Thunderbird's message composer emits U+FFFC object replacement characters
when creating or removing paragraph objects. Treating these structural
updates as inserted or deleted text sent raw placeholders to speech,
including multi-character updates that bypass ordinary key-echo settings.
Remove object markers from insertion and deletion presentation while
preserving surrounding text. Normalize typing-echo payloads before deciding
the echo kind or applying character, word, and sentence echo policies.
Return no text or echo kind for updates containing only object markers.
Add regression cases for single and repeated markers, mixed marker/text
payloads, and ordinary text through both event presentation and typing echo.
Validation:
- go test ./... -count=1, with the local cache and dependency network disabled.
- Built the reader and the BrlAPI-tagged Braille helper.
- A disposable Thunderbird profile on a private X11 display and session bus
produced one marker-containing presentation before the fix and none after.
- Private Thunderbird checks preserved character, word, and line navigation
after typing several paragraphs.
- The user confirmed main-arrow reading worked with the restarted build.
The original silent-arrow failure was not reproduced in isolation. Restarting
restored reading in the user's session; this commit does not claim to fix
its underlying cause or change arrow-key handling. Physical Braille output
was not tested. Preserve the original and fresh diagnostic logs locally.
Resolve the parity-ledger conflict by retaining testing's updated LibreOffice entry and Steam progress entry alongside deedra's edit-label and Mozilla records.
Verified: go test ./... -count=1, go vet ./..., parity ledger validation, and whitespace checks. Native GUI and physical speech/Braille acceptance were not run.
Thunderbird HTML messages expose readable text on containers around embedded
objects. The browse cursor previously considered only text leaves, and native
document capture omitted heading/link character counts. This skipped message
content, headings and link text. A caret event following Tab could also replace
a complete link announcement with its first character.
Capture owned hypertext child identities and native placeholder offsets on the
GLib thread, releasing native interfaces before returning. Require every child
to occupy one embedded-object character so providers exposing duplicate text
through static descendants retain their ordinary traversal.
Traverse native text ranges before and after embedded children without changing
their offset domains. Preserve character, word, line and document-boundary
navigation; normalize native caret and Braille routing positions at adjacent
embedded links. Keep the Braille cursor on the correct fragment when a parent
contributes text more than once to an assembled line.
Ignore same-input link caret echoes after focus while retaining subsequent
arrow and selection feedback. Add regression tests, extend the private Firefox
fixture, and record behavior and remaining acceptance in the parity ledger.
Validation:
- Full Go test suite with count=1; focused app/atspi/review race checks.
- go vet ./...; reader and brlapi helper builds.
- Private Firefox baseline, mixed-parent offsets and tab navigation.
- Private Chromium fragments, named links, document navigation and selection.
- Parity ledger validation and whitespace checks.
The existing Firefox tab test intermittently timed out on both an untouched
HEAD copy and the changed tree; the final run passed. The unchanged
TestRunReachesSoundStopAfterEarlierServiceStopError also failed once during
startup cancellation and passed in subsequent full-suite and race runs.
Both installed binaries were verified and the reader reloaded in its original
desktop session. The maintainer confirmed readable message text and later
working links after an earlier intermittent omission. That omission remains
untraced; repeat live acceptance and physical Braille verification remain open.
Combine relation-only edit-label capture while retaining bounded text reads,
password-target exclusion, UTF-8 output limits and menu checkability capture.
Keep both label regression fixtures and the existing foreground test dispatch.
Preserve Wine helper diagnostics and recovery documentation, and retain the
current parity records alongside the incoming edit-label entry.
Validation:
- go test ./... -count=1 -timeout=180s
- go vet ./...
- CTHULHU_NATIVE_EDIT_LABELS=1 python3 tests/native_atspi.py
- Isolated TestNativeTextLabelRelations
- python3 tests/native_atspi.py
- Parity ledger validation: 65 rows
- Python syntax and staged diff checks
Native GTK table fixtures emitted ATK warnings; all assertions passed.
