code-coordinator 0.5.248__tar.gz → 0.5.250__tar.gz

{code_coordinator-0.5.248 → code_coordinator-0.5.250}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: code-coordinator
-Version: 0.5.248
+Version: 0.5.250
 Summary: Coordinate Claude Code workers across multiple machines and repos.
 Author-email: John Donaghy <johnfdonaghy@gmail.com>
 License-Expression: LicenseRef-FSL-1.1-MIT

{code_coordinator-0.5.248 → code_coordinator-0.5.250}/code_coordinator.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: code-coordinator
-Version: 0.5.248
+Version: 0.5.250
 Summary: Coordinate Claude Code workers across multiple machines and repos.
 Author-email: John Donaghy <johnfdonaghy@gmail.com>
 License-Expression: LicenseRef-FSL-1.1-MIT

{code_coordinator-0.5.248 → code_coordinator-0.5.250}/code_coordinator.egg-info/scm_version.json

@@ -1,7 +1,7 @@
 {
-  "tag": "0.5.248",
+  "tag": "0.5.250",
   "distance": 0,
-  "node": "g698bc8431c005f357304c6d6731da95766129363",
+  "node": "g736f450ea5a6401a7a9b51d286e49fba0c0068e4",
   "dirty": false,
   "branch": "HEAD",
   "node_date": "2026-08-27"

{code_coordinator-0.5.248 → code_coordinator-0.5.250}/coord/reports.py

@@ -1762,7 +1762,12 @@ _USAGE_COLUMN_META: dict[str, ColumnMeta] = {
     "repo": ColumnMeta(id="repo", label="Repo", kind="text", weight=1.5),
     "title": ColumnMeta(id="title", label="Title", kind="text", weight=4.0),
     "legs": ColumnMeta(id="legs", label="Legs", kind="int", align="right", weight=0.6),
-    "tokens_in": ColumnMeta(id="tokens_in", label="Tok In", kind="int", align="right"),
+    # #2825: labelled "Raw In", not "Tok In" — sitting next to a full "Tok
+    # Out", "Tok In" reads as its matched pair, and it is not: this is raw
+    # UNCACHED input, ~0.001% of what a `work` leg actually sends (the other
+    # ~98% is `cache_read`, below). The column `id` stays `tokens_in` so no
+    # historical number or saved sort changes.
+    "tokens_in": ColumnMeta(id="tokens_in", label="Raw In", kind="int", align="right"),
     "tokens_out": ColumnMeta(id="tokens_out", label="Tok Out", kind="int", align="right"),
     # #2786: the column that carries the money — see module comment above.
     "cache_read": ColumnMeta(id="cache_read", label="Cache Rd", kind="int", align="right"),
@@ -2658,12 +2663,24 @@ COMPLETED_COLUMNS = [
     "title",
     "started_at",
     "ended_at",
-    # #2472's four. Appended, never interleaved: the client addresses columns
-    # by INDEX for sorting (`reports_sort_by_column`), so inserting one in the
-    # middle would silently re-point a saved sort at a different column.
+    # #2472's four, plus #2825's `cache_read`. #2472's own five (`repo`
+    # through `ended_at` above) keep the indices they had — the client
+    # addresses columns by INDEX for sorting (`reports_sort_by_column`), so
+    # THAT boundary is never interleaved. `cost_total` shifting from index 8
+    # to 9 here is accepted, not an oversight: it mirrors `usage`'s own
+    # precedent (#2786 inserted `cache_read`/`cache_create`/`turns` ahead of
+    # `cost_captured`/`cost_est`/`cost_total` there too), and grouping token
+    # counts together beats preserving one more index.
     "legs",
     "tokens_in",
     "tokens_out",
+    # #2825: `cache_read` was already on the wire as a row key (see
+    # `_completed_spend`) — ~98% of a `work` leg's input, sitting next to a
+    # `tokens_in` that is ~0.001% of it. Without this column, `Tok Out`
+    # (uncapped) next to `tokens_in` (raw uncached input) reads as "output
+    # dwarfs input", which is backwards by five orders of magnitude. Reuses
+    # `_USAGE_COLUMN_META`'s `cache_read` verbatim.
+    "cache_read",
     "cost_total",
 ]
 
