Not compiling with clang19

[bgoodri]

trafficstars

Summary:

Models do not compile with clang19. @SteveBronder or @WardBrian do you know what this means?

Description:

From https://www.stats.ox.ac.uk/pub/bdr/clang19/AovBay.log , which like rstanarm depends on rstan, StanHeaders, etc.

In file included from stanExports_onewaymodel.cc:5:
In file included from ./stanExports_onewaymodel.h:25:
In file included from /data/gannet/ripley/R/test-clang/StanHeaders/include/src/stan/model/model_header.hpp:11:
/data/gannet/ripley/R/test-clang/StanHeaders/include/src/stan/model/model_base_crtp.hpp:140:50: error: a template argument list is expected after a name prefixed by the template keyword [-Wmissing-template-arg-list-after-template-kw]
  140 |     return static_cast<const M*>(this)->template write_array(
      |                                                  ^
/data/gannet/ripley/R/test-clang/StanHeaders/include/src/stan/model/model_base_crtp.hpp:209:50: error: a template argument list is expected after a name prefixed by the template keyword [-Wmissing-template-arg-list-after-template-kw]
  209 |     return static_cast<const M*>(this)->template write_array(
      |                                                  ^

Reproducible Steps:

Try to compile a model with clang19

Current Output:

None

Expected Output:

A compiled stanmodel object

RStan Version:

2.31.x

R Version:

4.4.x

Operating System:

Linux

[bob-carpenter]

From the error message, it looks like Clang 19 wants the empty template args <>, so it would look like this: (this)->template<> write_array(....

[WardBrian]

Clang19 isn’t actually released yet, correct?

[WardBrian]

I believe Bob is correct: https://github.com/llvm/llvm-project/pull/80801

[andrjohns]

Just in case it might help with CI - the RHub project has a variety of docker containers with setups including clang19

[bgoodri]

OK. Does the code need to test for clang version 19 or later or would making those lines be like Bob says be fine for other compilers too?

[WardBrian]

When we open a PR to change model base our CI should tell us whether or not it’s a problem on older clang and GCC

[bob-carpenter]

We should just be able to update. The syntax with the explicit <> should work on earlier compilers. But as @WardBrian says, the proof is in the CI.

[WardBrian]

Ah, upstream resolved this in https://github.com/stan-dev/stan/pull/3228 (part of 2.34) by removing the template keyword in this case. I think adding the <> is equivalent, but removing the keyword seemed more direct