Live speech and physical Braille were not retested.
Irctest exposes its settings-field labels through AT-SPI LABELLED_BY
relations while leaving the fields' accessible names empty. Cthulhu
previously captured only the name, so focus speech announced the text
role and contents without Nickname, Username, or Real name.
Resolve direct label targets in the shared native snapshot capture for
unnamed text, entry, and password-text objects. Preserve explicit names,
combine multiple label targets in relation order, and use label/static
text only when a target has no accessible name. Bound relation scans,
target queries, and UTF-8 output; release all native references and avoid
reading password targets. No application-specific handling is required.
Add an isolated GTK fixture and native regression test exercising capture
and speech formatting for single labels, multiple labels, explicit names,
unlabelled fields, and labelled password fields without password contents.
Expose the probe through CTHULHU_NATIVE_EDIT_LABELS=1 in native_atspi.py.
Record verification and installed-reader acceptance in the parity ledger.
Validation:
- go test ./... -count=1
- CTHULHU_NATIVE_EDIT_LABELS=1 python3 tests/native_atspi.py
- go vet ./...
- Build the reader and the brlapi-tagged Braille helper.
- Validate configuration and testdata/parity-ledger.csv.
- Read-only live irctest capture includes all three settings-field labels
in formatted speech.
- Install both tested binaries and replace the reader in its original
desktop session; verify service ownership and installed build hashes.
- Maintainer confirms the installed reader now speaks the field labels.
The full suite passes outside the sandbox; its socket and process checks
require access unavailable inside the sandbox. Physical Braille acceptance
remains pending. Preserve cthulhu.log outside the commit.
Record the user-local Go installation used by the working reader: install
cthulhu and cthulhu-braille together, validate the configuration, and use
--replace from the original desktop session. Explain that another session
bus can start a second reader instead of replacing the intended instance.
Document the complete five-file standalone wine-access installation and
the legacy helper path supported by the Go supervisor. Older Python
Cthulhu builds can overwrite the launcher and sidecar at that path while
leaving the native hook files behind. Such a helper can connect to D-Bus
without supplying the current native menu and editor support.
Describe recovery by reinstalling the standalone helper and reloading the
reader, including restarting Wine applications when older hook DLLs are
already loaded. Keep the source/build distinction explicit: rebuilding
the Go reader alone does not refresh wine-access.
Validation for these documentation changes:
- Reviewed the complete diff and ran git diff --check.
- Checked the documented shell commands with bash -n.
- Installed both documented Go build outputs into a temporary directory,
compared their bytes with the builds, and passed --check-config.
- Verified the installed reader, Braille helper, and all five Wine helper
files match the builds used for the restored installation.
Runtime validation during the preceding repair:
- Cthulhu internal/wine and internal/remote tests passed; Wine-specific
internal/app tests passed.
- The legacy installed helper failed the isolated 32-bit editor check;
the standalone helper passed it.
- Standalone controller, 32-bit late-attachment/restart, and hook-safety
checks passed on private displays and session buses.
- The user confirmed MUSHclient reading was restored. Physical Braille
was not checked.
These are documentation changes; the installed behavior already comes
from the existing source revisions, with no new runtime source patch.
Wine editor insertion events previously reached speech and Braille without
checking the user's typing echo preferences. Turning off key echo or an
individual character category therefore left Wine typing feedback enabled.
Read the live settings for each insertion and reuse the native text
presentation policy for alphabetic, numeric, punctuation, symbol, space and
diacritical input. Honor character echo and the master key echo switch;
apply the action-key setting to the helper's tab and new-line labels.
Reject unexpected multi-character insertion labels rather than treating
these ordered character events as paste notifications.
Preserve navigation, selection and deletion feedback when typing echo is
disabled. Retain repeated characters, case, uppercase/custom voice routing,
focus ordering and the existing stale-source/password protections.