@@ -2677,10 +2694,14 @@ COMPLETED_COLUMNS = [
 # also what `issue-activity` already does, and what `RowIdentity` reads.
 #
 # #2472's spend columns reuse `_USAGE_COLUMN_META`'s labels/kinds verbatim
-# (`Legs`, `Tok In`, `Tok Out`, `Total $`) rather than inventing near-synonyms:
-# an operator reading `completed` and `usage` side by side is looking at the
-# same numbers out of the same rollup, and two spellings of "Tok In" would
-# imply otherwise.
+# (`Legs`, `Raw In`, `Tok Out`, `Cache Rd`, `Total $`) rather than inventing
+# near-synonyms: an operator reading `completed` and `usage` side by side is
+# looking at the same numbers out of the same rollup, and two spellings of
+# the same column would imply otherwise.
+#
+# #2825: `tokens_in` is labelled "Raw In", not "Tok In" — see the comment on
+# `_USAGE_COLUMN_META["tokens_in"]` for why. Applied here too so the two
+# reports keep agreeing.
 #
 # ONE cost column, not `usage`'s `cost_captured`/`cost_est` split. `usage`
 # splits because it is the dedicated cost report, where "some legs have no
@@ -2689,7 +2710,9 @@ COMPLETED_COLUMNS = [
 # would push `title` off a narrow pane. Both halves still ship as extra ROW
 # keys (see `_completed_spend`) — the wire contract allows row keys beyond
 # `columns`, so a client that wants the split has it without another column,
-# and `usage` remains the report to visit for it on screen.
+# and `usage` remains the report to visit for it on screen. `cache_read` is
+# the one exception (#2825): it isn't a split of an existing column, it's the
+# ~98% of input spend that `tokens_in` alone cannot represent.
 COMPLETED_COLUMN_META = [
     ColumnMeta(id="repo", label="Repo", kind="text"),
     ColumnMeta(id="issue", label="Issue", kind="int", align="right"),
@@ -2697,8 +2720,9 @@ COMPLETED_COLUMN_META = [
     ColumnMeta(id="started_at", label="Started", kind="timestamp"),
     ColumnMeta(id="ended_at", label="Ended", kind="timestamp"),
     ColumnMeta(id="legs", label="Legs", kind="int", align="right", weight=0.6),
-    ColumnMeta(id="tokens_in", label="Tok In", kind="int", align="right"),
+    ColumnMeta(id="tokens_in", label="Raw In", kind="int", align="right"),
     ColumnMeta(id="tokens_out", label="Tok Out", kind="int", align="right"),
+    ColumnMeta(id="cache_read", label="Cache Rd", kind="int", align="right"),
     ColumnMeta(id="cost_total", label="Total $", kind="money", align="right"),
 ]
 

{code_coordinator-0.5.248 → code_coordinator-0.5.250}/tests/test_reports.py

@@ -2635,6 +2635,18 @@ class TestUsageSurfacesCacheReadAndTurns:
         # (100 + 100 + 800 + 0) / 10 = 100.
         assert result.rows[0]["tok_per_turn"] == 100
 
+    def test_tokens_in_is_labelled_raw_in_not_tok_in(self) -> None:
+        """#2825: "Tok In" next to a full "Tok Out" reads as a matched pair —
+        it is not one, `tokens_in` is ~0.001% of real input. The column `id`
+        stays `tokens_in` (no historical number or saved sort changes), only
+        the display `label` moves."""
+        from coord.reports import fold_usage
+
+        result = fold_usage(_usage_fixture_rows(), _unbounded_window())
+        meta = {m.id: m for m in result.column_meta}
+        assert meta["tokens_in"].label == "Raw In"
+        assert meta["tokens_out"].label == "Tok Out"
+
     def test_totals_row_also_carries_the_new_columns(self) -> None:
         from coord.reports import fold_usage
 
