It was a change I did not notice when upgrading Docsy from 0.8.0 to 0.11.0. Docsy changed the way heading self links are generated: they used to be rendered with JS and now they use a Hugo render hook.
This means two things:
- We need to enable them explicitly by overriding the `_default/_markup/render-heading.html` template.
- We need to add the self heading ourselves to headings that are not rendered by Hugo, i.e. HTML headings that we create ourselves.
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Use json instead of json5 for syntax highlighting
Chroma, the library used for syntax highlighting in Hugo, does not support JSON5 so those code blocks were not highlighted.
However it supports comments in JSON so they are highlighted correctly in the rendered spec.
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Add changelog
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
---------
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
When used with a text that includes multiple paragraphs, the partial created invalid HTML by nesting `<p>` elements.
It also changed the rendering by making "Changed in vX.XX:" a separate paragraph, when it is inline with a single paragraph.
To change that we do as with "Required" and add "Changed in vX.XX:" to the text before it is rendered, making it inline with the first paragraph.
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
If the description is not set in the object definition, Hugo generates a weird string after "Required": `%!s(<nil>)`.
To avoid that, we default the description to an empty string when it is not set.
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
According to the W3C's HTML validator, trailing slashes in void-element have no effect,
and might interact badly in some cases.
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
The version of Inter is updated to use a variable font, which is supported widely among modern browsers.
Using `display: swap` means that the browser will render the text of the spec even before Inter is loaded, making the website appear to load faster on mobile.
Allow the browser to use the local Inter font if it exists.
The first commit allows to lazy-load the diagrams, which should improve the loading time of the CS API on mobile. In the process it also improves the alt text of the images.
The second commit serves the diagrams as high-resolution WebPs. Encoding a high resolution diagram as WebP gives a file of approximately the same size as the lower resolution PNG. For maximum compatibility we also serve them as a lower resolution WebP and a fallback PNG. WebP was chosen because it is one of the export formats of draw.io/diagrams.net, and it is widely available in modern browsers.
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
The code relied on an IntersectionOberver, so the ToC was only updated when a heading was in the viewport.
It meant that if we jumped to a part of the text that has no heading, the ToC would still point to the old entry.
The new code looks for the correct heading when the view is scrolled so the correct entry is always selected.
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
While 1405184765 redefined most fields to
be the appropriate object type, it missed one of them.
Signed-off-by: Nicolas Werner <nicolas.werner@hotmail.de>
The `<>` delimiters are not necessary for the shortcode to be rendered inline, and in some cases they break some expectations: a shortcode that is separated from other text to be in its own paragraph is not actually wrapped by a `p` element, breaking the spacing between paragraphs.
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
Docsy does not set the `height` anymore, but the `min-height`, which doesn't seem to affect scroll-anchoring.
See f22a70ec56
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
Use `p` elements to separate paragraphs instead of `br` and enforce single paragraphs to be wrapped in `p` for consistency.
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Rename .htmltest.yaml to .htmltest.yml
This is almost nothing but htmltest expects this exact name by default, so when running the
binary locally, it avoids to have to bother with providing the name of the config file.
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Add changelog
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
---------
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
I did a quick search of the "deprecated" word in the data folder and set the
field where the description says that a property is deprecated.
This does not change the rendering of the spec because the
descriptions already talk about the deprecation,
but it can be used by tools that rely on the OpenAPI definitions and JSON Schemas.
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
This updates the sign-off requirements to match what most other
matrix-org and element-hq repos already have. The change was first made
in synapse: https://github.com/matrix-org/synapse/pull/3467
Signed-off-by: Tulir Asokan <tulir@maunium.net>
* Fix ToC for room versions pages
Like for the cs-module shortcode, use .RenderShortcodes
instead of .Content for the rver-fragment shortcode,
so the headings are detected by Hugo.
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Change the way "this version" is detected in added-in and changed-in shortcodes
Now that we use .RenderShortcodes in the rver-fragment shortcode,
we cannot remove the output of these shortcodes dynamically
because they are replaced by a temporary placeholder due to Hugo's internals.
Instead, since the `this` parameter was only used for room version,
we always use the `v` parameter and compare with the version
provided in the page's front matter.
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Add changelog
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Add version front matter for v11
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Update changelogs/room_versions/newsfragments/1884.clarification
---------
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
Co-authored-by: Travis Ralston <travpc@gmail.com>
* Allow to specify a prefix for generated HTML IDs of API endpoints
Allows to deduplicate IDs of duplicate endpoints
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Add changelog
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
---------
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Include method in all API endpoint children's IDs
Avoids duplicate IDs for object of endpoints
that use the same path but a different method.
