taskops-cli 0.5.3__tar.gz → 0.5.5__tar.gz

{taskops_cli-0.5.3 → taskops_cli-0.5.5}/CHANGELOG.md

@@ -3,28 +3,64 @@
 The source of truth for release notes — GitHub Releases are extracted from
 here, never written twice.
 
-## 0.5.3 — the Files-changed pane renders for a landed card
+## 0.5.5 — the Worktrees page renders too (it was never the drawer)
+
+The failure reported since 0.5.0 was always on the WORKTREES page, and three
+chapters fixed the card drawer instead — the drawer was verified, declared
+working, and the page nobody had rendered kept asking
+`compare/ms%2F<chapter>...tk-<id>`. Both branches of a landed card are pruned
+when its chapter lands, so on every landed card the page drew a refusal over
+work sitting in the trunk. Verifying a neighbour is not verifying.
+
+- **`pages/WorktreeDiff.tsx` takes its range from `links.tsx::cardRange`** — the
+  same rule the drawer uses, never a second one that can drift: the card's
+  recorded commits, base = the oldest commit's PARENT. A tree with nothing
+  recorded keeps the two branch tips, which is right for a live tree whose work
+  is not committed yet and where the branch is the only handle there is.
+- **Pinned on the rendered PAGE** (`ui/smoke/sections/worktree-page-range.tsx`),
+  read off its `data-range`, with the call-site mutation measured red — the
+  previous section pinned the rule one layer below the call site and a call-site
+  mutation was green while production stayed broken.
+- **The history that was missing is on the host now**: 91 card branches and 19
+  chapter branches, pushed at the tips the board recorded, plus two cards whose
+  oldest commit was not an ancestor of their newest (rebased in their day) and
+  so needed its own ref. Audited across every card on the board: 87 of 89 render
+  a real diff on both surfaces, and the two that do not have no commit recorded
+  at all — there is nothing to diff, and their refusal is literally true.
+
+## 0.5.4 — the Files-changed pane renders for a landed card
+
+**0.5.3 is withdrawn in all but name.** It carries this release's source fix but
+the PREVIOUS release's committed UI bundle, so a hosted page served by it still
+asks for branch names and the pane stays empty. The cause is worth writing down:
+the version bump, the changelog and the tag were made while the shell was still
+inside a card's worktree after a conflict resolution, so `v0.5.3` points at that
+card's branch — which holds `links.tsx::cardRange` but not the chapter-closing
+rebuild of `src/taskops/ui/app.js`. Every gate passed there, because that tree
+was internally consistent; what was missing was the rest of the chapter. A
+release is cut from the trunk, and the check that catches this is comparing the
+bundle a host SERVES against the one the tag committed — which is now a deploy
+step, not a habit.
 
 - **The pane asked for two branch names, and a landed card has neither.**
   0.5.2 made the HEAD durable — the card's newest recorded sha instead of its
-  branch — and stopped there. The BASE stayed the chapter's branch name, and
-  for every card that landed before this host became the board's remote that
-  branch was pruned on the forge, so the compare refused and the pane was
-  empty while the commits sat in the trunk. Reported live on tk-bffa26.
+  branch — and stopped there. The BASE stayed the chapter's branch name, and for
+  every card that landed before this host became the board's remote that branch
+  was pruned on the forge, so the compare refused and the pane was empty while
+  the commits sat in the trunk. Reported live on tk-bffa26 and tk-c37061.
 - **The range now comes from the card's own commits, both ends**
   (`ui/src/links.tsx::cardRange`): one recorded commit goes through the commit
   door, which already diffs a commit against its first parent; several become
   `compare/<oldest>^...<newest>`. The base is the oldest commit's PARENT, not
-  the oldest commit — `git diff` is exclusive of its base, so `<oldest>...<newest>`
-  answers 200 with that commit's own changes silently missing, which is a pane
-  that looks complete and is not. No server change was needed: `<sha>^` already
-  passed the ref shape and resolves through `^{commit}`.
-- **Pinned where it broke.** The pane renders the route it asked for, and a
-  smoke section reads that route off the rendered component — the previous
-  attempt was pinned one layer below the call site, so a call-site mutation
-  went green while the pane stayed broken. `smoke.sh` now derives the same
-  route from the LIVE card payload and asserts the first commit's files are in
-  the answer.
+  the oldest commit — `git diff` is exclusive of its base, so
+  `<oldest>...<newest>` answers 200 with that commit's own changes silently
+  missing, which is a pane that looks complete and is not. No server change was
+  needed: `<sha>^` already passed the ref shape and resolves through `^{commit}`.
+- **Pinned where it broke.** The pane renders the route it asked for and a smoke
+  section reads that route off the rendered component — the previous attempt was
+  pinned one layer below the call site, so a call-site mutation went green while
+  the pane stayed broken. `smoke.sh` derives the same route from the LIVE card
+  payload and asserts the first commit's files are in the answer.
 - **"so them was never pushed here"** — the host's missing-ref refusal had a
   plural substitution in a subject slot. The number now runs through the whole
   sentence, and both forms are pinned as bytes.

