Min runtime ubsan messages are ambiguous

[vitalybuka]

null: https://godbolt.org/z/vqTd9sTPd
alignment: https://godbolt.org/z/zo196vGTY

both cases output "ubsan: type-mismatch by"

[vitalybuka]

CC @fmayer @alazarev

Full ubsan runtime passes parameter which differentiate errors type, but minimal runtime has no parameters.

To avoid code size increase in min runtime we can fork __ubsan_handle_type_mismatch_minimal into more specific handlers,
e.g. __ubsan_handle_null_minimal and __ubsan_handle_alignment_minimal.

It would be nice to have consistency with full runtime, so we would like to fork same way __ubsan_handle_type_mismatch_v1.

Concern is backward compatibility between new compiler and old runtime.

I remember Apple needed such compartibility for asan 6f026ff

Is such compatibility a requirement for Ubsan, either standalone or bundled with asan?

I am not sure how to ask, please share with interested folks.

@devincoughlin @rsundahl @wrotki @delcypher @yln @thetruestblue

[thetruestblue]

@usama54321

[usama54321]

Hi VItaly, Thanks for letting us know in advance. We want to maintain compatibility in the full runtimes, and changing the minimal runtime is not an issue for Apple. Can you add me to the PR once you make the change? Thanks

[vitalybuka]

Can you add me to the PR once you make the change? Thanks

Will do.

@usama54321 What is a horizon for compatibility? E.g. if we land new runtime handlers, when it would be OK to update compiler? It should be possible to limit the change to min ubsan only, it will complicate clang code, depending on wait time, we may consider to go that way.

I guess it's a general problem. E.g. someone decide to add a new ubsan check which is so useful, so it goes into default set of ubsan.

I assume there is not way to tell by Darwin target which set of handlers is availible?