@@ -4182,7 +4194,10 @@ class TestCompletedFold:
             "repo", "issue", "title", "started_at", "ended_at",
             # #2472's four, APPENDED — the client sorts by column index, so
             # #2454's five must keep the indices they had.
-            "legs", "tokens_in", "tokens_out", "cost_total",
+            # #2825's `cache_read`, appended after `tokens_out` (before
+            # `cost_total`, which shifts by one index — see the module
+            # comment).
+            "legs", "tokens_in", "tokens_out", "cache_read", "cost_total",
         ]
         assert [m.id for m in result.column_meta] == result.columns
 
@@ -4298,6 +4313,33 @@ class TestCompletedSpend:
         # captured cost so its 1M sonnet input tokens estimate to $3.00.
         assert row["cost_total"] == pytest.approx(5.0)
 
+    def test_cache_read_is_a_declared_column_not_just_a_row_key(self) -> None:
+        """#2825: `completed` showed a full `Tok Out` next to a `tokens_in`
+        that is ~0.001% of a leg's real input — the raw uncached count sitting
+        alone next to output read as "output dwarfs input", backwards by five
+        orders of magnitude. `cache_read` was already computed into every row
+        by `_usage_metrics`; this only had to become a declared column.
+        """
+        result = fold_completed(
+            [{"repo_name": "myrepo", "number": 21, "title": "t", "state": "closed"}],
+            [{"repo_name": "myrepo", "issue_number": 21,
+              "dispatched_at": 10.0, "finished_at": 20.0, "model": "sonnet",
+              "input_tokens": 634, "output_tokens": 186_209,
+              "cache_read_tokens": 59_908_777}],
+            [],
+            (C_START, C_END),
+            generated_at=C_END,
+        )
+        assert "cache_read" in result.columns
+        meta = {m.id: m for m in result.column_meta}
+        assert meta["cache_read"].label == "Cache Rd"
+        row = result.rows[0]
+        assert row["tokens_in"] == 634
+        assert row["tokens_out"] == 186_209
+        # The number that actually carries the input volume — orders of
+        # magnitude above `tokens_in`, exactly the #2825 scenario.
+        assert row["cache_read"] == 59_908_777
+
     def test_the_captured_estimated_split_ships_as_extra_row_keys(self) -> None:
         """One `Total $` COLUMN, but no information thrown away — a client
         that wants `usage`'s split can read it off the row."""
@@ -4384,7 +4426,7 @@ class TestCompletedSpend:
             if peer is None:
                 assert row["legs"] == 0, "only a no-leg issue may be absent from usage"
                 continue
-            for key in ("legs", "tokens_in", "tokens_out"):
+            for key in ("legs", "tokens_in", "tokens_out", "cache_read"):
                 assert row[key] == peer[key], f"{row['repo']}#{row['issue']} {key}"
             assert row["cost_total"] == pytest.approx(peer["cost_total"])
             checked += 1
@@ -4646,14 +4688,17 @@ class TestCompletedCatalogue:
         text = result_to_csv(_fold_completed())
         # The header row is the `column_meta` LABELS (the shared serializer's
         # rule), and the values stay raw epochs — not display strings.
-        # #2472's four labels are `usage`'s verbatim, so a spreadsheet built
-        # off one export lines up with the other.
-        assert "Repo,Issue,Title,Started,Ended,Legs,Tok In,Tok Out,Total $" in text
-        # myrepo#9 merged with no assignment: a real 0/0/0/0.0, not four
+        # #2472's four labels (plus #2825's `cache_read`) are `usage`'s
+        # verbatim, so a spreadsheet built off one export lines up with the
+        # other. `tokens_in` is labelled "Raw In", not "Tok In" (#2825) — it
+        # is not `Tok Out`'s pair, it's ~0.001% of a leg's real input.
+        assert "Repo,Issue,Title,Started,Ended,Legs,Raw In,Tok Out,Cache Rd,Total $" in text
+        # myrepo#9 merged with no assignment: a real 0/0/0/0/0.0, not five
         # empty cells.
-        assert 'myrepo,9,"Merged, still open",,400.0,0,0,0,0.0' in text
-        # myrepo#7's retry: 2 legs, $2.00 captured + $3.00 estimated.
-        assert "myrepo,7,Closed one,100.0,250.0,2,1000010,4,5.0" in text
+        assert 'myrepo,9,"Merged, still open",,400.0,0,0,0,0,0.0' in text
+        # myrepo#7's retry: 2 legs, $2.00 captured + $3.00 estimated, no
+        # cache_read_tokens in the fixture so that column reads 0.
+        assert "myrepo,7,Closed one,100.0,250.0,2,1000010,4,0,5.0" in text
 
