Skip to content

Firmware page: notes, steps and error messages are translated - #182

Merged
UniqueDroid merged 11 commits into
mainfrom
fix/firmware-note-i18n
Nov 19, 2026
Merged

UniqueDroid merged 11 commits into
mainfrom
fix/firmware-note-i18n

Conversation

@UniqueDroid

@UniqueDroid UniqueDroid commented Nov 18, 2026 •

Copy link
Copy Markdown
Owner

Found by Jan in the German interface: under a checked firmware file the note read "Uploaded file – verified: checksums of all parts, made for this device (without .sha256 file). Nicht vom ShareOS-Projekt signiert. Settings und packages are kept." – and other messages of the update were English too.

Stacked on #179 (CHANGELOG.md, numbers in tools/i18n) – merge #176, then #179, then this one. #181 touches CHANGELOG.md too; whichever of the two is merged second gets main merged in.

How it was checked

System → Firmware drawn in Chromium against tests/mock-ubus.py in every state (idle, not checked, no release, downloading, verifying, ready in five variants, each of the eight writing steps, written, not started from flash, every error of the updater), plus the toasts and dialogs of the page – once in English and once in each of the nine languages. Text that is the same in both, or still contains English word pairs, is not translated.

What was English

  1. The note under a checked file. The translations knew it as one text. Since the box checks signatures, the sentence about the signature sits in the middle (as a <b> when the file is unsigned) and splits it into parts no translation knew. Each sentence is an element of its own now (E1700–E1703).
  2. The step while writing. Preparing the system for the restart … was one text with its … and matched nothing; the fallback for single words made "Prüfunging the package again …" out of others. The label is an element of its own.
  3. Every error message of the updater (shareos-update, parts/rpc/update.uc) and of shareos-flash-bin – none was in the catalogue. 33 texts and 20 patterns added, all nine languages. The two results of the installation at boot (failed <version>: …) too.
  4. Messages behind a prefix (Download: …, Install: …, Firmware file: …, Check failed: …) and other firmware in the dialogs: the pattern translated the prefix and left the value. i18n.js now translates a value inside a pattern when it is a text of its own – several words, known as a whole or as a pattern. Names and numbers stay as they are. docs/TRANSLATING.md says so.

What stays English

Output of system tools passed through in a message: curl exit 6, tar: short read, No space left on device, flash_erase …, nandwrite ….

Error messages of the other tools of the box (users, storage, apps, …) are not in the catalogue either – not part of this pull request.

Also in here: storage setup waits for its partitions

test-storage failed in the run of #181 with mkswap: cannot open /dev/loop1p1: No such device or address – the failure that has come and gone in the storage tests before. partition() in shareos-storage waited only for the node of the data partition. Where the table is read once more after sfdisk has closed the disk (udev on the build machine; a box has none), a partition is removed and added back: the node is there, opening it fails. It now waits until every new partition can be opened and tries mkswap again.

tests/check-i18n.py, tests/check-i18n-coverage.js and tests/static-analysis.sh pass.

https://claude.ai/code/session_01Sb3gb1FHs6DNkZHKinKAJv


Generated by Claude Code

A copy or a move to another drive that goes on in the background of the box
counts what the sources take (total_kb) and what the targets take so far
(done_kb); job_cancel stops it with everything it started. App SDK:
fs.copy/move(…, { onProgress }) with { total, done, cancel() }, fs.progress.
The background jobs get their own part (files-jobs.uc) and a host test.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Sb3gb1FHs6DNkZHKinKAJv
…oks at that group

The measuring starts new processes between the look at the group and the
signal, so one pass could miss one. The test looked for the leftovers by
name ('sleep 60'), which also matches processes that are not the job's.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Sb3gb1FHs6DNkZHKinKAJv
appfs.open with upload: a ticket of mode 'put' – cgi/file takes any size
for it and moves the part into place; the part is removed when the request
is stopped. save() of a document says 'file too large (at most 64 MB)'
instead of ending in a cut connection. SDK: ShareOS.fs.upload(root, path,
name, blob, { onProgress }) over XMLHttpRequest, with cancel().

