timothy
fba5233caf
168 records -> docs/decisions/records/<area>/<topic>.md (163 active, 23 dirs) and docs/decisions/archive/<area>/<topic>.md (5 archived). The filename IS the key, so one-active-record-per-key becomes a filesystem property rather than a validator check, and supersession becomes a `git mv`. WHY: the monolith was a concurrency problem before an aesthetic one. A 3,900-line append target made parallel sessions collide -- PR #605 and PR #614 both hit append-vs-append conflicts during routine rebases, and hand-resolving those inside the corpus is exactly the operation the rationale-rewrite guard exists to police. HOW IT IS VERIFIED: a ~170-file diff cannot be meaningfully read, so correctness does not rest on reading it. The parser was taught BOTH formats first, so the body-diff guard parses the old form at the merge-base and the new form at head -- the migration validates itself, no bypass. The proof is a field-level equivalence harness: 168 records before and after, zero lost, zero gained, zero field mismatches, zero rationale bodies differing. Reviewers should scrutinise the harness; it is the actual evidence. What measuring caught that reading would not have: - ~500 lines sit OUTSIDE any record -- decisions.md's lifecycle schema and each topic file's preamble, mostly the only copy. Source files are kept and stripped, never deleted. They also cannot be filed per-area: topic files hold several areas and 4 of 23 areas span several files. - Archive discovery was a non-recursive glob; after the split it found ZERO archived records, surfacing as four bogus "supersedes points to unknown key" errors rather than an obvious failure. - ~32 live docs point into the corpus BY DATE, which the split dangles. Each stripped file now ends with a generated "Records formerly in this file" index, which also rescues the identical breadcrumbs in old issue comments. - decisions.md's "In this file:" list was 97 same-file anchor bullets that the split makes WRONG, not merely stale. Dropped; the generated index replaces them with links that resolve. The equivalence harness now runs against a checked-in FIXTURE, not the live corpus. The earlier version migrated the real tree, which made it a one-shot: the moment the migration landed there was nothing left to move and the tests failed for reasons unrelated to the code. A fixture keeps them testing the SCRIPT rather than the repo's current state. Keys preserved verbatim, warts included: `sched` (12) and `scheduling` (1) remain two directories for one concept. Renaming a key is not a move -- it changes identity, breaks the equivalence proof, and invalidates MemPalace's per-key drawers. Taxonomy normalisation is separate work. refs #610
3.7 KiB
key, title, status, since, supersedes, superseded-by, rule, signals, mechanics
| key | title | status | since | supersedes | superseded-by | rule | signals | mechanics |
|---|---|---|---|---|---|---|---|---|
| ffmpeg.work-ahead-slot-release-never-negative | 2026-07-21 — `WorkAheadSlots.Release()` clamps before decrementing and reports unbalance in-band (#539) | active | 2026-07-21 | none | none | `Release()` reads the count and compare-exchanges `current - 1` only when `current > 0`; a release against an empty pool records an unbalanced release and returns `false` **without ever writing a negative value**. It never decrements first and clamps afterward. The single caller (`HlsSessionWorker.Transcode`'s `finally`) logs a warning on the `false` return. | work-ahead slot release, unbalanced release, negative slot count, over-admit at limit 1, phantom work-ahead room, UnbalancedReleases counter · paths: `ErsatzTV.Application/Streaming/WorkAheadSlots.cs`, `HlsSessionWorker.Transcode`, `ErsatzTV.Core.Tests/Streaming/WorkAheadSlotsTests.cs` · issues: #539, #536, #529, #231, #250 | `bool Release()` CAS loop guarded by `current <= 0`; call site `if (ownsWorkAheadSlot && !_workAheadSlots.Release()) _logger.LogWarning(...)` |
Why never-negative, given the release is already unreachable-when-unbalanced. Today HlsSessionWorker has exactly one release site, guarded by ownsWorkAheadSlot, so an unbalanced release cannot happen — these were three Low findings from the #536 re-review, filed against the day someone adds a second release site. The pre-#539 shape decremented first (0 → −1) and clamped afterward, leaving a reachable interleaving where a concurrent TryAcquire(limit) reads the −1, sees phantom room, and admits a holder the budget doesn't have (a second acquirer then reads 0 and admits another) — two unthrottled transcodes at limit 1, re-opening the #529 QSV pool exhaustion. Because TryAcquire is the sole, CAS-guarded increment path, a count proven never-negative is exactly what forecloses that over-admit. Clamping before the decrement also records the breakage synchronously on the offending thread, instead of the decrement-first shape where the blame lands on a later, innocent release.
Release() returns bool so the breach is visible in the logs, not only to tests. #536's motivation was "a silently inflated budget with nothing in the logs to find it by," yet UnbalancedReleases was reachable only from a debugger or a test. WorkAheadSlots has no logger by design (it is a leaf primitive), so rather than plumb one in, Release() returns false on an unbalanced release and the caller — which already holds an ILogger — logs the warning. This is the one in-band signal that the ownership contract broke.
UnbalancedReleases can under-count, and that is documented rather than fixed. It only increments when a release finds the pool already empty. An over-release while the count is positive — e.g. one cancelling out a coexisting leak — decrements a real-looking slot and is never recorded, so the two bugs hide each other. There are no false positives (non-zero still means the contract broke), but zero does not prove correctness. Exact accounting would need per-owner tokens, which the #536 "ownership is a discipline, not a token" decision deliberately avoids; the docstring now states the limitation instead.
Negative control (inherited from #231/#250). A dedicated test hammers unbalanced releases on an empty pool while reader threads sample the count; none may ever observe a value below zero. Reinstating the pre-#539 decrement-first body makes it fail (sawNegative > 0 — the readers catch the transient −1); verified. As with the #536 tests, break the primitive by reverting the real body, not if (true) (CS0219 under warnings-as-errors leaves --no-build running a stale, still-fixed dll).