Burndown: ~107 pre-existing integration test failures surfaced by #618 fix

[jsavin]

Pre-existing integration test failures surfaced by #618

PR #619 fixes issue #618 (protocol script/eval was reporting success: true on compile errors due to langruntraperror's trap-and-continue behavior + a missed fllangerror signal). Fixing that surfaces ~107 integration tests that were silently green on develop because the eval-trap masked compile errors AND silent-success outcomes.

These are pre-existing bugs, not regressions from #618. The eval-trap was hiding them. Filing this as the umbrella issue to triage them.

Categories (counts after #618 + #586 + yaml fixture cleanup)

Category	Count	Sample failure	Likely root cause class
tcp	52	tcp.countConnections - track stream lifecycle returns "false" because during=2 (counts listener + stream) where test expects during=1 (stream only)	Verb-vs-test semantic mismatch. Need to decide: does countConnections count listeners?
menu	16	menu.setCommandKey("S") fails because headless overload signature is (@leaf, 'X'), not (key)	Test fixtures written for Mac-UI cursor-based API; need rewrite for headless overload
HTTP	7	webserver/inetd request handling	needs investigation
html	5	sanitization / framework	needs investigation
inetd	4	listener-related	needs investigation
webserver	3	response-related	needs investigation
cursor	3	menu cursor navigation	likely same as menu
persistence	2	db.save/db.load of menu data	menu serialization round-trip
workflow	2	end-to-end menu	composite
misc	13	mainresponder, html.entityencode, deep tests	scattered

Recommended triage approach

Per-category PR or worktree. Don't try to fix them all in one chain — each category likely has its own root cause family.

Until then

Integration suite should NOT block #619 / #617 from landing — the failures pre-existed and the eval-trap is a worse hazard than visible red tests. Two options to keep CI green:

1. Mark the 107 tests as skip: with reason "blocked on #N for category-X cleanup" (where N is a per-category sub-issue of this). Same pattern Frontier already uses for 188 prior skipped tests.
2. Accept temporary red — land fix(lang+parser): honor fllangerror (#618) + bare-LF parser fix (#586) + test cleanup #619 + fix(parser): bare-LF support in // comments and whitespace (#586) — DRAFT, do not merge #617, treat the 107 as the new floor, work them down PR by PR.

Recommended: option 1, because it makes the debt visible AND auditable.

Source of analysis

• PR fix(lang+parser): honor fllangerror (#618) + bare-LF parser fix (#586) + test cleanup #619 — the Integration tests pass on compile errors due to repl_eval_with_variables_value error trap #618 fix
• The agent's analysis in issue body of Integration tests pass on compile errors due to repl_eval_with_variables_value error trap #618 surfaced the trap mechanism
• Categorization done by running full integration suite at PR fix(lang+parser): honor fllangerror (#618) + bare-LF parser fix (#586) + test cleanup #619 HEAD + parser cherry-pick

Labels

tech-debt, P1, scope/large, testing