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Sb3gb1FHs6DNkZHKinKAJv
…, no-cache)

The packages are built with the file time 0, uhttpd named 1970 as
Last-Modified, and a browser used its copy of such a file for years
without asking again: after an update the old program kept running
(found with ShareExplorer on Jan's box).

shareos-apps: every installation of a web app sets the file times and
makes a link r<time> -> . in the app's folder; 'entry' in the list goes
through it, so every file of the app has a new address. Apps installed
by an older version get both in fix-perms (at boot).

uhttpd patch 0004: with UHTTPD_NO_CACHE in the environment static files
carry Cache-Control: no-cache; the apps server runs with it.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Sb3gb1FHs6DNkZHKinKAJv
Every NFS client acts as nobody (all_squash, anonuid=65534), who gets
the rights on a share's folder by its ACL. Personal folders keep their own
rights (0700, the account's), so such a share could never be used. A share
that is a personal folder or lies in one gets anonuid=<uid of the account
the folder belongs to>; everything else stays with nobody. Note in Shares,
translated; three checks in test-users.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Sb3gb1FHs6DNkZHKinKAJv
…ses the current address

ShareOS.appInfo() reads the app's app.json ({id, name, version, publisher}),
and an element with data-sos-version gets the version as its text – apps
kept the number by hand in the title bar, the About text and each language.

openApp opened /<id>/index.html; with 'files' the box now names the entry of
the app's current installation with the grant, so the browser does not take
files it kept from an older one.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Sb3gb1FHs6DNkZHKinKAJv
With signature checking on, the sentence about the signature sat in the
middle of the note and split its text into pieces that are in no
translation. Each sentence is an element of its own now, with its own
text in the catalogue (all nine languages, taken from the existing
translations of the whole note).

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Sb3gb1FHs6DNkZHKinKAJv
Found with a rendered check of System -> Firmware in every state and all
nine languages:
- the step shown while writing was one text with its ' …' and matched no
  translation ('Prüfunging the package again …'); the label is an element
  of its own now
- no error message of shareos-update, of the update part of the plugin and
  of shareos-flash-bin was in the catalogue: 33 texts and 20 patterns added,
  with all nine translations
- i18n.js: a value inside a pattern ($1 in 'Download: $1') is translated
  when it is a text of its own (several words, known as a whole or as a
  pattern), so the messages are translated behind a prefix too, and 'other
  firmware' in the dialogs

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Sb3gb1FHs6DNkZHKinKAJv
@UniqueDroid UniqueDroid changed the title Firmware: the note under a checked file is translated again Firmware page: notes, steps and error messages are translated Nov 18, 2026
test-storage failed on GitHub with 'mkswap: cannot open /dev/loop1p1: No
such device or address' (run 44358572104): partition() waited for the
node of the data partition only. Where the table is read again after
sfdisk has closed the disk, a partition is removed and added back; its
node is there, opening it fails. Wait until every new partition opens,
and try mkswap again.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Sb3gb1FHs6DNkZHKinKAJv
The run with the first fix got past mkswap and failed one step later:
'mdadm: cannot open /dev/loop2p2: No such device or address' (run
44374928238) – the first disk's partitions were gone again for a moment
while the second one was partitioned. wait_parts before mdadm --create,
mdadm --add and mkfs. test-storage prints the partitions, swaps and the
udev state when a setup fails.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Sb3gb1FHs6DNkZHKinKAJv
@UniqueDroid
UniqueDroid marked this pull request as ready for review November 19, 2026 15:09
@UniqueDroid
UniqueDroid merged commit 554af7c into main Nov 19, 2026
7 checks passed
@UniqueDroid
UniqueDroid deleted the fix/firmware-note-i18n branch November 19, 2026 15:10
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