Is the code in the wrappers identical? Can we make them linkonce_odr symbols instead? Although I'm not sure how we would do that with non-LLVM AOT if it needs this, too.

Also, is this going to be a nontrivial size increase if we're doing this for every wrapper?

Making them linkonce is kinda complicated, we have a dedup pass instead which only emits them once into a dummy assembly, but that is not used on ios right now.

It will lead to a size increase, but only if the interpreter is enabled.

Making them linkonce is kinda complicated, we have a dedup pass instead which only emits them once into a dummy assembly, but that is not used on ios right now.

Is the naming change going to interfere with the dedup pass on platforms that use it?

It will lead to a size increase, but only if the interpreter is enabled.

We should make an issue to revisit this if there will be situations where we'll run the interpreter together with AOT on ios

Making them linkonce is kinda complicated, we have a dedup pass instead which only emits them once into a dummy assembly, but that is not used on ios right now.

Is the naming change going to interfere with the dedup pass on platforms that use it?
No.

It will lead to a size increase, but only if the interpreter is enabled.

We should make an issue to revisit this if there will be situations where we'll run the interpreter together with AOT on ios

@monojenkins build failed

@vargaz @lambdageek @marek-safar Any objections to backporting this to 2020-02? I installed a fresh copy of VSMac and Mono recently and #20417 still happens.

@imhameed I think it's ok to backport

/backport to 2020-02

Started backporting to 2020-02: https://github.com/mono/mono/actions/runs/1463343749