Repository navigation
v0.9.3
Added
-
A draft state for a lab:
status: draftinlab.yaml(#369). Optional,
publishedby default, so no existing catalog changes. A draft lab is left
out oflist-labs(shown, marked "draft", with the newlist-labs --drafts),
ofnextandstart, ofcourse(unless the newcourse <id> --drafts) and
ofprogress. Its attempts are neither exported nor pushed, and
export --lab <id>refuses it.run,check,submitandshowstill
accept it by its id, with a line saying it is a draft.validate-structure
validates it like any other lab, and flags an unknown value, which behaves as
published, asfeedbackdoes.scoresis unchanged: it is the local
history of what was played.list-labs --jsonandshow --jsoncarry
statusfor every lab, an additive field. Documented indocs/contract-v1,
docs/machine-outputandschemas/lab.schema.json. -
A draft state for a catalog:
draft: truein the packaged manifest.
catalog listand the welcome screen no longer offer such an entry;
catalog list --draftsshows it, andcatalog add <url>or<id>still
installs it.catalog list --jsoncarriesdraftfor every known entry.
certificatsgoes back to draft until it publishes a lab. -
doc_urlandcriteria[].lessonsare translatable inlab.<lang>.yaml.
A lesson exists in each language of the site at a URL that cannot be derived
from the other, and a lab had a singledoc_url:guidesent an English
learner to the French page.lab.yamlnow carries the English URLs, as it
carries the Englishtitle, andlab.fr.yamlmay give the French ones;
guide,showandlist-labs --jsongive the URL of the session's language,
and a translation without them inherits the English ones.validate-structure
checks the form of every URL a translation substitutes, without counting an
inherited one twice, and--check-urlsqueries translateddoc_urls too. No
existing catalog changes: alab.yamlthat keeps its French URL behaves as
before.
Fixed
-
The engine documentation said two false things.
lab.<lang>.yaml
overrides thelabelof each criterion too, not onlytitleand
description; and indocs/machine-output,dsoxlab statusgives the state
of a lab, while host reachability isdsoxlab infra status. -
scripts/gen-catalogues.pyrefuses a manifest entry that publishes no lab,
unless it is a draft. Version 0.9.2 measuredcertificatsat zero labs,
left it out of the README table, and letcatalog listoffer it to learners
anyway. Such an entry is now a discrepancy (exit 1), with a message that says
to mark itdraft: trueor remove it. A draft entry is measured, stays out of
the table even with labs, and is never reported, nor as a forgotten public
repository.
Changed
dsoxlab new cataloglinks to the engine documentation instead of copying
it (#201). The README it wrote carried a copy of the contract and of a
lab's lifecycle, frozen on the day ofnew catalog; an inventory on
2026-11-14 found 50 false files in 21 catalogues, most born from that
skeleton. The section is now "The engine, documented where it lives", with
links tocontract-v1,commands,learnerandcatalog-author. The
generated CI workflow no longer saysvalidate-structureonly sees what
loaded, false since 0.1.97.validate-structurereports engine documentation that a catalogue copied
and that became false (new familystale_doc, #201). It looks for known
false sentences in versioned Markdown outsidelabs/andsolution/,
CHANGELOGs excluded:list-labsto run beforevalidate-structure(false
since 0.1.97),dsoxlab statusas the host reachability test (it is
dsoxlab infra status),runtime.host(never read), fixtures copied flat
(false since 0.1.37). Measured on the catalogues on 2026-11-14: 38 files in
18 catalogues, no false positive.validate-structure --jsongains the
stale_doccount, an additive key.
Documentation
- Durable decisions now live in
docs/decisions/, one record each (#203).
Several modules carried architecture reasoning in their comments: why a
flockand why its file is never deleted, whyansible-coreis declared
despite its 100 MB, why first-boot packages do not block. That reasoning was
in the right words but in the wrong place: it made files long, and could not
be found when asking "why this choice?". Fourteen records, each in English
and French, with the same short shape: context, decision, consequences, what
would reverse it, status and date. The eight decisions listed in the issue
are joined by six taken since: the fixture reversal, thedegradedstate of
a lab without its engine, the three outcomes of a check, the stable index of
optional hosts, the refusal of out-of-contract hosts, and the catalogue table
measured on the published repositories. The reasoning was moved, not
summarized: every measurement keeps its date. - The code keeps one line and a pointer.
locking.py,pyproject.toml,
the two template READMEs,test_journal_en_anglais.py,infra/terraform.py,
scripts/gen-catalogues.pyand a few docstrings now say the essential in one
line and point at their record. Comments that explain a single line stay
where they are. No behaviour changes. tests/test_decisions.pykeeps the directory honest: each record has its
sections and its twin in the other language, the index lists every record,
and everydocs/decisions/…pointer written in the repository leads to an
existing file. It is shown to refuse a broken pointer.docs/README.mdanddocs/design-principles.md(and their French versions)
point at the new directory: the principles say what is true, the decisions
say why it was chosen.