[Lldb-commits] [clang] [lldb] [llvm] Patch series to reapply #118734 and substantially improve it (PR #120534)

[via lldb-commits]

dyung wrote:

> @dyung - OK, I think the current just-pushed version of this PR is worth another test.
> 
> I've taught the TableGen string table emission to go back to working around the MSVC issues using a different table form that we used to use in LLVM when MSVC had a reliable error on it. It appears that now it only sometimes gets miscompiled, and so this workaround got accidentally removed without really being in a reliable place.
> 
> But that still requires that all of the large string tables go through TableGen to get emitted in this form. So I've sent two PRs (separately, they seem independently good) to move NVPTX and Hexagon to TableGen, and this PR contains patches to move more of the ARM builtins to TableGen.
> 
> I _think_ this gets all of the ones that produced too-large string tables, but testing should show. Let me know. If this works and there isn't a really short path to upgrade, I'll start pulling some of the pre-requisite PRs out for independent review, and then update this PR once I'm down to the minimal chain of PRs that do the full conversion of builtins to string tables.
> 
> Thanks, -Chandler

Some good news, everything seems to pass after your latest changes in this PR! I didn't believe it at first and did a clean rebuild and test to verify. In the end everything passed again.

That being said, I am working on deploying an updated version of VS2019 to our internal builders. I have gotten agreement with the plan, the only possible hold-up is that we would need to rebuild our internal builders which likely may not get done for a few weeks. I can apply quick work-arounds to the machines, but they ultimately will need to be rebuilt.

Would you prefer to move forward with the solution currently implemented in this PR or for us to update our internal builders?

https://github.com/llvm/llvm-project/pull/120534