Committed as LLDB r314225

Updates and small refactors requested by Bruno.

Applied Greg's and Zachary's suggestions:

At @bruno 's request, abandoning this revision in favor of the updated https://reviews.llvm.org/D38208.
That revision has all the changes requested.

Thank you. I am testing a new revision with your suggestions.

Updated to reflect Bruno's suggestions.

Added DEBUG_PRINTF code to enable introspection into what's happening.

Removed unnecessary code and improved code coverage. Also made a few operational changes:

Added more tests, and added an option to clang-import-test to actually use the origin-forwarding code. Also fixed some errors in CompleteType that were caught by these new tests.

Updated the patch to make it possible for external clients to record the origins of types manually, a requirement for LLDB.

Updated the patch to eliminate the need to pass in an OriginMap for the target. The OriginMap can be extracted afterward, and merged by other algorithms if necessary.

Eliminate sensitive dependence on sizeof(int). bool should still be rendered as i1 though.

Added a passing test for a global struct, so we have something that'll fail if the IR dumping breaks.

Upon reflection, it's not worth coming up with a new pattern if

Eliminated a useless Expected.

Responded to Lang's comments inline.

Looks good to me. Upon reflection, I think it's more important to adhere to the standard of the surrounding code in the ASTDumper, otherwise you're hurting reusability.

Thanks for working on this! I have a few requests (some may arise from ignorance of capture semantics).

Hmm, the transforming in place of SelLocs reads a little weirdly to me, but other than that the code seems fine.
Is your concern that you don't know how to write an Objective-C test that would cover this? It looks to me like an Objective-C interface with a method:

@interface MyClass { }
-(int)addInt:(int)a toInt:(int)b moduloInt:(int)c;
@end

might be enough for an ASTMerge test. If you want to make clang-import-test use this code, then we might need to add one or two things into that tester to handle Objective-C method lookup.

Thanks for the test. This looks fine.

Sorry for the late review, I saw your ping but the last week was busy. Anyway, thanks for the tests!

ModifiedType gets assigned during construction and getModifiedType() is just an accessor, so if that is nullptr something very bad is happening. We should not expect that except in case of a parser bug of some sort.

Used CFLAGS_EXTRAS instead of CFLAGS at Greg Clayton's suggestion.

Updated to reflect Pavel Labath's suggestions:

Two changes after Greg and Pavel's suggestions:

Committed r303110.

Updated the Makefile to fix a problem caught by Pavel Labath.
Also relocated the new test to lang/objc.

Improved symbol lookup per Greg and Jim's suggestion, with flipped steps 2 and 3.
Also added testing for the error case.

I also would flip 2 and 3. I'll give this a shot.

• Added a FIXME per Aleksei Sidorin's suggestion.

Eliminated a missing newline at end of file error

A few minor nits, but the operation of ignoring certain sanitizer flags seems to be happening in the wrong place.

Closed by r301273

Looks good to me. This is a good step toward simplifying the ASTImporter and breaking out its parts to be useful on their own.
Obviously make sure the tests pass :)