Repository navigation
Refraction tools: three regexes with quadratic scaling (baselined by the #3214 sweep) #3222
Description
Part of #3122. Found by the #3214 sweep (tests/tools/tool_regex_redos.py), which records these in tests/cobol_mainframe/tool_regex_redos_baseline.json so that CI fails only on new offenders.
| # | site | pattern | input it runs over | measured (local, 2026-09-19) |
|---|---|---|---|---|
| 1 | cobol_dag_architect.py ~L96 re.finditer |
\bOPEN\s+((?:INPUT|OUTPUT|I-O|EXTEND)\b[^.]*)\. |
whole program | 0.76s on 40k chars of OPEN INPUT with no period |
| 2 | cobol_graveyard_finder.py ~L308 re.findall (re.I) |
(?:==)?([A-Z0-9_\-]+)(?:==)?\s+BY\s+… |
one COPY statement's REPLACING clause (short in practice) | 0.63s on 4k chars of A |
| 3 | cobol_jcl_forge.py ~L55 (re.I) |
SELECT\s+([A-Z0-9\-]+)\s+ASSIGN\s+(?:TO\s+)?([A-Z0-9\-]+)[^.]*\. |
whole program | 1.0s at 160k chars of SELECT X ASSIGN X , ~16x for 4x the input |
Patterns 1 and 3 have the shape #3205 fixed for EXEC CICS … END-EXEC.: an unbounded run up to a required terminator is rescanned from every start. Pattern 2 is an unbounded name before a required BY.
Direction. Bound each run the way #3205 did, then drop its baseline entry.
Proof.
python tests/tools/tool_regex_redos.py --cireports each entry as "no longer reproduces".python tests/tools/refraction_snapshot.py check, plus--corpus …locally, shows the generated output unchanged, or the PR explains the change.
Activity
test(refraction): pinned mainframe corpora, generated-output snapshot…
fix(refraction): bound the three quadratic tool regexes (#3222) (#3230)