ci: upload translation sources to Weblate after releases by jplexer · Pull Request #2097 · coredevices/PebbleOS

jplexer · 2026-09-17T12:41:08Z

No description provided.

gmarull · 2026-09-17T13:00:48Z

        run: pbl build

+      - name: Prepare translation source artifact
+        if: inputs.collect_translation_sources && matrix.slot == 0 && github.event_name == 'push' && startsWith(github.ref, 'refs/tags/v')


should we upload on any push (ie, not only tags)?

gmarull · 2026-09-17T13:02:30Z

+    inputs:
+      collect_translation_sources:
+        description: Collect source catalogs for a firmware release
+        type: boolean
+        default: false


is this needed because on release other conditions would be true?

gmarull · 2026-09-17T13:04:34Z

+      - name: Test release source delivery
+        run: PYTHONPATH=tools python -m unittest discover -s tools/tests -p '*translation_source.py'


hmm unittest here feels weird, what's this doing?

gmarull · 2026-09-17T13:05:13Z

+    if: >-
+      github.repository == 'coredevices/PebbleOS' &&
+      github.event_name == 'push' && startsWith(github.ref, 'refs/tags/v') &&
+      vars.WEBLATE_UPLOAD_ENABLED == 'true'


why is var needed?

gmarull · 2026-09-17T13:07:25Z

+# Use Weblate's native PO source merge; do not follow redirects with credentials.
+curl --silent --show-error --fail-with-body --proto '=https' \
+  --connect-timeout 15 --max-time 180 --retry 3 --retry-max-time 600 \
+  --header "Authorization: Token $WEBLATE_TOKEN" \
+  --header 'Accept: application/json' \
+  --form-string 'method=source' --form "file=@$1;type=text/plain" \
+  --output weblate-response.json --write-out '%{http_code}' \
+  "$WEBLATE_SOURCE_URL" > weblate-status.txt
+# Redirects and unexpected successful responses must not masquerade as imports.
+test "$(cat weblate-status.txt)" = 200
+jq -e '.result == true' weblate-response.json > /dev/null
+echo 'Released source catalog uploaded to Weblate.'


Things that feel weird:

Use single language script (Python?) for all stuff

Unit tests for simple tools feel like overkill

gmarull

Thanks for putting this together. Following up on my inline comments with the full picture: the general direction I'd like is simplicity and consistency with the workflows we already have. Once a couple of assumptions are removed, this feature is roughly fifteen lines of YAML with no scripts and no tests.

The merge job is not needed

The POT is identical for every board. Every service that owns translatable strings is enabled in src/fw/prj_normal.conf (not per board), and src/fw/apps/CMakeLists.txt globs every app into apps.pot regardless of what a given board builds. So the multi-board msgcat union, the plural-conflict check, the per-board revision cross-check and the normalisation in tools/prepare_translation_source.py only reconcile catalogs that differ in their POT-Creation-Date header. Uploading the POT from one matrix leg is enough.

The release coupling is wrong, not just heavy

upload_translation_source.sh picks the newest release by published_at and skips otherwise. Maintenance branches break that: on 2026-09-15 both v4.30.3 and v4.27.3 were released, and v4.27.3 came out later. With this PR the 4.27 strings would be pushed to Weblate after the 4.30 ones.

Uploading on push to main removes the bug together with the gh api pagination, the concurrency group and the whole translation-sources.yml workflow. If per-push uploads feel too chatty, nightly.yml already builds main once a day and skips unchanged SHAs, so attaching the upload to its publish job is the other consistent option.

