Addressed comments

Address AddressesMap interface comments:

Indeed

Conditionalize relocations saving

In D158124#4651839, @avl wrote:

/usr/bin/time -l would be enough:

In D158124#4651700, @avl wrote:

I have a general question - Could you measure difference with run-time memory requirements(f.e. for clang binary) with and without this patch, please?

Is there a specific test suite/benchmark/tool somewhere that you want to be run here, and can point to?

Update doc and remove dead code

Rebase against main

Addressed feedback.

Address inline comments.
However, regarding all the DwarfLinkerForBinaryRelocMap and the discussion about wether or not some logic should be in DwrafLinker (as they're all related).
I agree in the sense that my initial incline was to limit all that logic to AddressManager.
The issue here is that with DwarfLinkerForBinary, we have access to Obj->Addresses, which is the AddressesMap, and there is no reasonable way do downcast it to AddressManager (as dyn_cast doesn't work here), which is why I had to resort to have it being part of the AddressesMap map interface and call via virtual dispatch

I have nothing necessarily against the suggestion to help the parallel implementation, only a few points:

Addressed inline comments

Properly add binary files

Rebased against main branch

Addressed review comments. Mainly:

• handle DW_AT_APPLE_origin in a more "standard" way.
• Improve computation of relocation offsets for attributes. This also means that the relocations can be simplified to not keep track of the offset within the compile unit. Similarly, the DW_AT_APPLE_origin is not tracked in particular for the slide calculation.
• Given that the relocations are not stored after applyValidRelocs, added a saveValidRelocs interface (naming is for consistency with applyValidRelocs and addValidRelocs).
• Renamed the parent relocation class to be more generic.

In D158124#4601107, @avl wrote:

3. Skipping and then inserting DW_AT_APPLE_origin attribute looks a bit screwed. Can we just update it if the attribute is presented(as it is done for other attributes) and add it if the attribute does not exist?

That sounds reasonable.

Missing files

Format

Addressed comments

Addressed comments

Add missing DWARFLinkerRelocs.h file

remove empty whitespace

Thanks for the review!

Rebase on top of HEAD.
Ping.

Ping.

It was tested against the clang extra unit tests, but not tests run with check-clang-tools.
Updated Yaml test files to match the new format, this might need further review.

Fix clang compilation warnings. These didn't appear when compiled on Windows.
Tested on a Linux distribution, should be fixed.

Rebase on top of HEAD.
Ping.

I don't have commit access.

Rebase on top of HEAD.

Ping

In D26137#602602, @malcolm.parsons wrote:

In D26137#602591, @Alpha wrote:

This shouldn't affect diagnostics without fixes. If there is no fix, there won't be anything to export, and the diagnostic just behaves normally.

That's a shame; I need a machine readable report of all diagnostics.

This shouldn't affect diagnostics without fixes. If there is no fix, there won't be anything to export, and the diagnostic just behaves normally.

Ping

Ping

Export effectively MainSourceFile.
Change CheckName field.
Add doxygen-style comments.

Ignore export of empty fixes.

Ping

Remove debug symbols.

Fix diagnostic deserialization bug for clang-apply-replacements.