Repository navigation
t27o: Python front end, with types inferred from oracle traces #8995
Description
Part of the t27o service epic. A Python front end for t27o. Python has no static types, so they are inferred from oracle traces: run the original on sample and boundary inputs on the lab, then record argument and result types. Python ints are unbounded, so pick widths from the traces and refuse beyond them.
Priority: this is the largest hand-written foreign debt in our own repos (scripts/.py, tools/.py; see the only-t27 rule, #8236).
Acceptance: a coverage and equivalence table over scripts/ and tools/, and the first verified ports land as .t27 with the Python deleted (net-negative PRs).
Activity
gHashTag commentedon Oct 19, 2026
Measured 2026-10-20 (census #11168, document
docs/now/2026-10-20-python-debt.mdin PR #11176): of 331,183 distinct hand-written, non-test Python lines in the five repositories, 812 (0.25%) sit in the 105 functions that fit a typed subset today and 4,090 (1.23%) in the 454 that fit once parameter types are inferred from traces (this issue's design). No file is a whole-file candidate of 150 lines or more, and the median in-subset function has 6 lines, so a Python port deletes functions, not files, andt27c gen-pythonlowers no function today. Pure code blocked only by heap values (owned strings, dict/set, comprehensions, union/None returns, pure stdlib imports) is a further 1,293 fns / 16,577 lines (5.0%): the same wall as the Rust census.Verdict from the loop, on the threshold set before the measurement (a whole-file candidate of at least 150 lines): not met, so no Python front end is being built now. This issue stays open as the roadmap item; what would change the answer is t27 gaining the heap values named above, which is the decision recorded in the t27o report (default: not started).
docs(t27o): Python debt census -- 2,126 tracked .py files, 12,547 fun…