Toolset update: VS 2022 17.7 Preview 1 #3711
Conversation
… repeat a type N times".
…n: !"If lookup found RDSymbol before it should find something this time, too", ParseTree.cpp 10376'.
…ebug mode with 'attempt to access expired storage' errors, part 3".
…nto requirements of class template members not implemented".
… remove workaround in `P2474R2_views_repeat`.
…tract`. Update `floating_point_model_matrix.lst` to silence these warnings.
…raint expressions effectively cannot be reproduced since each utterance is distinct).
…sues in `<ranges>` and `<type_traits>`.
…calls on Windows".
The changes are almost entirely positive, with no horrible mangling, and limited weirdness around `&`, `&&`, and `*`. No need for manual fixups.
`BreakBeforeInheritanceComma` and `BreakConstructorInitializersBeforeComma` were never documented and aren't printed out by `-dump-config` anymore. The weird empty `BasedOnStyle` isn't printed out by `-dump-config` anymore. `ConstructorInitializerAllOnOneLineOrOnePerLine` and `AllowAllConstructorInitializersOnNextLine` are now deprecated and aren't printed out by `-dump-config` anymore.
…used into `LineEnding`, no behavioral changes.
…ew options, no behavioral changes.
… no longer a bool, no behavioral changes.
|
I'm speculatively mirroring this to the MSVC-internal repo - please notify me if any further changes are pushed. |
| _Different_from<_TupleLike, _Tuple> && _Tuple_like<_TupleLike> | ||
| && (!_Is_subrange_v<remove_cvref_t<_TupleLike>>) &&( | ||
| tuple_size_v<_Tuple> | ||
| == tuple_size_v<remove_cvref_t<_TupleLike>>) &&_Can_construct_values_from_tuple_like_v<_Tuple, _TupleLike> |
There was a problem hiding this comment.
Ick. (No change requested, but I wouldn't complain.)
| // TRANSITION, VSO-1690654 | ||
| #ifdef __EDG__ |
There was a problem hiding this comment.
Not worth resetting testing, but I note that this will fit on one line now - which is how we'd conventionally write it.
| #define _STL_128_DIV_INTRINSICS 1 | ||
| #endif // ^^^ detect _udiv128 / _div128 ^^^ | ||
| #ifdef __clang__ | ||
| #define _STL_128_ADD_SUB_INTRINSICS 0 // clang doesn't have _addcarry_u64 / _subborrow_u64 |
There was a problem hiding this comment.
Naming nit: these are 64-bit intrinsics, not 128-bit intrinsics.
More serious: This is a lie. I can reproduce the issue with the clang in 17.7p1, but not when I self-install 16.0.1. Something has gone awry in the VS clang tree, and we need to run it down.
I don't really care about the naming given that I suspect this will be a temporary workaround. I've filed #3713 to investigate this.
…nsion to repeat a type N times"." This reverts commit eb6b4db.
|
I had to restore the workaround for VSO-1761088 "ICE when using a pack expansion to repeat a type N times". (Sorry, not enough time today to address the "nice to have" nitpicks even though I'm resetting testing.) |
|
I had to push an additional commit to skip the |
|
Thanks for enabling us to use current tools! 🔧 (No joke, these toolset updates are a ton of work.) |
New pool
Remove workarounds
Remove workaround for VSO-1761088 "ICE when using a pack expansion to repeat a type N times".LLVM 16 changes from @CaseyCarter
P2474R2_views_repeat./fp:strictwith-ffp-contract. Updatefloating_point_model_matrix.lstto silence these warnings.adjacent_transform_view(atomic constraint expressions effectively cannot be reproduced since each utterance is distinct).tuplecomparison operators.Xnested in class template is ill-formed whenXhas class template friend with specific constraints llvm/llvm-project#62290, which has been fixed for Clang 17, in<ranges>.TRANSITION, Clang 16issues in<ranges>and<type_traits>.P0896R4_views_take.LLVM 16 followups from @StephanTLavavej
_EXPORT_STDboth declarations and definitions.Add/update workarounds
_addcarry_u64/_subborrow_u64for Clang.LDBL_DECIMAL_DIGis now missing for Clang too.tgmath.htest because it fails in the MSVC-internal repo but passes in the GitHub repo (likely related to header overlaying weirdness).P2278R4_views_as_const.printfcalls on Windows llvm/llvm-project#62762 inP0067R5_charconv.clang-format 16
&,&&, and*. No need for manual fixups..clang-format: Update link to Clang 16, no behavioral changes..clang-format: Sort defaults, no behavioral changes..clang-format: Drop commented-out defaults, no behavioral changes.BreakBeforeInheritanceCommaandBreakConstructorInitializersBeforeCommawere never documented and aren't printed out by-dump-configanymore.BasedOnStyleisn't printed out by-dump-configanymore.ConstructorInitializerAllOnOneLineOrOnePerLineandAllowAllConstructorInitializersOnNextLineare now deprecated and aren't printed out by-dump-configanymore..clang-format:DeriveLineEndingandUseCRLFwere deprecated and fused intoLineEnding, no behavioral changes..clang-format:AlignTrailingCommentsis no longer a bool. Use the new options, no behavioral changes..clang-format:SortUsingDeclarations(which we didn't customize) is no longer a bool, no behavioral changes..clang-format: Add new defaults, no behavioral changes..clang-format:InsertNewlineAtEOFis new, set it totrue. Doesn't change any files right now..clang-format:RemoveSemicolonis new, set it totrue. Behavioral change!clang-format update process
As always, the idea is to run
clang-format --style=LLVM -dump-configwhen a new version is available, and then update our.clang-formatsuch that all of the defaults are commented out, followed by any customizations.