Skip to content

v0.9.5 cumulative release

Choose a tag to compare

@jumping2000 jumping2000 released this 10 Sep 05:53
· 18 commits to dev_en-it since this release

v0.9.5

  • Made the worker's compile task retry on OSError — three attempts, five seconds apart — instead of giving up on the first one.
  • Fixed the VS Code default interpreter path, which pointed at a bare relative .venv/Scripts/python.exe and failed to resolve; it is now anchored with ${workspaceFolder}.

v0.9.4

  • Moved every UI string out of the React components into frontend/public/translations/{en,it,en-it}.json (243 keys, flat key → string maps).
  • Added a t(key, vars?) helper on useLang(): {name} interpolation, _one / _other plural selection driven by count, and a missing key rendering as the key itself plus a console warning
  • Replaced ~121 inline it ? 'x' : 'y' ternaries and ~60 single-language literals; removed the duplicated const it = lang === 'it' || lang === 'en-it' flag from 8 components. isItalianUI() is now actually used, in Explore.tsx, where the flag still selects a localized data field rather than UI text.
  • Added en-it.json as a third file so the bilingual mode's chrome can diverge from Italian; it currently mirrors it.json, preserving the previous behaviour.
  • Fixed two latent bugs the translation surfaced: the Wiki and Ingest status banners picked their error styling by sniffing an "Error" / "Errore" prefix out of the message text, which breaks as soon as the text is translated; both now track an explicit boolean.
  • Fixed stale era labels in the Explore timeline when switching between it and en-it: the d3 effect depended on the italian flag, which does not change across that pair.
  • Added tests/test_translations.py: guards the three files against key drift and against unpaired _one / _other plural forms.

v0.9.3

  • Added an "All domains" option to the top-bar domain selector, and made it the initial value. The selector previously defaulted to generale and always sent it, so search and ask silently excluded every article assigned to a custom domain — on a knowledge base where all articles carry a domain, that hid nearly the whole corpus.
  • Made the Search page re-run the current search when the domain filter changes; results used to stay stale until the query was resubmitted.
  • Stamped the queried domain into filed-back Q&A outputs at creation time (_file_output), so an answer produced under a domain filter is reachable by that same filter. With no filter active nothing is written, exactly as before. Pre-existing files in wiki/outputs/ are unaffected — outputs remain outside the reach of bulk_assign_domain, which only resolves paths under concepts/.
  • Fixed malformed_record_count in llm_usage.recent_requests: the counter was incremented without being initialised in the recent-requests payload. Added it to the response and to the API reference (EN/IT)