{taskops_cli-0.5.3 → taskops_cli-0.5.5}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: taskops-cli
-Version: 0.5.3
+Version: 0.5.5
 Summary: A shared work board for teams of coding agents, with a human who decides.
 Project-URL: Repository, https://github.com/bernatch22/taskops
 License-Expression: MIT

{taskops_cli-0.5.3 → taskops_cli-0.5.5}/smoke.sh

@@ -15,7 +15,7 @@
 # (`tests/test_topology.py`), against a host it created.
 #
 #   sh smoke.sh                                  # the author's host, defaults below
-#   sh smoke.sh <host> <public-board> <private-board> <card-branch> <landed-branch>
+#   sh smoke.sh <host> <public-board> <private-board> <card-branch> <landed-card> <reported-card>
 #
 # Every assertion is here because it FAILED once, in production, against a real
 # reader — the comment above each one says which report it comes from. Nothing
@@ -37,6 +37,10 @@ BRANCH=${4:-tk-32d2ba}
 # the fault that turned the pull mirror inside out (§16, 2026-08-31). The host
 # holds the history now and never prunes, so this ref must answer forever.
 LANDED=${5:-tk-dfaff7}
+# The card the Files-changed pane was reported blank on (2026-08-31): two
+# recorded commits, so it exercises the RANGE arm of the rule below, not the
+# single-commit door.
+REPORTED=${6:-tk-bffa26}
 
 BASE="https://$HOST"
 C="curl -sS -m 15"
@@ -228,18 +232,74 @@ code=$(printf %s "$got" | sed -n '1s/[^ ]* \([0-9][0-9]*\).*/\1/p')
   "HTTP $code — a private board is not world-clonable"
 pass "GET /$PRIVATE/repo.git upload-pack (anonymous)" "HTTP 401"
 
