glaterza opened a new pull request, #43080:
URL: https://github.com/apache/superset/pull/43080
### SUMMARY
Two recently-added features — **API keys** and **semantic layers / views** —
shipped with
near-totally stranded Spanish catalogs. `pybabel update`'s fuzzy matcher
**copies** a
translation from a similar-looking `msgid`; when it guesses wrong, the
Spanish ends up
saying something the English never said.
Every entry corrected here is flagged `#, fuzzy`. Both the frontend and
backend builds
compile with `--use-fuzzy`, so **all of these are live in the UI today** —
the flag means
"needs review", not "withheld".
This is the same defect class as #42808, applied to two feature clusters
deliberately left
out of it: correcting one toast to say *capa semántica* while 33 neighbours
still said
*capa de anotación* would have read worse than being consistently wrong.
**These are not judgement calls about my Spanish.** The correct terminology
already exists
in this catalog on **non-fuzzy** entries, and the fuzzy entries contradict
it:
| Already in the catalog, not fuzzy | Contradicted by, fuzzy |
| --- | --- |
| `An error occurred while fetching semantic layers` → "…las capas
**semánticas**" | `Semantic Layer` → "Capa de **anotación**" |
| `%s Semantic View` → "%s Vista **semántica**" | `Semantic view does not
exist` → "El **conjunto de datos** no existe" |
| `API keys allow scoped programmatic access to Superset.` → "Las **claves
API**…" | `API Keys` → "**Clave privada**" |
| `Physical` → "Físico" | `Semantic` → "**Correo electrónico**" |
So the change is verifiable by a reviewer who does not read Spanish: it
makes the fuzzy
entries agree with the reviewed entries already in the file.
The worst of the set, all user-facing:
| msgid | Was | Literally | Now |
| --- | --- | --- | --- |
| `Inactive` | Activo | "Active" — **inverted** | Inactivo |
| `Semantic` | Correo electrónico | "Email" | Semántico |
| `Create API Key` | Creado por | "Created by" | Crear clave API |
| `Expired` | Explorar | "Explore" | Caducado |
| `Last Used` | Listar usuarios | "List users" | Último uso |
| `Definition` | desviación | "deviation" | Definición |
| `Revoked` | Trazado | "Plotted" | Revocado |
| `API Keys` | Clave privada | "Private key" | Claves API |
| `Mapbox (API key required)` | El correo electrónico es obligatorio |
"Email is required" | Mapbox (requiere clave API) |
| `Failed to fetch API keys` | No se han podido etiquetar los elementos |
"Could not tag the items" | No se han podido recuperar las claves API |
| `Add %s view(s)` | %s opción(es) | "%s option(s)" | Añadir %s vista(s) |
| `Semantic layer does not exist` | El gráfico no existe | "The chart does
not exist" | La capa semántica no existe |
| `Are you sure you want to revoke this API key? This action cannot be
undone.` | ¿Seguro que quieres eliminar las anotaciones seleccionadas? |
"…delete the selected annotations?" | ¿Seguro que quieres revocar esta clave
API? Esta acción no se puede deshacer. |
`Inactive` → "Activo" is the one I would most like a second pair of eyes on:
it is a status
badge asserting the opposite of the truth.
**On gender.** The API key status column is headed **Estado** and renders
one of Active /
Inactive / Expired / Revoked. `Active` → "Activo" is shared with the alert &
report list,
so it is effectively fixed masculine; the other three agree with it (and
with *estado*)
rather than with the feminine *clave*. Mixing genders down one column would
be worse than
either choice made consistently.
**Scope.** 64 entries: 63 get a corrected `msgstr`, and one (`API key name
is required`)
was already correct Spanish carrying a stale fuzzy flag. Every msgid was
checked to be
used **only** inside `features/apiKeys/`, `features/semanticViews/`, or
`features/semanticLayers/`, so nothing here changes wording elsewhere in the
app.
**Deliberately excluded:**
- Entries carrying a `backfill_po.py` attribution comment (12 in these
clusters). Those are
machine-translated drafts awaiting the review step in
`docs/developer_docs/contributing/howtos.md`; their problem is register,
not meaning, and
mixing them in would blur two different claims.
- `features/semanticLayers/label.ts`, which swaps "dataset" → "datasource"
and "database" →
"data connection" when the flag is on. All eight alternate labels
currently collapse to
the single string "Conexiones de la base de datos" regardless of case or
number, so
lower-case mid-sentence uses render capitalised and plural. That is a real
defect, but a
different one (case/number correctness rather than stranding), so it is
left for a
follow-up.
**On how these were found, including what the method misses.** #42808 used a
similarity
heuristic: a fuzzy entry whose `msgstr` is byte-identical to another
entry's, where the two
`msgid`s are dissimilar. That heuristic is structurally blind to the case
where a
legitimate synonym shares the stranded translation — `Revoke` → "Eliminar"
never surfaces,
because `Remove` → "Eliminar" is correct and so the pair scores as
*similar*. Lowering the
threshold does not help; a synonym is supposed to score similar.
Roughly a fifth of the entries here — including `Inactive`, `Expired`, `Last
Used`,
`Definition` and `Done` — were found instead by enumerating every `t()` call
in these three
feature directories and diffing that set against the catalog, then opening
the running UI
and reading the screens. Neither the msgid text nor the similarity score
would have flagged
them.
### BEFORE/AFTER SCREENSHOTS OR ANIMATED GIF
<!-- screenshots to be attached -->
### TESTING INSTRUCTIONS
Both features are behind default-off flags, so enable them first:
```python
FEATURE_FLAGS = {"SEMANTIC_LAYERS": True, "FAB_API_KEY_ENABLED": True}
FAB_API_KEY_ENABLED = True
```
Set the language to Spanish, then visit **Settings → API Keys**
(`/user_info/`) and the
**Semantic Layer** entry on the database connections page.
The catalog still compiles, and the entry counts move by exactly the 64
touched:
```bash
msgfmt -c --statistics -o /dev/null
superset/translations/es/LC_MESSAGES/messages.po
# master: 3931 translated, 978 fuzzy, 197 untranslated
# here: 3995 translated, 914 fuzzy, 197 untranslated
```
Only one of the 64 entries (`Add %s view(s)`) contains a format placeholder,
and its
specifier set is unchanged. `msgfmt` only checks format strings on non-fuzzy
entries, so
the check that proves clearing the flags introduced nothing is a blanket
un-fuzzy of the
whole file:
```bash
sed -E 's/^#,(.*)fuzzy(.*)$/#,\1\2/' \
superset/translations/es/LC_MESSAGES/messages.po > /tmp/t.po
msgfmt -c -o /dev/null /tmp/t.po
```
That reports **4 fatal errors on this branch and 4 on master** — the same
pre-existing
ones (`%s column`, `%s item`, `Added to 1 dashboard`, `%(suggestion)s
instead of …`),
untouched here.
### ADDITIONAL INFORMATION
- [ ] Has associated issue:
- [ ] Required feature flags:
- [x] Changes UI
- [ ] Includes DB Migration (follow approval process in
[SIP-59](https://github.com/apache/superset/issues/13351))
- [ ] Migration is atomic, supports rollback & is backwards-compatible
- [ ] Confirm DB migration upgrade and downgrade tested
- [ ] Runtime estimates and downtime expectations provided
- [ ] Introduces new feature or API
- [ ] Removes existing feature or API
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
