Skip to content

feat: import the GNOME screen-reader stack (orca, speech-dispatcher) - #466

Merged
hanthor merged 26 commits into
mainfrom
ccr-6715b6dc-lf5psn-screen-reader
Oct 15, 2026
Merged

hanthor merged 26 commits into
mainfrom
ccr-6715b6dc-lf5psn-screen-reader

Conversation

@hanthor

@hanthor hanthor commented Oct 15, 2026

Copy link
Copy Markdown
Member

Why

Utah needs orca and speech-dispatcher (+ -espeak-ng) for Bluefin parity. Its testsuite also checks them: orca.feature runs rpm -q orca speech-dispatcher and spd-say -O. That check currently fails on every Utah flavor and blocks the promotion gate (#447, utah#741).

Hummingbird ships none of this stack. I checked its repodata against Fedora's specs. The runtime closure is ten source packages, not three:

Package Why
orca the screen reader
speech-dispatcher spd-say and the speech server. Ships -espeak-ng, -libs, -utils and python3-speechd
espeak-ng the default synthesizer, needed by speech-dispatcher-espeak-ng
pcaudiolib, sonic espeak-ng's audio output and speed-up libraries
dotconf, libao libraries speech-dispatcher links against
pyatspi python3-pyatspi, a hard Requires of orca
brltty python3-brlapi, a hard Requires of orca
liblouis python3-louis, a hard Requires of orca

How

  • Nine of the ten came through Import Rawhide package: sonic, orca, speech-dispatcher, espeak-ng, pcaudiolib, dotconf, pyatspi, brltty, liblouis. That workflow had been broken since Oct 6 and was fixed in fix(import): let the isolated lock step write its output #456. Their bot branches are merged here as one change, because every import rewrites the same generated files. The source lock, .packit.yaml and docs/architecture.md were then regenerated with the repo tools.
  • libao by hand. Fedora's sources file pins its tarball by MD5, which validate rejects.
    • The upstream tarball from downloads.xiph.org has the same MD5 as Fedora's pin, 9f5dd20d….
    • It is now locked by SHA-512 with an https upstream url.
    • The Fedora lookaside is the fallback, on the /md5/ path, because no /sha512/ upload exists for it.
  • brltty is locked over https. brltty.app serves byte-identical content there.
  • orca is pinned to stage: 1. It BuildRequires brltty (brlapi-devel), and brltty BuildRequires libicu. Without the stage, test_icu_staging fails. feat(terminal): import Ptyxis with its VTE and simdutf dependencies #413 did the same for ptyxis and vte291.
  • Backlog. 17 names move to [resolved] and the snapshot is regenerated. The entries carry pr = 0 with a note citing Build orca, speech-dispatcher, speech-dispatcher-espeak-ng for Hummingbird #447; I'll set the real PR number in a follow-up commit.
  • Skill write-back in docs/skills/repeated-mistakes.md §23:
    • consolidating imports that share generated files;
    • MD5-pinned Fedora sources files.

Validation

All exit 0:

  • tools/factory_contract.py
  • tools/validate.py: 421 source RPMs
  • tools/multimedia_closure.py --check
  • tools/check_workflow_quoting.py
  • tools/runtime_contract.py --check
  • tools/factory_build_backlog.py --check
  • pytest tests: 1226 passed, 2 skipped

Not yet known

None of these ten has been built in this factory yet. That happens in the rebuild after merge, or in Canary if it picks them up. brltty is the largest and the most likely to need a build fix. After publication, Utah's install check will show whether any further runtime dependency is missing from Hummingbird.

Supersedes the individual import PRs (#455 and the bot import PRs for these nine packages).

