← board

FindTypeAlias prints a debug dump for puint8

Symptom

program pu;
var x: puint8;
begin WriteLn('hi'); end.



FindTypeAlias failed to find puint8! AliasCount=12
  Alias 0: PVarRecInt64 (noff=84, nlen=12)
  Alias 1: PVarRecDouble (noff=101, nlen=13)
  ... nine more ...
ok: /tmp/.../pu  [code=55344B ...]

Thirteen lines of compiler internals on stdout, then a successful compile.

Why it survived

Both halves of what makes a bug invisible were present:

  1. The compile SUCCEEDS. A miss in FindTypeAlias is normal — puint8 resolves through the pointer-alias path after that function returns -1. So the dump fires on a lookup whose failure means nothing, and the program builds and runs correctly. Nothing is wrong with the output binary.
  2. No test asserts on the compiler's own stdout. Every test here checks what the compiled PROGRAM prints. A compiler that prints thirteen extra lines passes all of them.

puint8 is not an obscure name to trip over, either — it is the natural spelling for a uint8* in C-interop Pascal, which is exactly the code most likely to hit it.

Fix

Deleted the block, and the two locals (s, j) that existed only to build its strings. Replaced with a comment saying why a miss here is normal, so the next reader does not add diagnostics back for the same reason.

Test

test/test_puint8_no_compiler_spew.pas — the assertion is on the compiler's output, not the program's: no FindTypeAlias/Alias 0: line, and exactly one line total (the ok: banner). That shape is worth reusing; it is the first test in the suite that checks the compiler is QUIET, which is the property that was missing rather than any property of the emitted code.

Bites: the pinned binary emits two matching lines.

Gate

make compiler/pascal26 (byte-identical fixedpoint) + tools/gate.sh quick GREEN.