Answers to the inline questions

collect_translation_sources and the four copies of the push && refs/tags/v guard are redundant: release.yml only runs on tags, and a called workflow inherits the caller's event context. With push-to-main they all go away.
WEBLATE_UPLOAD_ENABLED is not needed. Gate on github.repository == 'coredevices/PebbleOS' like nightly.yml and the eng-dash step do. WEBLATE_SOURCE_URL should be hardcoded in the workflow, the way ENGDASH_URL is in .github/actions/publish-engdash.
The unittest step in the release path runs tests nothing else runs (nothing in CI executes tools/tests). Both test files should go; test_upload_translation_source.py fakes gh and curl through PATH to exercise a bash script, which is far beyond what one curl call warrants.
Language mix: with the merge gone there is nothing left for Python to do. The upload is one curl --fail-with-body plus jq -e .result, inline in the workflow, matching the style of the eng-dash action.

Secrets

The call from release.yml passes no secrets: inherit; it only works because environment: translations on the job pulls environment secrets directly. A plain repository secret plus secrets: inherit, like ENGDASH_OTA_TOKEN, is simpler and matches the other workflows.

The one thing worth keeping

Weblate's method=source upload runs msgmerge over every catalog and commits whatever changed, and a fresh POT-Creation-Date changes every PO header. Stripping that line before upload avoids a churn commit in pebbleos-translations on every upload. That is a single sed line, not a script.

Proposed shape

In build-firmware.yml, add cp build/pebbleos.pot artifacts/ to the existing "Copy artifacts" step under if: matrix.board == 'asterix'. The POT then also lands on releases and nightlies as a normal asset, which is useful on its own since the pebbleos-translations tooling takes a POT as input.
In build-firmware.yml (push-to-main path) or nightly.yml, one job gated on the repository name that downloads the asterix artifact, drops the date header and does one curl to the hardcoded Weblate URL with a repository secret.
Drop translation-sources.yml, both Python files, the shell script, the tests, the find exclusion in release.yml and the paths-filter addition.

Housekeeping

The commit carries Signed-off-by: GPT-6 <noreply@openai.com>. The DCO sign-off is a human attestation, so the AI belongs only in Co-authored-by. The PR body is also empty; please describe the intended flow there.

gmarull · 2026-09-17T13:15:59Z

One correction to my review: I claimed the POT is identical for every board. That happens to be true today, but it is not guaranteed. src/fw/services/CMakeLists.txt adds each service with pbl_add_subdirectory_ifdef, so a board that disables a CONFIG_SERVICE_* option would silently lose that service's strings from its own POT.

Rather than merging catalogs across boards in CI, I'd make the property hold by construction. src/fw/apps/CMakeLists.txt already does this on purpose: it extracts strings from every app "whether or not this configuration builds the app". Services should follow the same rule, so that any single build produces the complete catalog. That can be a small separate PR or the first commit here. With that in place, the rest of my review stands: upload the POT from one matrix leg, no merge step.

If a CI-side union is still preferred, it is msgcat --use-first over the downloaded POTs, a couple of lines of bash in the upload job; the rest of prepare_translation_source.py is still not needed.

take review comments with a grain of salt, not all may be accurate

gmarull · 2026-09-17T13:23:24Z

In general, if we are going to make universal language packs (ie, not per watch model), we can simplify the POT generation by globbing everything, regardless of what gets configured. This would remove the need for merge, we could even just invoke gettext directly?

jplexer · 2026-09-17T14:54:27Z

we can simplify the POT generation by globbing everything, regardless of what gets configured.

I think this was the case before cmake (lmao), ill revert that behavior

gmarull · 2026-09-18T07:44:33Z

+  pbl_filter_sources(pot_sources EXCLUDE apps/prf/** applib/vendor/**)
+  pbl_firmware_pot(${PROJECT_BINARY_DIR}/pebbleos.pot ${pot_sources})


maybe make pbl_firmware_pot allow passing 'EXCLUDE' directly?

gmarull · 2026-09-18T07:45:06Z

+    if: >-
+      github.repository == 'coredevices/PebbleOS' &&
+      github.event_name == 'push' && github.ref == 'refs/heads/main' &&
+      vars.WEBLATE_SOURCE_URL != ''


