origin/master has advanced 1 commit(s) since this sha. Re-verify at current HEAD before acting — the callback is tagged to the sha that was tested, which may no longer be the state of the tree.

regression: test-core#src:test/test_algol_skeleton.alg red at ad8e212cf739 (auto-filed by twatch)

Type: regression (auto-filed by Track T watcher, host plexus). Untriaged.
Found: 2026-08-05T17:47:16Z
Test source: test/test_algol_skeleton.alg

Repro

tools/testmgr.py --tier native --job 'test-core#src:test/test_algol_skeleton.alg' at ad8e212cf739231cf0055e382af45703a8de4407

Range

bad ad8e212cf739, last good 030b9a625d8c, 4 commit(s) in range — the watcher narrows this by idle bisect; check tstate/TSTATE.md for the current range.

Log tail

pascal26:29: error: compiler error: PXXWriteFloatSci not found
(tail)
pascal26:29: error: compiler error: PXXWriteFloatSci not found
  near:    n   >>>

Stub ticket: signal only. Track T agent (face 2) enriches or a dev track takes it from the repro line.

Resolved (2026-08-05) — caused by e7d4f3336, repaired

Mine, from the correctly-rounded-float fix. Two distinct causes:

1. Algol / Fortran skeletons: a dependency the shim introduced. Collapsing the four float formatters into one made x86-64's EmitWriteFloatSci a SHIM onto the runtime's PXXWriteFloatSci. The esoteric skeleton frontends link no RTL and emit code immediately with no entry jump, so they cannot pull builtinheap — they died with compiler error: PXXWriteFloatSci not found. (Pulling the unit from ParseGProgram was tried: it compiles and then SEGFAULTS, because those frontends have no entry jump for unit bodies to land behind.)

Repaired by keeping the original hand-written emitter as EmitWriteFloatSciSelfContained, used ONLY when FindProc says the RTL writer is absent. Every program that links the RTL — i.e. every real one — still gets the exact formatter. The fallback carries a comment saying precisely what would let someone delete it: give the probe frontends an entry jump.

2. test_float_write / test_writeln_nonfinite_float: stale expectations. They encoded the OLD, WRONG digits:

expectation	was	correct	oracle
1234.5	`1.2345000000000002E+003`	`1.2345000000000000E+003`	CPython `f'{1234.5:.16e}'`
1e300	`1.0000000000000007E+300`	`1.0000000000000001E+300`	FPC and CPython

1234.5 is exactly representable, so trailing ...0002 was never right. Updated against the oracles, not against pxx's own output.

Verified: testmgr --tier native 1158/1158 pass (includes the self-host fixedpoint and the fpc-bootstrap).

2026-08-05 — resolved, commit d26d0093e.