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Differentiate API endpoints' request and response children's IDs
Ensures that the objects have a unique ID compared to other parts of the endpoint.
Mostly useful for the Error type that can be used for responses with different status codes.
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Differentiate the names of both SessionData formats
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Add changelog
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
---------
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Render added/changed in info on request and response content types
Fixes: #1774
Signed-off-by: Johannes Marbach <n0-0ne+github@mailbox.org>
Co-authored-by: Kévin Commaille <76261501+zecakeh@users.noreply.github.com>
* Add tr as child of thead in HTML tables
It is invalid HTML for th to be the direct children of thead
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Remove unnecessary HTML code end tag
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Avoid nesting p HTML elements
A p HTML element cannot contain other block elements,
so the "parent" element is closed when the first "child" one is opened.
We need to use Page.RenderString with options
to force Hugo to keep the wrapping p elements
even if the content contains a single paragraph.
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Add missing HTML details end tags
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Replace HTML a self-closing tag with start and end tags
The a element start and end tags are mandatory.
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Replace obsolete HTML name attribute with id
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Add changelog
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Update release documentation (Q2 2024 edition)
* changelog
* Drop the ranges we don't follow
* Don't discourage maintenance
* Patch releases just aren't a good idea
* C2S: Deprecate now-legacy endpoints
* C2S: Fix MXC URI code block while we're here
* C2S: Describe the authentication and deprecation requirements
* C2S: Intro the upload/download endpoints differently
* C2S: Literally copy/paste the `content-repo.yaml` spec
* C2S: Drop `/upload` and `/create` because we aren't replacing them today
* C2S: Fix notes while we're here
* C2S: Update metadata for new endpoints
* C2S: Add authentication to new endpoints
* C2S: Drop `allow_remote` and `allow_redirect` on new endpoints
* C2S: Append backwards compatibility notes
* C2S: Decorate old media endpoints with pointers to the new ones
The server-server spec might have a harder time linking to these, but that can be fixed with verbiage.
* C2S: Annotate IdP icon spec with media auth implications
* S2S: Modernize section text
* S2S: Create content repository API
This is largely a copy/paste of the new authed content repo API in the Client-Server API, though some keywords (like "client") have been changed. Paths and response formats have also been changed to support the federation-specific requirements.
* C2S & S2S: Add plethora of changelogs
* Reference RFC 1341
* Upgrade keywords in changed text
* Mention caching
* Cross-reference IdP icons
* Update content/client-server-api/modules/content_repo.md
As far as I can tell, these header files only encourage people to create
badly-formatted PRs.
Also we only have one template so let's give it the default name.
Also, some other editorial improvements, including factoring out our two definitions of the same key encoding algorithm.
Co-authored-by: Travis Ralston <travisr@matrix.org>
This was commented prior to the
port to OpenAPI 3.1 for technical reasons (#1127).
Now we can use it just fine.
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
Previously, titles would appear that do not link to a subchema definition.
It would also mean that named subschemas would appear without being clearly referenced.
Now, the type clearly shows the nesting of objects
and subschema definitions should be clearly referenced.
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
Otherwise the version might change depending on the runner.
We just use the same version as other jobs.
This removes a GitHub warning.
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
Co-authored-by: Richard van der Hoff <1389908+richvdh@users.noreply.github.com>
* Upgrade version of Hugo used to build the spec in CI
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Escape HTML manually in property-type partial
The behavior of `delimit` changed,
so Hugo doesn't recognize "safe" HTML passed to it anymore, so it escapes nested HTML links.
To fix that we escape the schema data manually
and consider the output of the partial as "safe".
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Add changelog
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
---------
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
Makes it easier to use, like resolve-refs. It just needs to be called once.
Fixes an issue with m.call.* events not displaying the common fields
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
Co-authored-by: Richard van der Hoff <1389908+richvdh@users.noreply.github.com>
Since we already have three of these, and I'm about to add a fourth, let's pull
it out to a common definition.
We could, of course, keep defining the grammar each time it's used, but
defining it in an appendix helps us be consistent for future API design.
The split was not clear between property-type and type-or-title,
so it was not obvious which partial should be called for recursion.
That resulted in an error where type-or-title was only called for objects and array items, even if it also resolves
arrays of types.
This makes the split clearer. property-type must be called for any schema,
and object-type-or-title is only called for object schemas.
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Use the resolve-refs partial as soon as possible
Call it right after accessing the site.Data,
since it is recursing it will solve all references in the tree.
That way we don't need to wonder where to call it,
we trust the validators that the refs will be used in the right place.