Extract strings from all normal firmware sources regardless of board or configured services. Include headers and definitions, while excluding recovery-only apps and applib vendor sources. Replace the per-area catalogs and merge machinery with one gettext pass so every normal firmware build produces the complete source catalog. Co-authored-by: GPT-6 <noreply@openai.com> Signed-off-by: Joshua Jun <lets@throw.rocks>

Include the POT in the asterix artifact and upload it to Weblate after successful main builds. Strip POT-Creation-Date to avoid catalog churn. Use WEBLATE_SOURCE_URL as a repository variable and WEBLATE_TOKEN as a repository secret. Release and maintenance builds do not upload sources. Co-authored-by: GPT-6 <noreply@openai.com> Signed-off-by: Joshua Jun <lets@throw.rocks>

jplexer marked this pull request as ready for review September 17, 2026 12:41

jplexer requested a review from gmarull as a code owner September 17, 2026 12:41

gmarull reviewed Sep 17, 2026

View reviewed changes

gmarull previously requested changes Sep 17, 2026

View reviewed changes

jplexer force-pushed the codex/weblate-release-upload branch from 6cd53f3 to 26327be Compare September 17, 2026 15:04

jplexer requested a review from gmarull September 17, 2026 15:04

gmarull reviewed Sep 18, 2026

View reviewed changes

jplexer and others added 2 commits September 18, 2026 11:29

jplexer force-pushed the codex/weblate-release-upload branch from 26327be to 2179773 Compare September 18, 2026 09:30

gmarull approved these changes Sep 18, 2026

View reviewed changes

jplexer merged commit 0225541 into coredevices:main Sep 18, 2026
49 checks passed

Navigation Menu

ci: upload translation sources to Weblate after releases - #2097

ci: upload translation sources to Weblate after releases#2097
jplexer merged 2 commits into
coredevices:mainfrom
jplexer:codex/weblate-release-upload

jplexer commented Sep 17, 2026

Uh oh!

gmarull Sep 17, 2026

Uh oh!

gmarull Sep 17, 2026

Uh oh!

gmarull Sep 17, 2026

Uh oh!

gmarull Sep 17, 2026

Uh oh!

gmarull Sep 17, 2026

Uh oh!

gmarull left a comment

Uh oh!

gmarull commented Sep 17, 2026

Uh oh!

gmarull commented Sep 17, 2026

Uh oh!

jplexer commented Sep 17, 2026

Uh oh!

gmarull Sep 18, 2026

Uh oh!

gmarull Sep 18, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

		- name: Test release source delivery
		run: PYTHONPATH=tools python -m unittest discover -s tools/tests -p '*translation_source.py'

		pbl_filter_sources(pot_sources EXCLUDE apps/prf/ applib/vendor/)
		pbl_firmware_pot(${PROJECT_BINARY_DIR}/pebbleos.pot ${pot_sources})

Conversation

jplexer commented Sep 17, 2026

Uh oh!

gmarull Sep 17, 2026

Choose a reason for hiding this comment

Uh oh!

gmarull Sep 17, 2026

Choose a reason for hiding this comment

Uh oh!

gmarull Sep 17, 2026

Choose a reason for hiding this comment

Uh oh!

gmarull Sep 17, 2026

Choose a reason for hiding this comment

Uh oh!

gmarull Sep 17, 2026

Choose a reason for hiding this comment

Uh oh!

gmarull left a comment

Choose a reason for hiding this comment

The merge job is not needed

The release coupling is wrong, not just heavy

Answers to the inline questions

Secrets

The one thing worth keeping

Proposed shape

Housekeeping

Uh oh!

gmarull commented Sep 17, 2026

Uh oh!

gmarull commented Sep 17, 2026

Uh oh!

jplexer commented Sep 17, 2026

Uh oh!

gmarull Sep 18, 2026

Choose a reason for hiding this comment

Uh oh!

gmarull Sep 18, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants