Address concerns about table subscripting. Add warnings for that.

Remove references to table subscript. Simplify patch.

Rebase on main.

Thanks all for the comments, fixed and will soon land after running a check-all.

@erichkeane @aaron.ballman any further comments on this?

Address comments regarding tables and remove some unused code.

Address the comments.

Add forgotten comment to test.

Patch with all outstanding comments addressed.

Thanks @aaron.ballman . I think I have now address all outstanding comments.

Rebased on current HEAD.

Rebased on current HEAD.

Rebased against HEAD. Waiting for @vitalybuka go ahead to land.

@aaron.ballman I have finished addressing all the concerns on this patch. Do you have any further comments?

Remove unnecessary code.

Further refinement of the patch with more diagnostics tested.

Addressed a few comments.

Address some of the pending comments.

Address last comment before landing.

Rebase on main and fix a test diagnostic expectation.

Rebase on main.

Rebase on main.

@aaron.ballman Thanks for all the comments, I have now finished addressing those. What do you think of the current patch?

Add missing newline at end of file.

Simplify and add tests for remaining diagnostics

Further tests for use of tables in conditional branches.

Finish fixing tests after dianostic message changes

Fix some tests whose diagnostics changed

Address more comments.

Remove some unnecessary diagnostics.

Add new Attr Subject Function Pointer for __funcref attribute.

I feel like the latest patch addresses all concerns until now. What do you think?

Rebase on main.

Updating tests after some changes to previous diagnostics.

Address a few comments, add new test for funcref keyword diagnostic.

Rename test that still had tables in its name.

Fix tests after changing name mangling and error messages.

Update mangling for Microsoft.

Now under the correct revision: @aaron.ballman Added support for externref mangling in Microsoft. This follows the pattern implemented by OpenCL types.

Undo last patch. Submitted change to incorrect revision.

In D128440#4047535, @pmatos wrote:

@aaron.ballman Added support for externref mangling in Microsoft. This follows the pattern implemented by OpenCL types.

@aaron.ballman Added support for externref mangling in Microsoft. This follows the pattern implemented by OpenCL types.

@aaron.ballman many thanks for the thorough reviews on the patches. Happy to see this moving in the right direction.

Address all comments except a couple of issues related to ABI and mangling.

Deal with Itanium Mangle and assert false outside Wasm triple.

Address a couple of comments by @aaron.ballman

Addressed a few comments downstream. Will upload new patch next.

@aaron.ballman Hi Aaron, since you requested an RFC for this. I wonder if you have any comments for this or the other related patches: D122215, D139010

Fix test broken after recent change to opt call. Rebase again on main.

Rebase on top of main.

Rebase on top of main.

Rebasing on top of main.

In D122215#4000612, @MoritzS wrote:

In D122215#3998105, @pmatos wrote:

In D122215#3991648, @MoritzS wrote:

Thanks for the patch! I just tried it out and I think this enables many interesting use cases for WebAssembly when using C/C++.

Currently the lowering to wasm does not work when using externrefs when compiling without optimizations. Is that intended behavior?

I have posted a new patch that fixed this. Can you please try it again and see if it works as you expect?

Thanks for the quick update, it works now!

Does this affect the debugging experience of the generated wasm programs? If you force mem2reg to run for all programs, no function arguments will be written to the "stack" referenced by the __stack_pointer global variable anymore. I don't know whether any wasm runtime actually uses that for debugging, though.

In D122215#3991648, @MoritzS wrote:

Thanks for the patch! I just tried it out and I think this enables many interesting use cases for WebAssembly when using C/C++.

Currently the lowering to wasm does not work when using externrefs when compiling without optimizations. Is that intended behavior?

Force mem2reg on all optimization levels to avoid allocas.

In D122215#3995105, @pmatos wrote:

In D122215#3991648, @MoritzS wrote:

Thanks for the patch! I just tried it out and I think this enables many interesting use cases for WebAssembly when using C/C++.

Currently the lowering to wasm does not work when using externrefs when compiling without optimizations. Is that intended behavior?

No - thanks for trying this out. This is definitely not intended behaviour. Somehow I probably didn't test enough without optimizations. I was expecting that the pass mem2reg would get rid of the allocas, but maybe it's not running at those optimization levels. I will have to take a closer look.

In D122215#3991648, @MoritzS wrote:

Thanks for the patch! I just tried it out and I think this enables many interesting use cases for WebAssembly when using C/C++.

Currently the lowering to wasm does not work when using externrefs when compiling without optimizations. Is that intended behavior?

Ready for review.

Ready for review.

This is ready for review.
RFC has been published here: https://discourse.llvm.org/t/rfc-webassembly-reference-types-in-clang/66939

Remove unnecessary attribute that lingered around after some refactoring.

Subsumes D124162 and D123510.

Please wait to review until the RFC (coming soon) is published.

Updated this revision on top of D122215.

Whole integration plan still in the works, so marking as planning changes.

Removed address space wasm_externref which is unnecessary with current design.