← board

Revert the __pxx_clock Int64-cast workaround

Track A owns the fix but not the file: lib/rtl/pxxcio.pas is Track B's lane, so this is handed over rather than done in place.

What to do

devdocs/dev/track-b-workarounds.md row 1 and the comment block in lib/rtl/pxxcio.pas's __pxx_clock both say "revert to the one-liner when that closes". It has closed. Replace the two-step implicit widening with the idiomatic form:

Result := Int64(ts.Sec) * 1000000 + Int64(ts.Nsec) div 1000;

Then delete the WORKAROUND comment block above it and drop the row from track-b-workarounds.md.

Already verified (so this is mechanical, not exploratory)

The exact expression, on the exact shape (NativeInt record fields), gives 1234567890 for Sec=1234, Nsec=567890000 on x86-64, i386 and arm32 with the fixed compiler — measured. And C's clock() compiled through crtl now returns plausible values with a non-negative delta between successive calls on all three, which is the behaviour the workaround was protecting.

The same shape is covered by test/test_int64_cast_of_nativeint.pas (Track A's regression test), so a reintroduction of the compiler bug would be caught before this line is.

Gate

Track B's usual: build with $(PXX_STABLE), make lib-test. Note the pin boundary — PXX_STABLE is pinned, and the compiler fix is not pinned yet. Until Track A runs make pin, pinned still has the bug and the reverted one-liner would be WRONG when built against it. So either wait for the pin, or verify against compiler/pascal26 and land the revert only once pinned has moved. That ordering is the only real hazard here.