Bound the idle wait so a busy page cannot hang select()

[nathanfallet]

DefaultTab.select() / find() / findAll() all open with wait(), which lands in the t == null branch of DefaultConnection.wait() — the one commented "Wait indefinitely". Its only exit is 100 ms without a single CDP event.

On a page that streams continuously — polling, server-sent events, ads, a refreshing interstitial — that quiet window never comes, so wait() never returns and select() blocks before it ever evaluates its own timeout. The timeout parameter is inert in that state, timeout = 0 included.

Changes

wait() takes an idleTimeout, capping how long it watches for the connection to settle. Waiting is best-effort — a page that never settles is not an error — so past the bound it stops watching and returns.

t stays a floor and keeps its meaning. Splitting the two makes the signature say what the body used to hide: one number served as both cap and floor, which is why the two branches existed. They are now a single path.

⚠️ Note for anyone tempted by the smaller change: giving t a non-null default does not work. The t != null branch guarantees a wait of at least t, and select() calls wait() once per poll — every selector would cost that much. It would trade a hang for a slowdown.

Idleness is now read, not guessed. The receive loop timestamps the last frame it saw, and isIdle() compares against it. The previous check re-subscribed to events on every poll and only observed the 100 ms window it had just opened, so a connection quiet for a while still looked busy, and a busy one could look quiet.

Tests

A clock seam mirrors the existing createTransport() seam, so the tests decide "traffic kept arriving" and "nothing arrived" off the scheduler's virtual clock rather than off how fast the machine runs.

• a page that never goes quiet — wait() returns, and only at the bound
• a silent connection — idle at once, no polling window spent rediscovering it
• wait(t) on an already-idle connection — the floor is still honoured

All three fail on main. Mutation-checked: removing the bound, removing the floor, never writing the timestamp, treating "nothing received" as busy, and dropping the idle threshold each turn one red.

Full :core:jvmTest + :opentelemetry:jvmTest green (184 tests). :cdp:detekt and kotlinStoreYarnLock fail identically on main and are untouched; :core detekt violations are unchanged at 142.

[codecov]

Codecov Report

❌ Patch coverage is 86.66667% with 2 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
...n/dev/kdriver/core/connection/DefaultConnection.kt	88.88%	0 Missing and 1 partial ⚠️
...tlin/dev/kdriver/opentelemetry/OpenTelemetryTab.kt	0.00%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!