@yuneta/gobj-ui 7.25.4 → 7.25.6

package/README.md

@@ -790,6 +790,49 @@ writes in flight when the transport closes are answered refused on the
 transport edge (`form_writes_in_flight.js`), so the form comes back on what
 was typed instead of staying busy for ever.
 
+**Where the transport edge comes from** (7.25.5). Until 7.25.4 only from the
+host, as `EV_TRANSPORT_STATE` — and two hosts (wattyzer, yunovatios) never
+forwarded it, so a websocket drop during a Save left Save and Cancel busy for
+ever. The view now also hears it from its SHELL: `yui_shell_set_connection_state()`
+— which every single-backend app already calls for the toolbar dot — publishes
+`EV_CONNECTION_STATE {connected}` on each edge, and `C_YUI_TREEDB_TOPICS`
+subscribes to it in `mt_start`. Nothing to add in such a host. A host with
+one transport per view (gui_treedb's `C_TREEDB_VIEW`, gui_agent's
+`C_AGENT_TREEDB_VIEW`) keeps forwarding `EV_TRANSPORT_STATE`; both paths can
+fire for one edge, and the view answers each write once. The view never
+subscribes to the transport itself: a subscription to a `C_IEVENT_CLI` is
+forwarded to the backend.
+
+`EV_CONNECTION_STATE` reaches only the subscribers that DECLARE it: an app
+that subscribed to every event of its shell with the `subscriber` attr does
+not get it (the shell's `mt_publication_pre_filter`), so adding it broke no
+app.
+
+```js
+// The app, from its transport handlers -- as it already did for the dot:
+yui_shell_set_connection_state(shell, false);     // EV_ON_CLOSE
+// -> every C_YUI_TREEDB_TOPICS under that shell answers its forms'
+//    writes in flight EV_WRITE_REFUSED, and they come back on what was typed.
+```
+
+A write is known by its topic AND its serial (`<topic>^<form_write>`): each
+topic's form counts its serials from 1, and keyed by the serial alone two
+forms saving at once shared one entry.
+
+**The pkey2 of an update goes back as the record had it** (7.25.5). It names
+the instance the write is for, and a `time` pkey2 that crossed the form's
+`datetime-local` came back without its seconds — another instance, or none.
+`instance_keys_of()` (`treedb_write_plan.js`) keeps it from the record the
+form was opened on.
+
+**Known limit — the change of hour** (`form_time_value.js`). A
+`datetime-local` holds a local wall time and no offset. The night the clocks
+go back, one hour happens twice: an epoch in its second pass is shown with the
+wall time of the first and read back one hour earlier, so saving that record —
+any field of it — moves a WRITABLE time column holding it by -3600 s. The
+hour skipped in spring holds no epoch, so nothing read there comes back wrong.
+Read-only time columns never go back, and a time pkey2 goes back as it was.
+
 ```js
 // A host of C_YUI_TREEDB_TOPIC_WITH_FORM that answers the form's writes
 gobj_create_pure_child(name, "C_YUI_TREEDB_TOPIC_WITH_FORM",
@@ -802,7 +845,24 @@ gobj_send_event(topic_form, ok ? "EV_WRITE_DONE" : "EV_WRITE_REFUSED",
 **A delete is resolved by IDENTITY, never by position** (7.23.193). The row's
 `id` crosses the confirmation dialog; a Tabulator position is regenerated by
 every row the node events add or remove while the dialog is open, and it used
-to delete the row that sat there when the person answered.
+to delete the row that sat there when the person answered. Rows (table) or
+cards (`C_G6_NODES_TREE`) that another writer deleted while the question was
+open are not deleted and the person is told (7.25.4); the rest go on.
+
+Consumer i18n keys of the form's writes and deletes — define them in the app,
+or they render as the key itself in every language:
+
+| key | shown when |
+|---|---|
+| `no session` | a Save asked with no backend session: the write is not sent (7.25.2) |
+| `some records were gone before the delete` | a confirmed delete named rows or cards that no longer exist (7.25.4). The text must not promise a refresh: nothing is reloaded, and in the graph there is no table |
+
+```js
+// en.js / es.js of the app
+"no session": "No session",
+"some records were gone before the delete":
+    "Some records no longer existed when the delete was confirmed; they were not deleted",
+```
 
 ### Reading a topic table: filters, columns, CSV
 
@@ -2030,6 +2090,17 @@ host sends it as `EV_DRAFTS`; `drafts_of_saved_answer()` reads the unnamed
 answer into the shape the event takes. What the session wrote stays marked;
 what the host said is applied again after every reload of the model.
 
+**`EV_DRAFTS` REPLACES what the host said before** (7.25.5). It only added,
+and a topic the host had named before a Save stayed a draft after the Save
+that published it ("unsaved schema changes" again). The editor keeps the
+host's marks apart from its session's writes (`host_draft_ids()`) and swaps
+them whole on each `EV_DRAFTS`, so a host sends the COMPLETE set every time —
+an empty `{drafts: {}}` included, which is how "nothing is a draft any more"
+is said. What `draft_changed` measures is the node's: the draft against the
+SAVED schema on a node newer than yunetas 7.25.3, against the file in use on
+a 7.25.3 node (where a topic saved and not yet applied is still a draft until
+the Apply).
+
 ```js
 // The host, when its saved-schema answers land (rows = the unnamed answer's data)
 import {drafts_of_saved_answer} from "@yuneta/gobj-ui/src/host_drafts.js";
@@ -2083,6 +2154,20 @@ else.
 - In: `EV_SHOW` (`{subpath}` — the tail it owns is `<treedb>[/<topic>]` or
   `<treedb>/diagram`), `EV_HIDE`, `EV_TRANSPORT_STATE`, `EV_REFRESH`,
   `EV_LANGUAGE_CHANGED`, `EV_MT_COMMAND_ANSWER`.
+- A **drop** of the transport — `EV_TRANSPORT_STATE {connected: false}`, or a
+  failed answer that arrives while the transport is out of `ST_SESSION` — ends
+  the load or the write it cut, and the model is asked again on the next
+  `{connected: true}`. A write cut this way is shown as
+  `the connection dropped during the write` (an i18n key the consumer's
+  locales carry). Each request carries its round in `__md_command__`
+  (`round` / `write`), so an answer of what the drop cut is ignored with a
+  warning.
+
+  ```js
+  // the host forwards its session edges; nothing else is needed
+  gobj_send_event(editor, "EV_TRANSPORT_STATE", {connected: false}, host);
+  gobj_send_event(editor, "EV_TRANSPORT_STATE", {connected: true}, host);
+  ```
 - Out: `EV_POSITION_CHANGED` (`{subpath}` — the host writes the url; this view
   navigates nothing itself), `EV_RECORD_WRITTEN` (whoever owns the Apply needs
   to know the yuno has not re-read its schema yet), `EV_SCHEMA_CHECKED`