Ping :) @libcxx @libcxxabi

Refactor with macros

improve include's

Resolve merge conflicts

kickoff CI

resolve lit failure

@libc++abi Can I get a review please? Thanks

@libc++ Can I get a review please? Thanks

LGTM

LGTM

Closing since this patch is divided into the following (with comments addressed and better approach):

• Foundational A/E changes: https://reviews.llvm.org/D114813
• Localize runtime error messages: https://reviews.llvm.org/D118179
• Add ASCII/32bit variant: https://reviews.llvm.org/D118503
• Exception derived classes as a separate library for z/OS: https://reviews.llvm.org/D118620

LGTM

Ping :)

Ping :)

addressed comment to be more readable

Great points! Updated to account for this.

safeguard to z/OS, use double underscore function names

Using inline functions instead of macros

ping @libc++ @libc++ab

Add default encoding

Remove LIBCXX_BUILD_ZOS_ENCODING

Ping

LGTM

In D114457#3192772, @shraiysh wrote:

Hey folks, it looks like this commit has caused the buildbot here to run into failures. Can someone please revert this change, fix it and then commit it again?

Added comment

Ping

Remove LIBCXXABI_BUILD_ZOS_ENCODING

formalize LLVM_EXTERNAL_UNWIND_SOURCE_DIR definition in libcxxabi/CMakeLists.txt

Resolve merge conflict with main

Thanks @ldionne for your detailed feedback and suggestions.

renamed zos cache file name

definition of LLVM_EXTERNAL_UNWIND_SOURCE_DIR added

Ping, all comments addressed and green CI.

better format

Verbose description, added correct symbol on abilist(s).

In D110647#3122355, @ldionne wrote:

In D110647#3122226, @muiez wrote:

Any insight as to why the CI is red? It seems to complain about a new symbol being added _ZNSt3__15ctypeIcE10table_sizeE, which is defined. However, that is the intention since we want to resolve the missing symbol.

The CI is testing that there are no ABI changes. This patch contains an ABI change, since it adds a symbol. The change is not an ABI *break* so it is acceptable, but it is nonetheless a change. You need to update the ABI lists to contain that symbol.

You can see how to do that here: https://libcxx.llvm.org/Contributing.html#exporting-new-symbols-from-the-library. Basically, go to https://buildkite.com/llvm-project/libcxx-ci/builds/6180#75b25b94-130a-472d-a409-72a7f5507f17 and click on Artifacts. The libcxx/lib/abi/x86_64-unknown-linux-gnu.libcxxabi.v1.stable.exceptions.no_new_in_libcxx.abilist file will contain the new symbol for table_size -- add it to the corresponding file in libcxx/. You'll then need to do the same on macOS (look at the macOS ABI lists in the macOS jobs once the CI gets that far).

Finally, please also add an entry in libcxx/lib/abi/CHANGELOG.TXT.

Added symbol in abilists. More detail in test case.

Ping :)

Ping @libc++abi

kickoff CI

Ping :)

Ping :)

Any insight as to why the CI is red? It seems to complain about a new symbol being added _ZNSt3__15ctypeIcE10table_sizeE, which is defined. However, that is the intention since we want to resolve the missing symbol.

@ldionne, any thoughts on the approach and/or discussion so far? Your insight would be appreciated.

Use external unwind.h header.

Removed whitespace

Address comments. Use common API by getting LSDA from the unwinder using _Unwind_GetLanguageSpecificdata.

Ping

Update design doc to account for alternate solution to translate libcxx err messages based on char-mode.

Improved solution for error messages, eliminate #ifs.

Clang-format

Any updates @ldionne? Comments have been addressed.

Using out-of-line definition instead of constexpr. This way is more consistent and allows the test case to pass for standards prior to c++17.

Commented out unimplemented functions.

In D111892#3067012, @Quuxplusone wrote:

Above my pay grade, but...

(1) Isn't a runtime error vastly less user-friendly than a link-time error? Perhaps you could show an example of the kind of program that would run into this linker error and/or assertion?
(2) Is a plain assert really more appropriate than either _LIBCPP_ASSERT or throw? (I think _LIBCPP_ASSERT would be less appropriate because it compiles away in non-debug modes; but then, so does regular assert in NDEBUG mode.)
(3) Alternatively, how about either just commenting-out, or =delete'ing, these functions, to turn the linker error into a compile-time (and SFINAE'able) ill-formedness?

Cause compile time error instead.

ping