Add regression coverage for live disable/re-enable, independent categories,
Unicode, character-echo overrides, and unaffected editor actions. Extend
the private D-Bus test to exercise settings changes through the real Wine
service signature. Document behavior, current word/sentence echo limits,
and live Notepad acceptance checks in the README.
Validation:
- Reproduced disabled typing reaching output with the new test before fixing.
- Full go test -race ./... -count=1 passed outside the socket-restricted sandbox.
- Private-session TestNativeSessionBusWineEvents passed with -race.
- Focused typing/editor regressions passed, including a fresh pre-commit run.
- go vet passed for internal/app, presentation, remote and wine.
- Reader and BrlAPI helper builds passed; reader --version succeeded.
- gofmt and git diff --check passed; independent review found no blockers.
Live audible speech, physical keyboard and physical Braille acceptance
remain pending. No running reader was replaced. Changes target deedra.
Wine review establishes navigation ownership and stores the window title
without speaking an activation. The native title controller treated that
metadata as an already announced title, suppressing the introduction when
review focus arrived first. Frame-to-control focus updates also reset the
window title controller, cancelling pending introductions and discarding
duplicate history.
Distinguish Wine review ownership from spoken AT-SPI activations when
classifying and deduplicating native introductions. Use the native Wine
application identity when waiting for committed application preferences,
so introductions remain audible with automatic title changes disabled
while subsequent title changes honor that application's settings.
Separate object focus cleanup from window title cleanup. Wine review
revisions now retain the pending title and its duplicate history; actual
foreground transitions, native window lifecycle cleanup, screen lock,
and shutdown continue to reset title state. Preserve XTerm/Fenrir handoff
and ordinary AT-SPI activation deduplication.
Add regressions for both native/review arrival orders, ordinary Wine
windows and virtual desktop children, duplicate reports, return to the
same window, and application preference handoff. Add a private D-Bus
event-loop test for a frame-to-control focus burst and document focused
verification and live acceptance procedures in tests/window_titles.md.
Validation:
- Confirmed the new title introduction, preference handoff, and private
focus-burst regressions fail before the fix and pass afterward.
- go test ./... -count=1 passed with local socket access; the sandboxed
attempt was blocked by Unix and loopback speech fixture sockets.
- Focused Title, WineReview, XTerm, and Foreground tests passed with
the race detector.
- Private D-Bus TestPrivateWineTitleSurvivesReviewFocusBurst passed
with the race detector.
- go vet ./internal/app, the reader build, and git diff --check passed.
Installed-reader behavior, real Wine/X11 switching, physical Alt+Tab,
audible speech, and physical Braille acceptance remain pending. The
private fixture uses copied foreground metadata and a real registry,
not a real Wine window. No installed reader was replaced.
@@ -48,6 +48,25 @@ go run ./cmd/parityledger validate testdata/parity-ledger.csv
Start verification with the affected packages; run broader checks when the change affects shared behavior or before submission where practical. Format edited Go files with `gofmt`; run `shellcheck` for edited shell scripts. Use deterministic tests and synchronization instead of timing sleeps. Keep fixtures generic, without real usernames or personal identifiers.
### Version update pre-commit hook
The optional `hooks/pre-commit` hook updates the Go `Version` constant to the
local date and `Channel` to the current branch, then stages the version file.
It skips merge commits and honors `SKIP_VERSION_UPDATE=1` for deliberate
version overrides. If an update is needed and the version file has unstaged
changes, the hook stops so those edits are not silently included in the commit.
In a clone with no existing pre-commit hook, enable it with:
```sh
cp hooks/pre-commit .git/hooks/pre-commit
```
The local copy stays active when switching to branches without the tracked
hook file; copy it again after updating the hook. Preserve and integrate an
existing hook rather than overwriting it. The old Python version hook and
`.git/versionpath` are not used by this Go hook.
### Speech regression push check
Run `python3 tests/check_speech.py` for speech queue, title, Wine review metadata,
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.