Repository navigation
Page dates move only when the text does, a page changed after the release says so, every Dataset names who made it, and a published llms shard name keeps answering - #1154
Open
jmrplens wants to merge 6 commits into
Conversation
… key only machines read moves no page's date, an unclassed key stops the build by name, and the conformance stamp says which commit the report was generated from The date a page last changed passed over a commit that changed only its title, description or sidebar entry, and nothing else in the frontmatter. The commit that gave each errata page the works it corrects added a key read only by the structured data, and 124 errata pages took 14 November 2026 as the day their content changed, in the footer, in dateModified and in the sitemap, with no word of them changed. site/src/lib/frontmatter-keys.mjs now classes every key a page can carry: content (what a reader reads, or how the page is laid out), label (the title, description, sidebar entry and topic) or metadata (corrects, searchTitle, searchWeight, pagefind: read by the structured data or the site search, never shown). The date rule's dateless keys are the label and metadata classes. The rule refuses a key in no class, on either side of a change and whatever the body did, naming the key, the page and the commit; the content collection refuses a key its schema declares and the list does not class, so a new key is classed the day it is added. A key is read as YAML reads it, bare, dotted or in quotes, so a quoted key is classed and refused like a bare one and is never taken for part of the key above it. On today's history the 124 errata pages fall back to the dates of their last real change (121 to 28 October, 3 to 2 November); the Spanish IEC 60268 page keeps 14 November, its last commit being a visible edit. The conformance report's stamp said its rows last changed in the commit that last touched docs/conformance.json, and the release commit changed only the version field. It now reads "Generated for phonometry 4.0.0 from commit 39f1973, on 17 November 2026", and the same in Spanish, which is what the line measures.
…ery Dataset names who made it and its other language, every work names its publishing body by one identity, and each Environments page says when its commands last ran and on what - The Spanish landing page gives the CNOSSOS-EU environment row the scope the English one states: the source side of Annex II, sections 2.2 and 2.3, with propagation through ISO 9613-2. The site-wide graph every page carries reads both rows, so it no longer claims the whole annex in Spanish. check-i18n-parity.mjs now holds the two landing pages' designations to the same scope, area by area, through scripts/shared/designation-scope.mjs: each designation is read as the words and numbers it names, with the Spanish words for the same thing read as the English ones, so a limit one language states and the other leaves out stops the build. - The 48 catalogue Datasets name the site's Person as creator and publisher and say they are free to read; their rows stay attributed to the printed works through isBasedOn and claim no licence. The two conformance report Datasets name each other as workTranslation and translationOfWork, as their articles do. check-structured-data.mjs requires both. - citations.mjs looks every kind's publisher up among the standards bodies, not a standard's alone, and knows ECAC and EASA: an RFC, an AES paper or ECAC Doc 29 now names its body by the identity its standards use. The check fails on any work of the bibliography that writes such a body out anonymously. - The four answer-first guides on predicted, facade and field insulation and on occupational exposure say "in Python" in their descriptions, in both languages. - llms.txt describes the Bibliography and About entries with their pages' own descriptions, and a test holds every line that links a page outside the API section to say something of it. - Each Environments page, in English and in Spanish, opens with the day its commands were last run, the phonometry release and the versions of every tool and image they ran on (17 November 2026, phonometry 4.0.0 from PyPI). tests/environment_checks.json records that run and tests/test_environment_checks.py holds every page to it, both ways. The pages link pip, uv, conda, conda-forge, mamba, JupyterLab, JupyterLite, Pyodide, micropip, Docker, the python image, Podman, mypy and pyright once each; the JupyterLite page links Getting Started and the Jupyter page, and the type checking page the API reference.
…ts title and in its Markdown copy, and the next release takes the notice off with nothing typed for it The release is the commit the tag v<VERSION> names, or, before the release workflow has pushed that tag, the commit that set VERSION. A page is newer when the commit that last changed its content was committed later. The copy's Unreleased: line is a release line, so IndexNow does not count its removal as a change, and routes.json names it so the llms shards are cut alike from a dated dump and an undated one. The two landing pages took their text from one module, src/data/home.ts, and date-sources.json dated both by every commit to it, so a commit that changed only the Spanish text dated the English page as well and would have shown it the notice. Each language's text is now a module of its own, home-en.ts and home-es.ts, with home.ts keeping the shape and what both pages show alike, and each landing page is dated by its own. A test on the site's own entries edits the Spanish text alone after a release and finds only the Spanish landing page newer.