* Enable strict $ref rule in OpenAPI validator
* Document use of $ref to compose examples
* Fix schema path in event-fields shortcode
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Fix `v` tag_name prefix sneaking into npm version
* Fix `yarn version` failing in CI due to no git global ident name
* Add changelog
* Rename 1765.misc to 1765.clarification
* Break out non-JSON request/response content types as tables
Currently we display this as a table like "image/png|image/jpeg" and description on a single line, but we're using a table. This breaks the join out to individual rows.
* changelog
It was not enabled before the docsy update and it messes with
the TOC highlighting during the transition.
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Deprecate the `font` HTML tag
Rationale:
MSC4077 allows to deprecate HTML tags
that are deprecated in the WHATWG standard,
if they can be replaced by tags with the same feature.
`font` is deprecated and can be replaced by `span`
with the `data-mx-bg-color` and `data-mx-color` attributes.
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Add changelog
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Add deprecation info box
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
---------
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Spec for MSC3981
This writes up https://github.com/matrix-org/matrix-spec-proposals/pull/3981
Hopefully this is relatively straightforward, apart from having to add
the parameters and response field in all three places. I tried to factor
these out but it seems references just aren't supported in the right
places currently (see https://github.com/matrix-org/matrix-spec/pull/1745
for my efforts). Path parameters can't be optional, so it can't be done
that way either.
* Missed schemas
* newsfile
* Actually it clearly isn't going to support markdown, is it?
* grammar
Co-authored-by: Richard van der Hoff <1389908+richvdh@users.noreply.github.com>
* grammar
Co-authored-by: Richard van der Hoff <1389908+richvdh@users.noreply.github.com>
* Clarity
Co-authored-by: Richard van der Hoff <1389908+richvdh@users.noreply.github.com>
* Clarity
Co-authored-by: Richard van der Hoff <1389908+richvdh@users.noreply.github.com>
* Typo
Co-authored-by: Richard van der Hoff <1389908+richvdh@users.noreply.github.com>
* More clarity.
Note this is counter what the MSC actually proposed to add, but
I think it's clear that this is what it meant.
---------
Co-authored-by: Richard van der Hoff <1389908+richvdh@users.noreply.github.com>
* clarification around implementation requirement, and mention new label
* add changelog
* fix typo
* Fix typos
Co-authored-by: Richard van der Hoff <1389908+richvdh@users.noreply.github.com>
---------
Co-authored-by: Richard van der Hoff <1389908+richvdh@users.noreply.github.com>
* Add missing 'in' in SSO specification
Signed-off-by: Johannes Marbach <n0-0ne+github@mailbox.org>
* Use standard changelog entry for typos
---------
Signed-off-by: Johannes Marbach <n0-0ne+github@mailbox.org>
Co-authored-by: Hubert Chathi <hubert@uhoreg.ca>
* Convert m.call.candidates schema to YAML
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Clarify that sdpMid and sdpMLineIndex are not required in `m.call.candidates`
MSC2746, merged in v1.17,
introduced the end-of-candidates candidate,
where only the `candidate` property is set to an empty string.
Besides, the [WebRTC specification](https://www.w3.org/TR/webrtc/)
says that only one of those fields is required in a normal candidate.
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Add changelog
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Link to the "End-of-candidates" section, and clarify what "empty" means
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
---------
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Disambiguate uses of PublicRoomsChunk
Make sure that different objects don't share the same title.
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Clarify that extra fields of space hierarchy children are not required
There is no `children_state` field,
the `room_type` is only set for spaces
and the description of `allowed_room_ids` says that the field can be omitted.
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Use `body` field as media caption
As per MSC2530.
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Add changelog
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Use `s` HTML tag in example
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* Move changed-in annotation
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
---------
Signed-off-by: Kévin Commaille <zecakeh@tedomum.fr>
* clarify otk and fallback key types in examples
- remove unsigned curve25519 keys from examples because we don't use those for
otks and fallback keys
- add missing `device_unused_fallback_key_types` property, which is required
* add changelog
* Clarify that the key backup MAC is implemented incorrectly
Due to a bug in libolm, all implementations of the
m.megolm_backup.v1.curve25519-aes-sha2 key backup algorithm incorrectly
pass an empty string through HMAC-SHA-256 to generate the `mac` property
of the `session_data`.
It was intended for the entire raw encrypted data to be passed through
HMAC-SHA-256, but the issue was caught too late in the process, and thus
we are stuck with this until a new key backup algorithm is introduced.
This commit clarifies the real-world behavior of all current
implementations.
Signed-off-by: Sumner Evans <sumner@beeper.com>
