This is nice; thanks for the cleanup!

Thanks!

This is largely incomplete, not filtering out platforms and architectures.

This is largely incomplete, not filtering out platforms and architectures.

• Make archive input files own their archives again

• Inlined the archive*.s file inputs
• Removed unused return value
• Retweaked the inputs for symbol-order to reflect the files that inspired them
• Re-ran check-lld once more

• Inlined the test files for symbol-order
• Changed symbol-order to check sections instead of running executables
• A couple other lil changes
• rebased

• Add a test for symbol order resolution based on @smeenai's wonderful work
• Get rid of the input section function in favor of immediately caching Archive inputs during fetch
• Test format cleanup: use *.o naming
• Tried to use llvm-nm -p, test started failing, punting for now
• Explicitly include Archive.h

Remove testing cruft, oops

Added two tests inspired by the ELF implementation -- a bad archive, and an archive with no symbol table.

• initialization list for SyntheticSection
• caching position for "default" sections
• objdump instead of readobj in test
• constants for __DATA_CONST
• llvm_unreachable

Rebase, go through most comments (still need to add an erroneous archive test)

clang-format

Rebased, aligned file offsets, and tested with UBSAN.

Comments. More significantly, added more checks to verify __text merging.

• Move comparator logic within output segment / section
• Refactored creation of synthetic classes to use a base class / add to segments within constructor
• removed instances of "this"

Rebase

tweak comment

Pruned unnecessary hidden checks, coalesced some functions, removed unused imports

Refactored to turn synthetic input sections into output sections. A new class represents merged input sections, the MergedOutputSection.

Rebased, various comments about renames, basic test

Added a test; unfortunately symbol export is blocking the text so I need to rebase

Passing most local tests now

Output is horribly wrong but at least it's not crashing now when running symtab.s

Now in a buildable state

Addressing request to split generating content and assigning addresses. This doesn't functionally
do anything different than before since nothing in the writer is parallelized, but in principle
this could be adapted to be so.

Since D76839 is based on this diff, it'd be easier to land this diff as-is so that D76839 doesn't have to rebase. Any qualms about landing as-is and then changing the write order in another diff?

My understanding of this patch is that this creates section contents early. But can you avoid doing that?

Address comments, rebase on D76252

Sorry for the delay in response, and thanks for the review so far!

Verify that address offsets for the text segment are reasonable

Addressing first round of comments (minus the value checks in the test)