…ads every retired page's Markdown copy and every retired shard name to where it went, and a docs run dispatched by hand can send IndexNow every URL once The llms shards were renamed as they grew: an area past the budget of one fetch was cut into -1 and -2 parts and its own name answered 404 from the next deploy on. Fifteen names went that way after the site moved to docs.jmrp.io, and llms-start.txt was about to follow. site/src/data/llms-shards.json now records every shard name the site has published. generate_llms.py writes each of them again: an area published whole that comes to need parts keeps its own name as a short file listing the parts (llms-start.txt today), and an area published in parts is cut into at least as many again. A recorded name a run cannot write stops it, unless the record redirects it to a file the run writes; the fifteen names retired before the record are redirected that way, each to its first part. make llms adds the names it writes to the record, and both drift gates hold it to the pages. site/scripts/cloudflare-redirects.mjs prints those fifteen redirects too, and one item per retired page for its Markdown copy, <address>index.md, to the copy of the page it moved to (a Spanish API page to its English twin's copy, the only one there is). check-redirects holds the list to the map and, on a build, every copy and shard it leads to to a file the build wrote. docs.yml takes an indexnow_full input on workflow_dispatch. The digest IndexNow is compared by leaves the release lines out, so a release that takes back something every page said reaches the engines through no page; a run dispatched with the input sends every URL of the sitemap once, with the removed pages, and keeps its record as any deploy does. CONTRIBUTING says when to run it.
…ge that imports the notice's link or rule starts a git of its own src/lib/release-commit.mjs started git and also gave out the changelog link, the tag name and the rule a page applies to the release, which the notice, the Markdown copy header and the copy writer import. check-render-spawns fails a file that starts a process and gives a page anything but its read once per build, and it was not listed. The link, the tag name, VERSION_FILE, the Release type and changedAfter move to src/lib/release-rules.mjs, which starts no process. release-commit.mjs gives out loadRelease alone and is listed in ONCE_PER_BUILD with its one call; a test holds the read to one per site root and version, keeping the first answer when the tag moves after it, and F66 names that test.
… in its footer and in its Markdown copy, so a second change on one day is no longer served as unmodified The edge serves a page's Last-Modified from the last <time datetime> of its footer and a copy's from its Updated: line. The copy header wrote the day alone, and so did the footer of a page published and last changed on the same day, whose one date carried the day of the commit that added it. Read as midnight, a second change on that day answered an If-Modified-Since revalidation with 304. The copy's Updated: line now writes the moment the footer carries, YYYY-MM-DDTHH:MM:SSZ (isoInstant in src/lib/date-rules.mjs). Every footer <time datetime> is a moment too: the publication date takes the time of the commit that added the file (firstPublishedAt in src/lib/page-dates.mjs), and the one date of a page published and changed on one day takes the later of its two moments. What readers see is the same day as before; the sitemap and the structured data keep their days. IndexNow hashes each copy with its Updated: line cut to the day it names (withUpdatedDay in src/lib/page-markdown.mjs), so the new form reports no page as changed: the manifest of a build with it equals the one before it, digest for digest. generate_llms.py measures the Updated: line in its new shape, so a dated dump and an undated one still cut every shard at the same places, and llms.txt is unchanged. check-markdown-twins holds Updated: to a moment, on the day of the sitemap's <lastmod>, equal to the footer's last <time datetime>; check-crawl fails a page whose footer ends in a day with no time.
|
Sorry @jmrplens, your pull request is larger than the review limit of 150,000 diff characters |
|
Important Review skippedToo many files! This PR contains 111 files, which is 11 over the limit of 100. To get a review, reduce the PR to 100 files or fewer by splitting it into smaller PRs or changing its base branch. Upgrade to a paid plan to raise the limit.
|
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What and why
The documentation site after the 4.0.0 release: page dates that only move when the text does, a notice on the pages that changed after the release, structured data that names who made each dataset, Environments pages that say when their commands last ran, and llms shard names that keep answering once published.
