[WIP] Upgrade frontend & libs to v2.111

[kinke]

No description provided.

[kinke]

Okay, this isn't in a too bad state, especially considering the many accumulated changes for v2.111; most tests are passing, only about 10 failures remaining. We could work on the rest in parallel, by creating PRs against this merge-2.111 branch.

TODO:

• I haven't resolved most merge conflicts for the druntime integration test Makefiles yet, which have been significantly reworked upstream:

  $ git grep "<<<< HEAD"
  runtime/druntime/test/common.mak:<<<<<<< HEAD
  runtime/druntime/test/shared/Makefile:<<<<<<< HEAD
  runtime/druntime/test/shared/Makefile:<<<<<<< HEAD
  runtime/druntime/test/shared/Makefile:<<<<<<< HEAD
  runtime/druntime/test/stdcpp/Makefile:<<<<<<< HEAD
  

  @the-horo: Maybe you wanna have a stab at it, or perhaps even did some LDC porting already? You know everything about the new system. I hope most LDC-specifics in these files are guarded by checking whether DMD contains ldmd2, or at least provide comments wrt. what was added/changed.
• timetrace - some things were upstreamed and now 'conflict' (yeah, even a couple of warnings wrt. duplicate C++ symbol definitions...). Maybe @JohanEngelen or @thewilsonator wanna consolidate these?
• apparent CTFE regression wrt. overlapping initializers for unions: ICE: overlapping initializers for struct literal #4374 (comment). I'll try to fix that in the DMD frontend, don't wanna have to re-deal with that crap in LDC codegen. - edit: Overlapping field initialization in CTFE-generated union literal dlang/dmd#20675; not a CTFE regression (existing issue), but now surfaces after std.json changes in v2.111; [stable] std.json: Avoid setting union members, set the whole union instead dlang/phobos#10683
• apparent frontend regression wrt. lvalue-ness of indexed cast AA (lit-test compilable/gh3162.d) - edit: [REG2.111] Indexing a *cast* AA yields no lvalue anymore dlang/dmd#21020
• adapt LDC codegen to removed _d_newThrowable (runnable/test19317.d)
• add support for placement new (runnable/placenew.d)
• reggae can't be compiled with v2.111, apparently due to a Phobos std.getopt regression, wrt. const(string)[] IIRC (=> const is the problem) - edit: [REG2.111] std.getopt doesn't accept const(string)[] anymore dlang/phobos#10680
• probably have to minimally adapt a unittest assertion in std.experimental.allocator.building_blocks.allocator_list for macOS arm64
• probably have to disable new runnable/exe1.c; seems to test edge cases for compiling C only (former DMC backend tests or something...)

[the-horo]

@the-horo: Maybe you wanna have a stab at it, or perhaps even did some LDC porting already? You know everything about the new system. I hope most LDC-specifics in these files are guarded by checking whether DMD contains ldmd2, or at least provide comments wrt. what was added/changed.

Sure. When I rewrote the makefiles upstream I did think somewhat of ldc but I didn't test anything. I'm a bit busy atm but I'll see what I can do

[the-horo]

• timetrace - some things were upstreamed and now 'conflict' (yeah, even a couple of warnings wrt. duplicate C++ symbol definitions...). Maybe @JohanEngelen or @thewilsonator wanna consolidate these?

#4880 (those warning are errors on my system)

[the-horo]

I haven't resolved most merge conflicts for the druntime integration test Makefiles yet, which have been significantly reworked upstream:

$ git grep "<<<< HEAD"
runtime/druntime/test/common.mak:<<<<<<< HEAD
runtime/druntime/test/shared/Makefile:<<<<<<< HEAD
runtime/druntime/test/shared/Makefile:<<<<<<< HEAD
runtime/druntime/test/shared/Makefile:<<<<<<< HEAD
runtime/druntime/test/stdcpp/Makefile:<<<<<<< HEAD

@the-horo: Maybe you wanna have a stab at it, or perhaps even did some LDC porting already? You know everything about the new system. I hope most LDC-specifics in these files are guarded by checking whether DMD contains ldmd2, or at least provide comments wrt. what was added/changed.

https://github.com/the-horo/ldc/tree/refs/heads/pr-2.111. It worked fine for my one test locally on linux but the other environments might still have issues.

[kinke]

Thanks a lot Andrei!

[the-horo]

Alright, some progress has been made. The tests are running but some of them are failing. importc_compare fails on every platform but there are some other failures as well.

[the-horo]

My action plan right now is:

1. isolate and fix the newly failing tests
2. go through the logs and check that no tests started being accidentally skipped
3. run the tests linking druntime both statically and dynamically when -DBUILD_SHARED_LIBS=BOTH

[kinke]

I suggest you open a PR with that branch; we can remove your 1st commit from #4880 once the timetrace mess has been dealt with (too bad you're getting errors instead of warnings, but oh well). Then we can discuss the druntime integration tests stuff there. And please don't worry too much about remaining failures; once I see them in the CI logs, I can help out as well. Resolving the merge conflicts was already a great help.

3. run the tests linking druntime both statically and dynamically when -DBUILD_SHARED_LIBS=BOTH

Oh, cool stuff! But please keep that separate, in another PR.

[the-horo]

I suggest you open a PR with that branch; we can remove your 1st commit from #4880 once the timetrace mess has been dealt with (too bad you're getting errors instead of warnings, but oh well). Then we can discuss the druntime integration tests stuff there. And please don't worry too much about remaining failures; once I see them in the CI logs, I can help out as well. Resolving the merge conflicts was already a great help.

Here you go: #4883