LGTM.

• Rebase to the tip.
• const auto *Contrib -> auto *Contrib.

• Rebase to the tip.

• Rebased to the tip.

Ping.

• Extended the test.

I've created D77302 for renaming getOffset functions.

• Added reporting the version of an unsupported index.
• Added a test for the unsupported TU index version case.
• Removed the actual CU in the test for the CU index version because it is not necessary.

• Added a helper function isKnownV5SectionID().

In D77146#1954186, @jhenderson wrote:

Looks good to me, although I'm not fully up to speed on .debug_loclists to be 100% certain, so might be worth a second opinion.

Thanks!

• Removed DWARFUnitIndex::getVersion() as it is related to the other patch.

• Updated the comment for DWARFSectionKind.
• Simplified the storing of raw section identifiers.
• Moved independent changes into separate patches.

• Use a named constant DW_SECT_INFO in the translation function instead of a numeric constant 1.
• Rename the functions.

LGTM. Please wait for the comments from @jhenderson and @dblaikie.

• Extended the test by adding an unknown ID which is greater than DW_SECT_MACRO.

• The test was extended to add a TU index case.
• Added a comment to the test that the additional sections are required to reach the test points in the code.

In D75929#1926834, @labath wrote:

(btw, is there a test case for the "unknown column" code path?)

• Fixed messages in llvm-dwp.cpp.
• Fixed comments in the tests.

I am mostly OK with the changes at the moment. Please reach the consent with @jhenderson about code duplication and other topics he mentioned.

@jhenderson, thank you for the comments!

We are almost there.

In D75929#1920691, @dblaikie wrote:

This is almost what you are doing right now -- the only difference is that the "internal" enum would no longer be internal -- it would actually match the on-disk format of a v5 index. This v5 enum would contain the official DWARFv5 constants as well as the new extensions we want to introduce for mixed 5+4 indices.

Yep, this would be the direction I would suggest/encourage. It seems that if the goal is to have one index in a DWP file (which seems reasonable) then all future index versions will have to support column indexes all previous DWARF sections - the DWARFvN enum can then be used to describe all the previous versions as well.

• Use values for clashing identifiers proposed by @dblaikie.
• Convert all unknown section identifiers into a special value, DW_SECT_EXT_unknown; Use an optional parallel array to keep the raw values of unknown identifiers.
• Split the refactoring part in llvm-dwp.cpp into a separate patch, D76067.

In D75929#1916466, @labath wrote:

In D75929#1916127, @ikudrin wrote:

I believe that this patch is more or less compatible with any approach which might be taken. The idea is that there is a set of constants for internal use and functions to translate them to/from external representation and both constants and translation functions might be adjusted when needed. In any case, the general design remains the same.

That's true, but I'm not sure it is really the best solution.

Thanks for the links! What a coincidence...

In my understanding, this should still depend on D73086 because this revision requires the dumper for the test(s).

While this can be applied to master cleanly, it cannot be built, because it requires D73086; it applies after D73086 with some conflicts. Please, prepare a diff which will be applied after D73086 cleanly and add D73086 as a parent revision to show the dependency.

The test should be extended to use all supported kinds of entries. There should also be another macro unit without the debug_line_offset field. There should be tests for other combinations of flags in the header to show that the correct diagnostics for unsupported flags are generated.