[Nawid3333]

Two small fixes the refactor review of 1.3.82.160 found (no refactor of working code).

1. A failure of the last source's player no longer touches the next source

setSource counts the next source at once, but the last source's player stays this.player until resetPlayer tears it down. What it reported in between was taken for the new source's:

• its hand-over to Firefox's own player (PLAY_DIRECTLY) played the old URL over the source the user had just picked;
• its ERROR / NEED_KEY showed the load error over the source on its way in;
• a refused link of the old source was asked for anew and rebuilt over it (renewSourceLink, also when the answer came after the next source was asked for).

tryNextSource and recoverPlayer already checked fallbacks.request === sourceRequests; every failure route does now (FastStreamClient.isCurrentAttempt, plus the check in renewSourceLink before asking and after the answer).

2. One list says who may send the background each message

runtime.onMessage answers 33 message types. Five checked fromOwnPage(sender) where they were answered; others did not, though only FastStream's own pages send them: LOAD_OPTIONS, MPV_TEST (its answer includes the mpv path), CLEAR_SOURCES, REQUEST_PLAYLIST_NAVIGATION. Web pages cannot message the background, and FastStream's content script never sends these, so this is hardening (a compromised content process), not a page exploit.

MessageSenders now classifies every answered type as 'own' (FastStream's pages only) or 'any' (content scripts too), read before any handler, replacing the five inline checks. backgroundMessages.test.mjs keeps it complete: every answered type is listed, no listed type is stale, and no type a content script sends is 'own'. The background test harness sends an 'own' message from the player page unless a test names another sender.

Tests

• New: client race tests (each fails on main; the late-link case also fails with only the first check), backgroundMessages.test.mjs, a CLEAR_SOURCES-from-a-page case; removing the gate fails 4 tests.
• Local: lint, tsc, 3368 unit tests; all 25 extension and 10 classic e2e specs except the mpv ones.
• reload-keeps-buffer failed in the full parallel e2e run (one segment that was still downloading at the reload was asked for again), and passes alone twice on this branch and twice on main: a timing race of the spec under load, not this change.
• glm reviewed both halves: no finding.

🤖 Generated with Claude Code