrome.

Thank you, Rui!

Hi Eric,

Hi Eric,

I have manually verified that this change gets rid of an assert during linking Chromium. Can you please also include a test case similar to the one mentioned in https://bugs.chromium.org/p/chromium/issues/detail?id=682773#c14?

I have just confirmed that this CL fixes the issue with Chromium + ThinLTO + debug info described here: https://crbug.com/682773#c9

This change makes Clang hardly incompatible with MSVC++. Consider the following program:

A proper fix had been submitted.

Add a comment to the suppression.

Remove leading space.

For the record, check-lld failures can be observed here:
http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux-fast/builds/2078

LGTM

LGTM

To reproduce the issue on the Chromium side:

NULL -> nullptr; fix the backreferences test cases.

switch -> if

Add backreferences back (except \0).

Fix handling escaped symbols: they are all self-escaped in LLVM.

Format nits.

Shorter in-class initialization (thx Peter!)

Address final nits. Also sync.

Add a test case proposed by Peter.

In D27188#610048, @pcc wrote:

Maybe I'm not making myself clear. This test doesn't pass on my machine.

TEST_F(TrigramIndexTest, foo) {
  std::unique_ptr<TrigramIndex> TI = makeTrigramIndex(
      {"class1.h", "class2.h", "class3.h", "class4.h", "class.h"});
  EXPECT_FALSE(TI->isDefeated());
}

Add one more test case.

Please, take another look.

nits.

Extract TrigramIndex{.h,.cpp,Test.cpp}.

Please, take another look.

Address the comments.

Peter, thank you for your comments. I have added some benchmarks to the description and will address the rest of your points tomorrow.

Speed improvements.

Get rid of build()

Remove debug leftovers.

Simplify code; ensure build() is called.

Add test cases for trigrams into SpecialCaseList unit test.

Oh, I have just learned about the existence of https://reviews.llvm.org/D26880
Probably, my request to review the CL is outdated (I have not yet looked into the other CL to make a real guess)

can you please take a look at this CL, if you have cycles? The CL does not look terribly complicated and it's the only known reason for the Chromium ThinLTO bot being broken:
https://build.chromium.org/p/chromium.fyi/builders/ThinLTO%20Linux%20ToT/builds/551