Native Save As proposed filename, 2026-11-13
Run: scratch/runs/20261113-native-dialog-input-1115/ (fixture evidence,
not game progress). Investigation began at
533bbba9527a81e89f1e954fdae6905dfd1fda66 on shared main.
The public GetSaveFileNameA handler reproduced an empty filename EDIT when
the caller supplied history.pgn in OPENFILENAME.lpstrFile. The extended
test/test-open-dialog-buffer-too-small.js failed with '' !== 'history.pgn'
before the production change. create_open_dialog always passed NULL text
when creating control 0x442; it never read the input filename. This finding
establishes a current common-dialog defect, not its historical introduction.
src/09c3-controls2-open-save-dialog.wat now scans the supplied filename
within nMaxFile, creates an independent ANSI control-text copy (converting
the W spelling), initializes the EDIT from it, and releases the temporary
copy after control creation. Opening and editing do not write the caller's
buffer. The existing acceptance path remains responsible for output.
The regression calls the public A/W Save As handlers and uses two real WAT
instances over shared memory, matching the common-modal ownership boundary.
For each spelling it checks the proposed name, replacement through native
EDIT selection/WM_CHAR, byte-for-byte preservation on Cancel, zero extended
error, modal completion, and Enter accepting C:\history.pgn on reopening.
Existing small-buffer failures and filename offsets remain covered.
Validation commands and outcomes:
node test/test-open-dialog-buffer-too-small.js: before FAIL on the ANSI proposed-name assertion; after PASS, including both A/W cases.node test/test-commdlg-extended-error.js: PASS.node test/test-open-dialog-filter.js: PASS.node test/test-modal-common-dialog-worker.js: PASS.node test/test-dialog-setfocus-tabstop.js: PASS.node test/test-dialog-key-posts-to-modal-pump.js: PASS.node test/test-worker-modal-key-route.js: PASS.node test/test-wat-edit-target-focus.js: PASS.flock /tmp/wine-build.lock bash tools/build.sh: PASS, including all build gates and compiled data-segment overlap checks.
The broader initial probe also passed test-renderer-shell-dialog.js and
test-menu-post-disabled-owner.js. It found two separate failures:
test-renderer-dialog-modal-input.js expects an event without the current
cursorWarpSerial: 0 field; test-renderer-worker-menu-escape.js fails its
menu-open state assertion (65537 versus 0). Neither test was modified and
neither failure is evidence of a Pawn save stall.
No Pawn game, browser runtime, or box was launched: available disk was below the required floor. No claim is made about Pawn's second turn, committed d4, history export, browser filename focus, or end-to-end Worker typing. The fixture directly dispatches native control messages and does not prove DOM keyboard delivery. Full-path initial-directory handling and default filename extension behavior were outside this change.
Next: after sustainable disk headroom is available, use authentic Pawn Save As interaction to establish game-specific behavior. Retain the existing history/second-turn uncertainty until independently observed in that run.