Can we factor the LLVM targets information out of `--version` and into its own flag?

[bradcray]

I’m finding the “available LLVM targets” in the chpl --version output to have made what used to be a fairly short and sweet message into something that's now much longer without adding information that I typically find myself caring about. In this issue, I'd like to propose pushing that additional information into another flag (like chpl --print-llvm-targets , chpl --help-llvm-targets or chpl --llvm-targets?)

If we imagine we’ll normally be the ones asking the users to send us this information, we could leave it at that. If we think they’re going to want to find it on their own, we could change the LLVM version line to read:

built with LLVM version 14.0.6  (use --print--llvm-targets to see a list of available targets)

(where my preference would be the former if we don’t think there’d be a downside). If we thought this was something we were going to want from users typically, we could add it to the GitHub bug template; or if we thought we'd primarily want it for GPU bugs, we could create a GPU-bug-specific template.

[mppf]

Note that listing the targets was added in PR #22170 by @stonea :

... it also seems relevant to include what targets LLVM was built to support (e.g. NVPTX, AMDGPU, etc.) and so I now have that information also be displayed when we pass --version.

My own opinion on this is that moving it to another flag sounds fine. I could go either way on the message along the lines of use --print--llvm-targets to see a list of available targets. If we include it, let's please put it on a new line just after the LLVM version.

[stonea]

I think moving that information so it only prints out if passing a special --print--llvm-targets flag would be fine.

Mostly I added this because of concerns expressed here: #22087 (comment)

Though thinking about it some more, I'm honestly not sure how much value adding a --print-llvm-targets flag would be, because:

• (at least with GPU support), if the user is in a state where they need a given target and its not built in chplenv already errors out letting you know:

  chapel/util/chplenv/chpl_gpu.py

  Line 175 in c8cf061

  def validateLlvmBuiltForTgt(expectedTgt):

• If I saw a user issue where this failed, I'd probably just suggest using CHPL_LLVM=bundled as a next step anyway.
• It's easy enough to just get target information from llvm-config --targets-built
  • Though I suppose it's possible that the llvm-config on the user's path could refer to a different llvm.