What is now true
Page dates. Every frontmatter key a page can carry is classed in
site/src/lib/frontmatter-keys.mjsas content, label (title, description, sidebar, topic) or metadata (corrects,searchTitle,searchWeight,pagefind). A commit that changes only label or metadata keys moves no page's date in the footer, indateModifiedor in the sitemap. A key in no class stops the build by name, with the page and the commit, and the content collection refuses a schema key the list does not class. Keys are read as YAML reads them, bare, dotted or quoted. On today's history the 124 errata pages that took 14 November 2026 from the commit that addedcorrectsgo back to the dates of their last real change (121 to 28 October, 3 to 2 November).Conformance stamp. The report says "Generated for phonometry 4.0.0 from commit 39f1973, on 17 November 2026" (and the same in Spanish), which is what the line measures.
Release notice. A page whose content changed after the commit the latest release was published from shows a short note under its title linking the Unreleased section of the changelog, and its Markdown copy carries an
Unreleased:header line. The release is the commitv<VERSION>names, or the commit that setVERSIONwhen the tag is not pushed yet, so the next release takes every notice off with nothing typed for it. The landing pages' text is now one module per language (home-en.ts,home-es.ts), so a Spanish-only edit no longer dates the English page. The release is read with git once per build insrc/lib/release-commit.mjs, which gives out nothing else; the rule pages apply and the changelog link live insrc/lib/release-rules.mjs, which starts no process.Structured data and text. The Spanish landing page claims CNOSSOS-EU with the same scope as the English one (Annex II, 2.2 and 2.3), and
check-i18n-parity.mjsholds the two landing pages' designations to the same scope area by area. The 48 catalogue Datasets name the site's Person as creator and publisher; the two conformance report Datasets name each other as translations. Every work in the bibliography names its publishing body by one identity, ECAC and EASA included. Four answer-first guides say "in Python" in their descriptions, and llms.txt describes Bibliography and About with their own descriptions.Environments pages. Each page, in both languages, opens with the day its commands were last run (17 November 2026, phonometry 4.0.0 from PyPI) and the versions of every tool and image they ran on.
tests/environment_checks.jsonrecords that run andtests/test_environment_checks.pyholds every page to it both ways.Stable shard names.
site/src/data/llms-shards.jsonrecords every llms shard name the site has published.generate_llms.pywrites each of them again: an area published whole that now needs parts keeps its name as a short listing of its parts (llms-start.txttoday), and an area published in parts is cut into at least as many. The fifteen names retired before the record redirect to their first part.cloudflare-redirects.mjsprints those redirects and one per retired page for its Markdown copy, andcheck-redirects.mjsholds the list to the map and, on a build, to files the build wrote.Moments in the footer and the copy. Each page's footer dates its last change to the second (
<time datetime="2026-11-16T21:36:13Z">), and theUpdated:line of its Markdown copy states the same moment, so a second change on one day moves the copy's Last-Modified instead of answering as unmodified.check-markdown-twins.mjsholds the copy's moment to the footer's and its day to the sitemap, andcheck-crawl.mjsrefuses a footer whose last date is a day with no time. The sitemap and the structured data keep days, and IndexNow hashes copies by their day, so no page reads as changed for this alone.IndexNow.
docs.ymltakes anindexnow_fullinput onworkflow_dispatch: a run dispatched with it sends every URL of the sitemap once, with the removed pages. CONTRIBUTING says when to use it.Validation
No new computation in the library. The site and its checks:
ruff check .,ruff format --check .,mypy src scripts stub/src tests/static_typing(781 files, no issues),zizmor(no findings), and the prose guards (em dashes, Spanish accents, decimal comma, digit grouping, span and fence readers, quoted extras, conformance claims and vocabulary).llms.txt,docs/andsite/src/data/llms-shards.jsonunchanged, and the builtllms.txtis the same file.test:scripts(744 passed),test:agent-md(290 passed),check:types(0 errors, 0 warnings, 0 hints),check:render-spawns,check:i18n,check:markdown-writers,check:icons,check:guides, the full build with its warning check, and after itcheck:markdown-twins, the restated snippets, the generated conformance claims,check:redirects,html-validate,check:search,check:cited-standards,check:structured-data,check:schema-vocabulary,check:crawl,check:descriptionsandcheck:page-size.astro sync, the landing pages split by a fixture that edits only the Spanish text, the shard record against the old generator (llms-start.txtmissing), the edge list against the old script (copies missing), the full IndexNow list against the old diff, and the footer and copy moments against the old writers.