@ndmspc/ndmspc-ui 1.4.4 → 1.4.6

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/README.md CHANGED
@@ -360,16 +360,40 @@ https://gitlab.com/api/v4/projects/ndmspc%2Fndmspc-ui/packages/generic/ndmspc-ui
360
360
 
361
361
  (use `permalink/latest` in place of `<tag>` to always follow the newest release)
362
362
 
363
- For a rolling development pointer, every build of `main` is published by the
363
+ Three names describe what is downloadable:
364
+
365
+ | Name | What it serves | Moved by |
366
+ | ------------- | ----------------------------------------------- | ----------------------------------------- |
367
+ | `v1.4.5` | that tag's page, never overwritten | the tag's own pipeline |
368
+ | `v1.4-stable` | the newest `v1.4.x` page, pre-releases included | every `v1.4.*` tag (`release-stable`) |
369
+ | `next` | the newest `develop` build | every `develop` pipeline (`release-next`) |
370
+
371
+ `v1.4-stable` is what lets ndmspc release a 1.4 patch without ndmspc-ui being
372
+ tagged again: ndmspc asks for `v1.4-stable` first, falls back to its exact tag
373
+ page (`v1.4.5`) when the series slot does not exist, and only then to `next`.
374
+ The slot is over-written in place (the previous package version is deleted
375
+ first), so downloading it always yields the latest page of the series:
376
+
377
+ ```
378
+ https://gitlab.com/api/v4/projects/ndmspc%2Fndmspc-ui/packages/generic/ndmspc-ui-page/v1.4-stable/ndmspc-ui-page.tar.gz
379
+ ```
380
+
381
+ The page keeps the version of the tag it was built from (a `v1.4-stable` download
382
+ still shows `UI v1.4.5` in the user menu). Two consequences: rebuilding an older
383
+ series tag after a newer one moves the slot backwards (the last pipeline wins),
384
+ and `v1.5-stable` only appears with the first 1.5 tag. Pin an exact tag when a
385
+ byte-identical page matters.
386
+
387
+ For a rolling development pointer, every build of `develop` is published by the
364
388
  `release-next` job to a single over-writable `next` slot in the Generic Package
365
- Registry, so the URL always serves the latest `main` page:
389
+ Registry, so the URL always serves the latest `develop` page:
366
390
 
367
391
  ```
368
392
  https://gitlab.com/api/v4/projects/ndmspc%2Fndmspc-ui/packages/generic/ndmspc-ui-page/next/ndmspc-ui-page.tar.gz
369
393
  ```
370
394
 
371
395
  Point `_ui_url` above at that URL instead of the release downloads path to track
372
- `main` (substitute the numeric project id for `ndmspc%2Fndmspc-ui` if you prefer).
396
+ `develop` (substitute the numeric project id for `ndmspc%2Fndmspc-ui` if you prefer).
373
397
 
374
398
  The archive extracts to a ready-to-serve web root (`index.html` + `assets/`).
375
399
  A minimal CMake consumption looks like:
package/dist/index.d.ts CHANGED
@@ -2013,8 +2013,27 @@ export declare class NdmspcRoomClient {
2013
2013
  constructor(options: NdmspcRoomClientOptions);
2014
2014
  /** The MCP endpoint in use (for display). */
2015
2015
  get url(): string;
2016
- /** Lists the rooms the router is tracking, with the idle TTL. */
2017
- list(): Promise<NdmspcRoomList>;
2016
+ /**
2017
+ * Lists the rooms the router is tracking, with the idle TTL.
2018
+ *
2019
+ * A plain GET on the router's own action, not an MCP `tools/call`: the list is what a view polls
2020
+ * while a room starts, and a GET has no session handshake that can get stuck (and is safe to
2021
+ * retry). The handler answers the same envelope the MCP call carries.
2022
+ */
2023
+ list(signal?: AbortSignal): Promise<NdmspcRoomList>;
2024
+ /**
2025
+ * Calls one of the router's own room actions and returns its payload.
2026
+ *
2027
+ * These are plain `/api/ndmspc/room/...` endpoints, not MCP `tools/call`: they answer the same
2028
+ * handler envelope, the server injects the verified identity, and there is no MCP session
2029
+ * handshake to get stuck on a connection that is gone (the failure an update used to hit).
2030
+ */
2031
+ private requestPayload;
2032
+ /**
2033
+ * Drops the MCP handshake so the next call starts a new one. A handshake pinned to a connection
2034
+ * that is gone (the router rolled, a socket died) must not be reused by every later call.
2035
+ */
2036
+ reset(): void;
2018
2037
  /**
2019
2038
  * The router's aggregate usage statistics for a window.
2020
2039
  *
@@ -2360,6 +2379,12 @@ export declare interface NdmspcRoomInfo {
2360
2379
  code: string;
2361
2380
  /** Epoch seconds the creation began. */
2362
2381
  startedAt: number;
2382
+ /**
2383
+ * Seconds the router took from the creation start to the room being ready, as the router measured
2384
+ * it. Absent while it is still starting, for a room that failed or is pending, and for one the
2385
+ * router adopted without its start (a router restart forgets it).
2386
+ */
2387
+ startedIn?: number;
2363
2388
  /** The tokens that open the room; absent for a room that enforces nothing. */
2364
2389
  access?: NdmspcRoomAccess;
2365
2390
  /** What the room's Service declares it may use; absent for one that declares nothing. */
@@ -3005,10 +3030,11 @@ export declare class NdmspcRoomService {
3005
3030
  * How long to wait before reading the list again.
3006
3031
  *
3007
3032
  * The push, when it is live, is what makes a change appear - so a watching view only needs the net
3008
- * under it, and takes the slow floor whatever is happening. Without the push, the configured interval
3009
- * is used while a room is still being created (a create whose row is `preparing`, or one waiting for
3010
- * cluster resources), and a slower one otherwise: the list rarely changes by itself, and the router
3011
- * serves one request at a time, so polling it hard spends the room's own requests for nothing.
3033
+ * under it. A room still being created is the exception: the push is one publish tick behind, and a
3034
+ * create is the one moment its row matters, so the configured interval carries it even while
3035
+ * watching. Without the push the same interval covers a room waiting for cluster resources; the
3036
+ * list rarely changes by itself otherwise, and polling it hard spends the router's requests for
3037
+ * nothing.
3012
3038
  */
3013
3039
  private pollIntervalMs;
3014
3040
  private clearTimer;