Disable RTTI in the toolchain

[danakj]

This reduces object sizes which is desirable for linking speed.

zygoloid did some analysis to determine if any of our code requires RTTI for dynamic_cast here: #7532 (comment):

The only thing I found is that libc++ requires dynamic_cast in order for std::print to correctly write Unicode to terminals on Windows

We use llvm::print functionality, not std::print, so this doesn't affect our toolchain.

[github-actions]

Depends on #7532, start review with these changes

(Last updated: 2026-07-22 17:14:15 UTC)

[danakj]

Looks like we use typeid in diagnostics:

In file included from ./toolchain/lex/helpers.h:8:
./toolchain/diagnostics/emitter.h:533:16: error: use of typeid requires -frtti
  533 |                typeid(Arg).name(), typeid(Storage).name());
      |                ^
./toolchain/diagnostics/emitter.h:533:36: error: use of typeid requires -frtti
  533 |                typeid(Arg).name(), typeid(Storage).name());
      |                                    ^
2 errors generated.

[zygoloid]

#7554 removes the typeid usage throughout the repository. We were only using it to get a static type name, and LLVM provides a better utility for that.

[chandlerc]

We should probably keep these. We'd need it for modules to work, and header parsing should give the same error messages.

[danakj]

These moved up to line 70 (this is in the dependent PR)

[chandlerc]

Rather than this workaround, I think we should just fix the runtimes to correctly force these features on. I think this is just about adding -fexceptions and -frtti to the libc++ (and libc++abi) copts:

https://github.com/carbon-language/carbon-lang/blob/trunk/bazel/llvm_project/0005_Introduce_basic_sources_exporting_for_libcxx_and_libcxxabi.patch#L181-L183

Those flags should be ordered after the project feature flags, and allow us to re-enable things.

[danakj]

I see, okay I will look at regenerating patches, or adding another patch (but not today).

[chandlerc]

You could also add them explicitly to the two libcxx rules instead of the variable. No patch generation and we already have -w in both for similar reasons so minimally hacky workaround