The component pin moves to bd01470 (long change-list names end before the
row's buttons). Real paths come from dunce, so Windows shows C:\... rather
than \?\C:\..., and 36 more ElyTree tests pass there. The lock returns to
windows-sys 0.61, which the last pin change had resolved down.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Release apps use the GUI subsystem; workers, Git and cmd start with
CREATE_NO_WINDOW. Each exe embeds its icon as resource 1, which gpui
also loads for the window. Patch 0044: Win32 devices take no client
buffer, so tiles and slide layers paint at full size, then copy.
ely.xcd gives both schemes one notes colour: Calc rewrote the comment
style on each switch and left an undo step. lo-patches.sh keeps LF.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Both build on packaging/stage.sh. The NSIS installer takes the staged
folder and removes only what it laid out; installer.sh drives it. The
AppImage gives linuxdeploy the app, the workers and libcef.so.
ElyPDF's test window installs the app's menus first, as ElyTree's
does. probe.yml reports what each Gitea runner holds.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
rustc and clippy are clean for the whole workspace, every target, on
both. clippy --fix moved test modules last and took the shorter
forms; long types got named aliases; ElyTree's form takes its title
and button words as one pair.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The kernel is the merged library on every platform; Windows and
Linux find it beside the app in libreoffice/program. All three run
unipoll: macOS through GCD, Linux by answering one request per poll,
Windows through a message-only window on the loop's thread (LOK's
own main thread had deadlocked Calc loads). Patch 0043 flushes
callbacks inline everywhere, so a request sees its result; 0042 stops
MSVC's Boost auto-link. Linux paints with cairo, BGRA; the start-up
check now paints red text and finds red in byte two on every
platform, since getTileMode does not see Skia.
ely-lok on Windows: 180 of 189 pass; on Linux 187. The rest are
listed after 0.1.3 (spelling, Calc dark tiles, show layers and
fields, a11y selection). Tests build file URLs portably and accept
CRLF where Windows writes it.
packaging/stage.sh lays out an app with its engines for Windows and
Linux; licenses.sh finds each licence on all three. ElyMark finds its
page beside the app.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
born() takes a HEAD as new only when it names a branch under
refs/heads/ that no ref holds yet; a broken ref, a tag target or any
other state is an error in Git's words. The branch name keeps every
byte. Round 3 is fixed; round 4 waits until 0.1.3 ships.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A failed read of what Git is doing shows as a danger alert in Git's
words. git am's Skip and Abort had no operation name in either
catalog and panicked; both catalogs now hold it. Tests cover the
unread stop and both buttons through their confirm to Git.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The branch list's click goes through Tree::pick_ref, which clears the
look like every other pick; the test covers file, commit, side,
branch and remote picks. The test window installs the app's menus
first, as tool_root does: Windows and Linux draw the menu bar in the
window and panicked on an empty one.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
WebView::open_inert makes a browser with JavaScript off; ElyMark's
staged print page uses it. webprobe opens one page both ways: the
scripted copy must report, the inert one never.
The CEF router drops a query that lacks onFailure, and queries made
before the router is ready are lost; agents.md says so.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
paper::stage takes the background executor and inlines images, reads
the style sheet and writes the page there. A gpui test checks nothing
is written before the executor runs.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A failed build removes its holder. When both editors fail, the page
tells the app and keeps the text; the next mode request rebuilds it.
mark-broken.mjs breaks the editors through setAttribute and checks
the fallback, the told failures, no leftover holders and the rebuild.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A document can forge both a console message and its source (KaTeX's
\errmessage, data: scripts, sourceURL, encoded paths), so the display
handler logs a line number and nothing the page wrote. The page's own
diagnostics reach the app over elyQuery: a missed reveal sends its
line; the visual failure and diagram lines already did. A test calls
the real handler and checks the captured log.
Round 5 found only the missing test; it is added and mutation-checked,
with no sixth review under the five-round cap. M-15 is new: the print
page runs a document's scripts.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
export_to splits picked, cancelled and failed as open_picked does;
a failure reaches panel_failed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The tally remembers why the page went down; until a load succeeds,
render() ends each new request with that reason instead of calling a
dead page. end_renders is a free function with a gpui test.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
restore_shown takes its entry through Keeping::take; the test runs a
crashed session through start, untitled, made and take, the second
tab first.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
GPUI clears windows after the quit observers, so the desk's release
ran close() and wiped what on_app_quit had kept. hold() lets the
lock go unclosed; release then finds nothing to close.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Preview clicks and Cmd/Ctrl clicks in the editor go through one
click handler; it stops the browser and sends the link mark's own
address, since Milkdown blanks unsafe hrefs in the DOM. Reviews run
at effort xhigh.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The window test now saves print jobs to a file (ELY_PRINT_TO) and
requires two distinct jobs printed ok and readable PDF text, not
a count of Chromium's log lines. It runs on macOS only.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every path that takes the disk text as saved goes through
adopt_disk, which sets the tab's mark and the view's baseline
together; the merge branch no longer repeats it by hand.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
MuPDF builds with MSBuild on Windows (/MT, OpenSSL from vcpkg);
Git for Windows' MinGit stands in for the source build there. CEF
loads from where libcef lives and sets its API version first on
every platform. File identity on Windows compares volume and index.
The checkout and reference/ keep LF on Windows.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Committed before review so Windows and WSL builds can start.
Each fix was tested and mutation-checked here; the codex review
of each item is still to come, so tasks/BUGS.md stays unticked.
M-02 is mid-review: round 1 findings are fixed (apps/ely-mark/src/tally.rs).
K-01 stays deferred. The ElyMark window test mark-outside.sh failed
once at its crash-recovery step in the last run; not yet looked into.
