@torrent-tv/proxy 2.89.5 → 2.89.6
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +11 -0
- package/package.json +1 -1
- package/test/auto-quality-step.test.js +762 -753
- package/test/decode-cost.test.js +475 -462
- package/test/format-keyed-outputs.test.js +78 -78
- package/test/helpers/manager.js +62 -57
- package/test/quality-variants.test.js +1270 -1257
- package/test/segment-judged-by-its-own-header.test.js +222 -221
- package/test/segment-serve-wiring.test.js +9 -3
- package/test/serving-existing-output.test.js +111 -105
- package/test/two-viewers-one-picture.test.js +508 -504
- package/test/viewer-requests-admission.test.js +92 -89
- package/test/viewing-generations.test.js +390 -390
package/CHANGELOG.md
CHANGED
|
@@ -1,3 +1,14 @@
|
|
|
1
|
+
## 2.89.6
|
|
2
|
+
|
|
3
|
+
- **Chore**: Seventeen checks that failed on an unmodified checkout are brought in line with the system they check; no service code changes. Each followed a change of the system that its test did not follow:
|
|
4
|
+
- the encoder's measured modes are read through the encoder in use since 2.87.0 and cannot be assigned afterwards, so the checks state them when the manager is built, and a second host is built where a check compares two (`auto-quality-step`, `quality-variants`, `two-viewers-one-picture`, `decode-cost`). The speed a file's supply demands is the file's, asked of the host's load, not a field on a session;
|
|
5
|
+
- a viewer is placed on an output only when its encoding cost is measured (2.87.0): fixtures that open or prepare an output state what one encoder costs, or what the output's own encoder was seen doing (`quality-variants`, `serving-existing-output`);
|
|
6
|
+
- checks that keep the plan out of the path under test do so before a viewer is registered, because registering one now asks for a plan (`quality-variants`);
|
|
7
|
+
- an output is named by its format, so a request clamped onto a picture already made is answered by that output; the fake opening honours it (`quality-variants`);
|
|
8
|
+
- the encoder in use is `encoders.current` (`format-keyed-outputs`), an output's header is the segment store's (`segment-judged-by-its-own-header`), the opening path reports to the viewer (`viewer-requests-admission`), and serving a session file lives in `services/server/transcode-session-files.js` (`viewing-generations`);
|
|
9
|
+
- the self-contained piece of `segment-serve-wiring` carries a whole version 0 `elst` entry (ISO/IEC 14496-12: twelve bytes, with the media rate), and the store keeps the header production keeps, prepared by `prepareSharedInit`. Without either the piece's position was read as zero on both sides;
|
|
10
|
+
- managers built by `test/helpers/manager.js` keep host measurements beside their own store instead of beside the installed proxy. Two checks passed only on a machine whose earlier runs had left `host-timings.json` there; they now state the first-segment time they rely on (8.4 s, field 2026-08-31).
|
|
11
|
+
|
|
1
12
|
## 2.89.5
|
|
2
13
|
|
|
3
14
|
- **Fix**: A segment store's watch on an output's directory ends when that directory is removed by anything other than `drop`. On Windows the watch emits no `error` for it, and reports `rename` for the removed directory without end — measured 2026-10-02 on Node 24.2, 232 618 events in eight seconds — which kept the process alive; `test/segments-are-shared.test.js` and `test/segment-store-eviction.test.js` passed every check and never exited. A directory made again under the same name is watched again: the dead watch used to stay registered and stop a new one being made, so a wait for a piece there ended only on its deadline.
|