FindTypeAlias failed to find puint8! on stderr, compile continues

• Type: bug (compiler core) — Track A. FindTypeAlias and the alias table are shared ground, not frontend ground; a core internal printing to stderr during a normal successful compile is A's to answer.
• Filed: 2026-08-20 by frank3, observed while driving the rtl-generics corpus stage for [[feature-pascal-corpus-generics]] (generics.defaults).
• Possible loose end, not a new defect: PUInt8 was cleared as an earlier wall of the same corpus climb. Check that fix first — this may be its remainder rather than anything new.

What was seen

The line appears on stderr, once, during a compile that exits 0 and produces a working binary. AliasCount=36 suggests the table is populated and puint8 specifically is not in it.

The fork, which is why this is filed rather than fixed

Two readings, and they want opposite fixes:

1. The lookup failure is real. Something asked for puint8, did not get it, and carried on with whatever the failure path leaves behind. Then this is a silent wrong type with a warning attached, and the priority above is too low.
2. It is a leftover debug print. The caller handles a miss correctly and the message is noise that should never have shipped — a compiler that prints internal diagnostics on a successful build trains everyone to ignore its stderr, which is how the next real message gets missed.

Establish which before fixing: find the FindTypeAlias call site that misses, and read what it does with the failure. Do not infer it from the wording.

Repro

Not yet reduced — seen only in the full rtl-generics drive. Reducing it is the first task; grep -rn "failed to find" compiler/ locates the print.

Resolved — it was already fixed, two days before this ticket was read

Reading 2 was right, and the fix landed on 2026-08-20 as 9d230105b, "remove the leftover puint8 debug dump from FindTypeAlias". Nothing in the tree prints it today: grep -rn "failed to find" compiler/ lib/ tools/ is empty.

From that commit's own message: "A lookup miss there is normal — puint8 resolves through the pointer-alias path afterwards — but the miss printed the whole alias table to STDOUT first." So the ticket's question is answered on both halves: the miss is not a wrong type (the pointer-alias path takes over), and the message was noise. The ticket's own priority note ("if reading 1, the priority above is too low") does not apply.

One correction to the ticket, worth keeping because it affected how the symptom was described: the dump went to stdout, not stderr. That is why it was seen at all — it landed in build output rather than in a stream people filter.

Closed as done citing the existing commit rather than re-fixed. The "establish which before fixing" instruction was the right call and cost nothing to honour: git log -S "failed to find" answered it in one command.