@kddnewton (Kevin Newton) any updates, or reproduction issues? I'm confident I fixed the issue OP reported.

Yeah, it broke CI on macOS https://github.com/ruby/ruby/actions/runs/16430139182/job/46429723627#step:9:179

So this was partially reverted via https://github.com/ruby/ruby/commit/7dbd9c26361719a45fa39838f46a76d67dc3c2e9. It reverts the ruby/prism changes but kept the compiler changes from https://github.com/ruby/ruby/pull/13966.

I think the macos issue might have stemmed from the fact that all flags were cleared. That includes the one from frozen_string_literal: false which rbconfig.rb has. Seems like something that could cause this, and indeed the mutable flag got cleared from "foo""bar" which rbconfig.rb contains on macos

You added a test that "'a' 'b'" should not have PM_NODE_FLAG_STATIC_LITERAL but that doesn't look correct to me? At least from the description about that flag from the docs (A flag to indicate that the value that the node represents is a value that can be determined at parse-time). It does not mean frozen is my understanding.

I feel like the prism side is already correct, just some compiler optimization is missing some extra condition somewhere to not apply. "'a' 'b'" is static, "a" "b#{c}" is not, which is my understanding of how it should be behave.

(For some reason I thought the compiler changes were enough by themselves but I tested now on 3.4.6 and master, and the behaviour OP describes is still present). So, indeed not fixed.

Turns out the prism commit was reverted before the compiler change was merged. Then the ruby/ruby PR was merged, which contained the reverted code again, essentially fixing this issue.

Then https://github.com/ruby/ruby/commit/e74524616013c616744ebf8168f1ad57eee74a05 came along and "broke" it again, since it gave the impression of being reverted during 3.4.6 release. A bit of a mess to be honest.

I did notice an issue with the original approach where these concatinated strings no longer warn about chilled strings. It has to preserve frozen-ness/mutability when concatinating string literals like "foo""bar". I have https://github.com/ruby/prism/pull/3667 for the prism side and https://github.com/ruby/ruby/pull/14697 for the compiler.

Turns out the prism commit was reverted before the compiler change was merged.
A bit of a mess to be honest.

I think the maintainers understand this, but the thing is

• A Prism change that needs a compiler fix should be merged to ruby/ruby first, and then reverse-synced to ruby/prism.
  • When you merge such a change to ruby/prism first and it breaks ruby/ruby CI, the change will be reverted, at least on the ruby/ruby side. This started the out-of-sync situation.
  • If you don't want this to happen, you probably want to have prism_compile.{c,h} in ruby/prism as well. cruby-bindings.yml would test it, so it's not impossible.
• When something is changed on ruby/ruby master (in this case, a sync from ruby/prism was reverted in ruby/ruby), it should be reverse-synced to ruby/prism as soon as possible.
  • There's nothing that prevents you from leaving the out-of-sync ruby/prism at the moment. This time, they were out of sync for a while without anybody noticing it.
  • Perhaps cruby-bindings.yml should fail at the tool/sync_default_gems.rb step when there's a diff not made by the Prism PR itself. It should let you notice ruby/prism is out of sync (due to a revert in ruby/ruby, for example) when it fails there. There's no point in running the CRuby test when it's not going to be the ruby/ruby code after merging the ruby/prism PR.

It was my bad, I did not look at the history and only noticed what happened in what order now. From what I gathered someone was notified of the revert.

Thank you for taking the time to explain the process in such detail. I was uncertain what should happen in which order. ruby/ruby first and clean up after makes sense to me.