Description
This is from Claude when investigating whether #772 applies to capgen-v1.
When looking at host vars imported by use ... only:, Claude found that two host variables sharing a local name with different standard names
v1 accepts this silently (exit 0, no warning) and emits a cap where the control-arg dummy shadows the use-associated host variable:
use test_host_mod, only: ncols, nthreads ! host var, std name test_host_thread_count
...
subroutine grp_one_run(..., nthreads, ...)
integer, intent(in) :: nthreads ! control arg, std name number_of_threads
...
call thr_use_run(ncol=..., ni=nthreads, ...) ! binds to the DUMMY
The scheme asked for test_host_thread_count but receives the framework's thread count. It compiles clean and runs with wrong data. I confirmed the Fortran semantics with a standalone test (dummy value 7 wins over host parameter 999), so this is a silent wrong-value bug, not a compile error.
Steps to Reproduce
commit ed284a8a4a6507f8c684839dbe2a2e2770b3a09d (HEAD, upstream/feature/capgen-v1)
see above.
Additional Context
Output
Description
This is from Claude when investigating whether #772 applies to capgen-v1.
When looking at host vars imported by
use ... only:, Claude found that two host variables sharing a local name with different standard namesSteps to Reproduce
commit ed284a8a4a6507f8c684839dbe2a2e2770b3a09d (HEAD, upstream/feature/capgen-v1)see above.
Additional Context
Output