[kddnewton]

No description provided.

[hasannadeem]

The two test-all failures (here and on #4339 and #4340, which stack on this) come from building CRuby, not from the tests. ruby/ruby still uses the old pm_constant_id_list_t API in two places. CI only shows the first one, because make stops at compile.o before it compiles vm.o:

1. prism_compile.c:3611 sets the removed field when it builds the synthetic pm_def_node_t for mandatory_only_iseq:

   prism_compile.c:3611:14: error: 'pm_constant_id_list_t' has no member named 'capacity'
    3611 |             .capacity = def_node->locals.capacity
   

   Deleting that line is enough, since .size there already comes from parameters_node->requireds.size.

2. vm_eval.c:1871, which is compiled as part of vm.c, calls pm_constant_id_list_init(&parent_scope->locals) with the old one-argument signature. pm_constant_id_list_init(result.arena, &parent_scope->locals, 0) keeps its behaviour, because the loop after it fills the list with pm_constant_id_list_append.

I compiled both call sites, unchanged, against this branch's constant_pool.h. They give exactly those two errors (too few arguments to function call, expected 3, have 1 for the second), and with both changes they compile cleanly. Nothing else in ruby/ruby outside prism/ uses the API this changes, so the sync needs both changes.

Also, the CHANGELOG entry covers the struct change, but pm_constant_id_list_init is exported and its signature changed too, from (list) to (arena, list, size). It also now sets size instead of starting the list empty. That could be worth a line for other C API users.

Edited: the first version of this comment missed vm_eval.c and said CRuby doesn't call pm_constant_id_list_init.