-# ── (h) a LANDED card still renders its diff ─────────────────────────────────
-# Reported 2026-08-31 against the pull mirror, and the reason this chapter
-# exists: `tk-dfaff7`'s branch was pruned on the forge when its chapter landed,
-# so the board — whose git was a --mirror of the forge — lost the diff of work
-# that is intact in the trunk. The host holds the history now and refuses ref
-# deletions, so this URL must answer for the life of the board.
-head_ln "a card whose branch was pruned on the forge"
-body=$($C "$BASE/$BOARD/api/git/commit/$LANDED") || fail "GET git/commit/$LANDED" "no answer"
-case $body in
-  *'"patch"'*) pass "GET /$BOARD/api/git/commit/$LANDED" \
-    "a diff ($(printf %s "$body" | wc -c | tr -d ' ') bytes)" ;;
-  *) fail "GET git/commit/$LANDED" "no diff: $(printf %s "$body" | head -c 300)" ;;
-esac
+# ── (h) the Files-changed pane's OWN range, for the two reported cards ───────
+# This section REPLACES the one that asked `git/commit/<the landed card's branch
+# name>` (written 2026-08-30 for the pull mirror). That assertion is not
+# recoverable and must not come back: `tk-dfaff7`'s branch was pruned on the
+# forge before this host became the board's remote, so the ref was never pushed
+# here and the host — which prunes nothing — has nothing to serve under that
+# name. Asking a landed card for its BRANCH was the fault itself, in the smoke
+# as much as in the pane.
+# Reported 2026-08-31 on tk-bffa26: the pane asked
+# compare(base=ms/promotion-survives-its-own-confi, head=tk-bffa26) and the
+# milestone branch does not exist on the host — pruned before the host became
+# the remote — so a landed card showed nothing while its diff sat right there
+# (proven on tk-dfaff7, whose recorded commit answers 45898 bytes through the
+# commit door). This chapter moved the pane onto the card's OWN recorded
+# commits, so the assertion has to ask what the pane asks and derive it the same
+# way, or it would go green against a broken pane.
+#
+# `links.tsx::cardRange` is the rule, verbatim: no sha → nothing to assert; ONE
+# recorded commit → `git/commit/<sha>`, which the door already diffs against its
+# first parent; SEVERAL → `git/compare/<oldest>^...<newest>`, the base being the
+# OLDEST commit's PARENT because git diff excludes its base. `gitRoute`
+# percent-encodes each ref (`^` → %5E) and leaves `...` literal, since the
+# separator is the route's grammar and not part of either ref.
+#
+# python3 reads the shas and nothing else does the arithmetic: the payload
+# carries `sha` in more places than `commits` (the history and the merge record
+# among them), so a grep over the whole body picks the wrong first and last ref
+# and would pass on a range no pane ever asks for.
+head_ln "the pane's own range, for both reported cards"
+for card in "$LANDED" "$REPORTED"; do
+  payload=$($C -X POST "$BASE/$BOARD/api/rpc" -H 'Content-Type: application/json' \
+    -d "{\"verb\": \"card\", \"args\": {\"task\": \"$card\"}}") \
+    || fail "POST card $card" "no answer"
+  # route on the first line, the first commit's own files on the rest — one read
+  # of the payload, and the same read the pane does.
+  asked=$(printf %s "$payload" | python3 -c '
+import json, sys, urllib.parse
+data = json.load(sys.stdin).get("data") or {}
+shas = [c["sha"] for c in (data.get("commits") or []) if isinstance(c.get("sha"), str) and c["sha"]]
+if not shas:
+    sys.exit("no recorded commit")
+q = urllib.parse.quote
+route = ("git/commit/" + q(shas[-1], safe="") if len(shas) == 1
+         else "git/compare/" + q(shas[0] + "^", safe="") + "..." + q(shas[-1], safe=""))
+print(route)
+for name in (data["commits"][0].get("files") or []):
+    print(name)
+') || fail "card $card" "could not derive the pane's range"
+  route=$(printf %s "$asked" | sed -n 1p)
+  body=$($C "$BASE/$BOARD/api/$route") || fail "GET $route" "no answer ($card)"
+  case $body in
+    *'"patch": "diff'*|*'"patch":"diff'*)
+      pass "GET /$BOARD/api/$route" \
+        "$card renders ($(printf %s "$body" | wc -c | tr -d ' ') bytes)" ;;
+    *) fail "GET $route" "$card: no patch: $(printf %s "$body" | head -c 300)" ;;
+  esac
+  # THE dropped-base assertion: with the base at the newest commit (or at a
+  # branch tip) the answer is still a patch — it is just the wrong one. Only the
+  # FIRST commit's own files prove the range reaches back to where it started.
+  printf '%s\n' "$asked" | sed -n '2,$p' | while read -r name; do
+    [ -n "$name" ] || continue
+    case $body in
+      *"\"$name\""*) : ;;
+      *) fail "GET $route" "$card: $name — the first commit's file is missing from stat: the base was dropped" ;;
+    esac
+  done || exit 1
+  pass "the first commit's files are in stat" \
+    "$card: $(printf '%s\n' "$asked" | sed -n '2,$p' | wc -l | tr -d ' ') file(s) from $(printf %s "$asked" | sed -n 2p)"
+done
 
 printf '\nall green — %s, board %s\n' "$HOST" "$BOARD"

{taskops_cli-0.5.3 → taskops_cli-0.5.5}/src/taskops/_version.py

@@ -1,4 +1,4 @@
 from __future__ import annotations
 
 __title__ = "taskops"
-__version__ = "0.5.3"  # single source of truth; pyproject reads it dynamically
+__version__ = "0.5.5"  # single source of truth; pyproject reads it dynamically