In #125558 we ran into inconsistent ordering of error messages between CI and our local machines.

Our fix was to re-order the top-level definitions, so that global_asm! comes before a function containing inline asm!.

Here the first commit failed on CI, the second one works. The only change is ordering of top-level definitions: 08c8887 .. 8fd7cc9

gist containing source files because idk how long force pushes are kept

@oli-obk had this theory

It's probably some code iterating over a hashmap and invoking queries as it does. Should be fairly easy to find as it must be in thr backtrace of the error (you can get that with -Ztreat-err-as-bug=index_of_the_error_in_the_output)