[ddehilster]

Compiled analyzers used to ignore zone and endzone lines and run every pass on the whole document. They now zone the same way an interpreted run does.

Compiled zone mode

• The generated run_analyzer() runs the passes before the zone line, then hands two generated functions to Arun::zones:
  • zonepasses(): the passes between zone and endzone;
  • endzonepasses(): the passes after endzone, or 0 if there is no endzone line.
• Parse::zoneLoop is the zone loop (find zones, hide the path, run, check, restore, stub). Interpreted and compiled runs now share it; executeZones and runZones are thin wrappers.
• Parse::zonePrefix gives compiled passes the same z001-ana###.tree names and the same three-zone -DEV cap as interpreted ones.

Pass numbers

• 4.2.48 stopped numbering zone/endzone lines in the run, but the sequence loader still numbered them. After a zone line, an NLP++ error therefore named the pass before the one that failed: "pass 4 (words)" for an error in report.
• Seqn::isZoneMarker now keeps zone lines unnumbered everywhere passes are numbered (Ana, Aseq intern and numberSeq, Seqn::gen), so generated pass names match too.

Build

• Parse::iniPass is no longer inline on Windows. Arun.cpp calls it, and it only linked while parse.cpp happened to keep an out-of-line copy. With this change it didn't, and the link failed.

Tests

• New Windows CI step: it builds the zone-mode fixture as a compiled analyzer and diffs its stdout and -DEV tree list against the interpreted run.
• Locally, x86 against hand-assembled compile-libs:
  • the compiled run matches the interpreted run line for line, including final.tree and the z00N-ana00M.tree files;
  • all 56 Windows fixture steps pass;
  • parse-en-us -DEV output on its three inputs is byte-identical to 4.2.48 (454 files);
  • an NLP++ error after a zone line now reports pass 4 (report).

🤖 Generated with Claude Code