 
 class TestRowIdentity:

{code_coordinator-0.5.248 → code_coordinator-0.5.250}/tui/src/app/fixtures.rs

@@ -141,7 +141,19 @@ pub fn make_test_app(data: BoardData) -> CoordApp {
         chat_transcript_cache_key: None,
         chat_last_activity: None,
         chat_spinner_throttle: 0,
-        paused_machines: read_paused_machines(),
+        // #2028: never call the real `read_paused_machines()` from a test
+        // fixture — it reads the LOCAL `~/.coord/paused_machines.json`
+        // unconditionally (mirrors `workspace` above: no real I/O). Before
+        // this, any host whose real pause file was non-empty (e.g. two
+        // long-destroyed ephemeral workers left paused on dellserver since
+        // 2026-08-04) failed `poll_paused_machines_replaces_snapshot_when_ready`
+        // and `..._reverts_an_operator_pause_..` (`app/tests.rs`) purely from
+        // ambient machine state — both tests assert `paused_machines.is_empty()`
+        // immediately after construction. Empty here matches what every
+        // caller of this fixture already expects; a test that wants a
+        // pre-populated set assigns `app.paused_machines` directly (see the
+        // tests further down this file).
+        paused_machines: std::collections::HashSet::new(),
         quiet_paused_machines: std::collections::HashSet::new(),
         cordoned_machines: std::collections::HashSet::new(),
         quiet_hours_windows: std::collections::HashMap::new(),

{code_coordinator-0.5.248 → code_coordinator-0.5.250}/tui/src/app/mod.rs

@@ -9756,12 +9756,15 @@ fn today_yyyy_mm_dd() -> String {
 ///
 /// #1563: on a thin client (board service configured) this file is no
 /// longer authoritative — `coord pause` on such a client now writes
-/// straight to the daemon instead. `CoordApp::new`/`make_test_app` still
-/// call this directly for the *initial* synchronous `paused_machines`
-/// snapshot (mirroring `live_tmux_sessions`/`fleet_terminals`/
-/// `drive_sessions` further up this file, each seeded from a cheap local
-/// read before their background remote sweep is armed), but every
-/// *periodic* refresh after that goes through
+/// straight to the daemon instead. `CoordApp::new` still calls this
+/// directly for the *initial* synchronous `paused_machines` snapshot
+/// (mirroring `live_tmux_sessions`/`fleet_terminals`/`drive_sessions`
+/// further up this file, each seeded from a cheap local read before their
+/// background remote sweep is armed). `app::fixtures::make_test_app`
+/// deliberately does NOT call this (#2028) — a test fixture must never
+/// read ambient machine state, so it seeds `paused_machines` empty instead
+/// and lets individual tests assign it directly. Every *periodic* refresh
+/// after startup goes through
 /// [`data::fetch_paused_machines`] / [`data::spawn_paused_machines_fetch`]
 /// instead, which route to the daemon's `GET /pause` when a board service
 /// is configured and fall back to this function otherwise — so a