Refs #447

🤖 Generated with Claude Code

https://claude.ai/code/session_01EvvFSAwG7AzVee1EtWLK15


Generated by Claude Code

claude and others added 26 commits October 15, 2026 18:42
7231f42 runs the source-lock step in the packit container with
--cap-drop=ALL. Root inside the container then has no CAP_DAC_OVERRIDE and
cannot write the runner-owned /output directory, so every import since has
died on "Permission denied: /output/candidate.json" before opening a pull
request. Make the disposable output directory world-writable; the container
keeps every other restriction.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EvvFSAwG7AzVee1EtWLK15
Review on #456: instead of a world-writable output directory, run the
packit container as the runner's uid. The output directory then needs no
mode change, and the untrusted spec expansion no longer runs as uid 0.
HOME points at /tmp, since that uid has no home in the image.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EvvFSAwG7AzVee1EtWLK15
…15b6dc-lf5psn-screen-reader

# Conflicts:
#	config/upstream-sources.json
…' into ccr-6715b6dc-lf5psn-screen-reader

# Conflicts:
#	config/upstream-sources.json
…cr-6715b6dc-lf5psn-screen-reader

# Conflicts:
#	config/upstream-sources.json
…ccr-6715b6dc-lf5psn-screen-reader

# Conflicts:
#	config/upstream-sources.json
…-6715b6dc-lf5psn-screen-reader

# Conflicts:
#	config/upstream-sources.json
…-6715b6dc-lf5psn-screen-reader

# Conflicts:
#	config/upstream-sources.json
…6715b6dc-lf5psn-screen-reader

# Conflicts:
#	config/upstream-sources.json
…r-6715b6dc-lf5psn-screen-reader

# Conflicts:
#	config/upstream-sources.json
Fedora's sources pins libao-1.2.0.tar.gz by MD5, which the import
workflow's lock step and tools/validate.py reject. The upstream release
at downloads.xiph.org (and the OSUOSL mirror) has MD5
9f5dd20d7e95fd0dd72df5353829f097, the bytes Fedora pins, so sources is
repinned to its SHA-512 and the lock names the upstream URL, the
release-monitoring feed (project 7629) and the md5-path lookaside as
fallback. Regenerates .packit.yaml and the architecture counts for the
consolidated screen-reader imports. Refs #447.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EvvFSAwG7AzVee1EtWLK15
Moves the 17 backlog names the imported recipes build (orca,
speech-dispatcher and its subpackages, espeak-ng, brltty, brlapi,
python3-brlapi, liblouis, liblouis-tables, python3-louis, pcaudiolib,
dotconf, python3-pyatspi, libao) to [resolved] and regenerates the
snapshot. Refs #447.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EvvFSAwG7AzVee1EtWLK15
Records how to merge several import/rawhide-* branches into one and
regenerate their shared files, and how to repin an MD5 Fedora sources
file to SHA-512 after verifying the upstream bytes (libao).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EvvFSAwG7AzVee1EtWLK15
orca BuildRequires brlapi-devel, brltty and python3-brlapi from brltty,
which BuildRequires libicu-devel. With both at stage 0,
tests/test_icu_staging.py flags orca as able to resolve Fedora's
ICU-linked brltty. Stage orca at 1, as ptyxis/vte291 were in #413,
refresh the architecture stage count, and note the cross-import edge in
repeated-mistakes section 23.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EvvFSAwG7AzVee1EtWLK15
The imported lock pointed at http://brltty.app. The https URL serves
byte-identical content (SHA-512 checked against the lock).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EvvFSAwG7AzVee1EtWLK15
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EvvFSAwG7AzVee1EtWLK15
@hanthor
hanthor merged commit 514e664 into main Oct 15, 2026
169 checks passed

@hivecommons-hive hivecommons-hive Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Reviewed — no findings from correctness, security, intent-alignment, style, docs-currency.

Confidence: 5/5 (safe)

— hive: agent=reviewer backend=copilot model=claude-fable-5.1 copilot=1.0.88

hanthor added a commit that referenced this pull request Oct 15, 2026
…#470)

The first main build of #466 failed liblouis and brltty at lock resolve:
their documentation BuildRequires (texlive-xetex, linuxdoc-tools) pull
TeX Live, whose xetex requires libicuuc.so.77, and the build root excludes
ICU 77. orca then failed on Fedora's ICU-77 brlapi for want of ours.

- liblouis does not link ICU: opt it in with buildroot_icu77, as libvdpau
  does for the same TeX Live chain.
- brltty links ICU, so ICU 77 in its root would ship an unusable brltty:
  drop linuxdoc-tools instead (configure then skips the manuals) and make
  the manual cleanup rm -f.

Refs #447


Claude-Session: https://claude.ai/code/session_01EvvFSAwG7AzVee1EtWLK15

Co-authored-by: Claude <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants