← board

Re-price feature-n-nilpy-ast-typing-module-scope?

The fork

feature-n-nilpy-ast-typing-module-scope carries prio: 55, which makes it the top ranked actionable Track N item (above it is only the META ticket feature-nilpy-thirdparty-libraries-as-targets, whose work lands as children). So every agent running tools/progress.sh next --track N is pointed at it.

Its own closing section, written 2026-08-09 after the last round of work, says otherwise:

An unreadable shape no longer costs a wrong value; it costs a tyVariant slot where a narrower one would have done. That is a performance cost, not a correctness one, so this ticket should now be read as an OPTIMISATION item rather than a correctness one — and re-priced accordingly.

The re-pricing it asks for was never applied.

What remains in it, for calibration

The route the ticket names as its real close is: make the module pre-pass able to trial-parse an as-yet-unseen name without Error() halting the compiler — Error calls Halt directly today. That is a Track A change to the compiler's error machinery with a wide blast radius, in service of a tyVariant-instead-of-narrower slot.

Everything that was a correctness defect in this ticket has shipped: item (2) (x = c.two(1)), and the block-nested binding widening (with the tkFloat gap that was hiding half of it).

Options

  1. Re-price to ~25-30 (what the ticket asks for). It stops steering the queue; the real correctness work in Track N ranks above it again. The Error-does-not-Halt refactor gets picked up when someone wants it for its own sake — and it likely has better justifications than this ticket.
  2. Leave at 55. Defensible if the Error/Halt refactor is wanted soon for other reasons (it would unblock several "trial-parse this and recover" ideas), and this ticket is the convenient carrier for it.
  3. Split: drop this ticket to ~25 and file the Error-must-not-Halt refactor as its own Track A ticket at whatever priority that capability deserves. Probably the honest shape — the two things are not the same size and are wanted for different reasons.

Recommendation: option 3. The capability and the optimisation are being ranked as one item, and it is the capability that carries the value.

DECIDED 2026-08-14 by the user — prio 8

"Eight. It makes no sense to optimize a halt()."

Lower than the ticket's own suggested 25-30, and for a better reason than "optimisation ranks below correctness". The work cannot be done as written: the inference needs a pre-pass that trial-parses an unseen name, and Error() calls Halt directly, so a speculative parse cannot back out. An optimisation gated behind a fatal path is not a low-priority item, it is a blocked one.

feature-n-nilpy-ast-typing-module-scope re-priced 55 -> 8, so it stops being the top actionable Track N item. The real work is filed as [[feature-a-error-does-not-halt-so-a-parse-can-be-speculative]] — which ranks on its own merits, and has better justifications than type inference: multiple-error reporting, better diagnostics, and any other speculative parse.

Log