Versions are actually on a scale of recommendations, and are expected to be created as needed. The scale presented here (develop/beta/default/recommended/mandatory) is a more wordy version of what was previously discussed/intended for room versions - the labels aren't final and may be changed.
The "Room Specification" (or "Room Version Specification") is the specification that defines which room versions do what and are intended to be documents which speak the truth about how rooms operate under the hood.
The approach taken here is a bit different than other specifications. For starters, the specification is versioned in this project instead of relying on the matrix.org repository to track compiled HTML. This is done for a couple reasons, the first being we're still developing the v1 specification while concurrently making a v2 spec and the second being trying to reduce the reliance on matrix.org's repository for specifications.
Because the room spec is built into versions, some changes needed to be made. The `targets.yaml` now has a special syntax for indicating what version something is at, and the changelog generator can handle rendering different versions of the same changelog (as parsed from the RST). Some additional work has been put in to the changelog parsing to allow us to reference the v1 room spec as "v1" without having to sacrifice clarity in the changelog headings.
Finally, this moves the state resolution algorithms into the versioned
spec as a result of MSC1759 (https://github.com/matrix-org/matrix-doc/pull/1759).
Note: this does not introduce the concept of versioned schemas (tabs) that I was previously working with. There's currently no use for them, so they are shelved elsewhere.
Fixes https://github.com/matrix-org/matrix-doc/issues/1499
This is done by using magic variables in the RST. The magic variables are generated based on the substitutions available, making them available for use at build-time.
Magic variables were chosen because it allows people to continue working on the spec and release process without having to worry about removing a chunk of text from the top of the file. Originally, this was attempted by using jinja2 if-statements, however the substitutions are replaced *after* the template is executed, so the condition would never match.
The format of the variable is to make the templating happy. Using colons or percent signs results in the templator thinking something else is going on, and then complaining about format.
Because this is the first release, it has several moving parts to it:
* The version variables have been defined.
* The towncrier changelog has been prepared for future modifications.
* The templating has been updated to better support future versions of the specification.
* A release process document has been created.
It's worth noting that Synapse does not make use of the `poll` or `unpoll` fields, and therefore the wording has been updated to permit servers to reject users. In the case of synapse, it would automatically reject everyone in the list by nature of ignoring it.
This is useful for when we start defining event schemas. This also has a sanity check for ensuring the directory exists, allowing the IS and push API paths to be uncommented.
There's two kinds of transactions currently: one with EDUs and one without. The one with EDUs is only used on /send, however the schema is still somewhat worth splitting out for simplicity.
The examples are brought apart to make them slightly more reusable for when they get dumped into the relevant sections of the spec (see TODO in server_server_api.rst)
Further, the Transactions stuff introduces tuples to the spec. The units.py has been updated to support this.
The changelog shows up via stdout, everything else via stderr. We dump as much information as we can into the changelog to make debugging errors easier.
Travis Ralston