| Age | Commit message (Collapse) | Author |
|---|
|
(#2108) (#2109)
On a fresh (no-consent) load the consent banner could render completely
unstyled and expand to full page height. Cookiebot delivers its dialog CSS as
a constructed stylesheet on document.adoptedStyleSheets; ReadTheDocs'
readthedocs-addons.js reassigns that whole-array property with a destructive
replace, and when that lands after Cookiebot's network-gated adoption it drops
Cookiebot's sheet, leaving #CybotCookiebotDialog at position:static expanded to
~6700px. It is a race, so it is intermittent and clears once consent is given
(the dialog then never renders).
- layout.html: a small adoptedStyleSheets shim, installed before any page
script, that preserves Cookiebot's sheet so another library cannot drop it.
- custom.css: a leak-safe safety net that keeps the dialog contained even if
its adopted sheet is ever absent. It uses only properties Cookiebot itself
sets, so it has verifiably no effect on the normal, styled banner.
π€ Generated by [robots](https://vyos.io)
(cherry picked from commit dce7bb521e3947af59fe18895b609ec353e0450e)
Co-authored-by: Yuriy Andamasov <yuriy@vyos.io>
|
|
π€ Generated by [robots](https://vyos.io)
|
|
Read COOKIEBOT_ID from the environment via conf.py html_context,
replacing the hardcoded data-cbid value in layout.html.
Guards tightened to {% if gtm_id and cookiebot_id %} so neither
GTM nor Cookiebot loads unless both IDs are present.
π€ Generated by [robots](https://vyos.io)
|
|
Prevents broken googletagmanager.com requests with empty container IDs
when GTM_ID is not set (local builds, preview environments).
π€ Generated by [robots](https://vyos.io)
|
|
Move the Google Tag Manager container ID out of the template and into
conf.py via os.environ.get('GTM_ID', ''), exposed to Jinja via
html_context. Set GTM_ID in the RTD project environment variables.
π€ Generated by [robots](https://vyos.io)
|
|
π€ Generated by [robots](https://vyos.io)
|
|
The `<noscript>` GTM iframe loads via a direct HTTP request to
googletagmanager.com/ns.html and bypasses both Cookiebot's auto-blocker
and Google Consent Mode v2 entirely β Consent Mode is JS-only
(`gtag('consent', 'default')` and the update API), so a noscript
request can't carry consent state. Under strict prior-consent (GDPR),
JS-disabled visitors get no analytics rather than unconsented analytics.
Addresses CodeRabbit review feedback on PR #1872.
\xf0\x9f\xa4\x96 Generated by [robots](https://vyos.io)
|
|
Make the external dependency explicit in the template comment: Cookiebot's
`uc.js` with `data-blockingmode="auto"` automatically pushes
`gtag('consent','update', β¦)` into the dataLayer when the visitor
accepts/declines categories. No in-repo update logic is needed; category
mapping happens in the Cookiebot admin UI.
Addresses Copilot review feedback on PR #1872.
\xf0\x9f\xa4\x96 Generated by [robots](https://vyos.io)
|
|
Agent-Logs-Url: https://github.com/vyos/vyos-documentation/sessions/ebb95457-26ce-4636-b018-fe7b8e31778c
Co-authored-by: andamasov <12631358+andamasov@users.noreply.github.com>
|
|
Inject Cookiebot CMP, Google Consent Mode v2 defaults, and Google Tag
Manager into the Sphinx layout. Cookiebot loads first so its
auto-blocker can scan and block other tags before they execute;
Consent Mode v2 starts denied for all categories except security_storage
and updates only after Cookiebot fires the consent update.
π€ Generated by [robots](https://vyos.io)
|
|
Per the [llms.txt spec](https://llmstxt.org/), `/llms.txt` links should
point at markdown files (the spec example: `[API docs](https://example.com/api.md)`).
We were using `.html` URLs, which still work but force LLM consumers
through HTML parsing for content that's already available as raw MyST
at the same path (the `.md` is copied to HTML root by
`_copy_md_sources`).
Pivot the curated llms.txt template to `.md` URLs across all 22
entries. Spec-compliant, no extra requests, LLMs get raw markdown
without the navigation chrome.
Verified locally: `/en/<ver>/<page>.md` returns 200 with raw MyST
content (frontmatter + body) for all linked paths on this branch.
Same pivot applied symmetrically across [#1874](https://github.com/vyos/vyos-documentation/pull/1874) (current),
[#1876](https://github.com/vyos/vyos-documentation/pull/1876) (sagitta), and [#1903](https://github.com/vyos/vyos-documentation/pull/1903) (circinus).
π€ Generated by [robots](https://vyos.io)
|
|
Parity with [#1874](https://github.com/vyos/vyos-documentation/pull/1874) (current) and [#1876](https://github.com/vyos/vyos-documentation/pull/1876)
(sagitta). Replaces the static `docs/_html_extra/llms.txt` with a Jinja
template (`docs/_templates/llms.txt.j2`) rendered at build time via a
`build-finished` hook in `conf.py`. URLs and the version line are
interpolated from `html_baseurl` and `release`, so the rendered file
always matches what circinus actually serves.
Hook behavior matches `current`/`sagitta` exactly, including the two
review-cycle hardenings landed on those PRs:
- Skips on builders other than `html` and `readthedocs` β the curated
template encodes `.html` URLs which don't match `dirhtml` output.
- `RuntimeError` if `html_baseurl` is empty rather than silently
rendering broken root-relative URLs.
- Jinja `Environment(undefined=StrictUndefined)` so a typo in
`llms.txt.j2` raises at build time instead of writing a half-blank
`llms.txt`.
The closed PR [#1875](https://github.com/vyos/vyos-documentation/pull/1875) was the previous attempt; this re-lands the change after the
post-flip rebase situation cleared.
Verified locally: hook tests (happy-path render, fail-fast on
`html_baseurl=None`/`''`, `StrictUndefined` catches `{{ relase }}`
typo, skips `dirhtml`, skips on build exception) pass against this
worktree.
\xf0\x9f\xa4\x96 Generated by [robots](https://vyos.io)
|
|
|
|
|
|
|
|
Per https://blog.vyos.io/vyos-1.3.0-lts-release
|
|
|
|
|
|
|
