The specification was a bit involved about that topic. Hopefully this
reformatting text improves the readability.
This commit is based on a different patch by Florian Schmaus and
hopefully further improves clarity by spelling out all cases clearly
in a list, as well as more thoroughly defining corner cases.
Plus (after ML discussion):
- Clarify intention of 'with' field rule
- Add 'ids' field for fetching specific messages
- Add archive metadata query
- relax language about inclusion of RSM count/index
These were always optional, but the XEP encouraged their use.
Implementation experience has shown that generating the count
has significant performance cost, while most/all current clients
ignore it anyway.
Incorporate feedback from OMEMO sprint
Improvements to wording and further example of exceptional elements
Content of rpad does not need to be limitted to base64
Credit where credit is due
Use author shortcut
Limit store hint rules MUST to chat messages only
Add revision block
Remove <validate/> from the registrar submission and instead state the
requirement that 'status-addresses' form field values must be of type
xs:anyURI in the text.
The reference to RFC 6120 was incorrect, what this really meant is RFC
6415. But instead of simply s/RFC 6120/RFC 6415/ here, I decided to
reorganize stating the requirements of XRD and JRD a little.
The result IQ's <query/> child element does not need to carry the node
information. And in this case the node information was actually wrong,
since "documents/test2.txt" was queried, and not "test2.txt".
This pipeline features the following:
- Building of an nginx image with the XEPs as static files,
in all formats.
- Incremental builds on the main branch and incremental builds
for MRs based on the last main build.
- Automatic archiving of changed XEPs to the attic
- Automatic announcement to the mailing lists
If it is known that the documents have already been built or if it
is imperative to use the versions built even if local changes have
been applied since the last build, this switch comes in handy.
Most jobs depend on build or one of its subdirectories. By default,
this causes make to take the timestamp of the `build` directory (or
the respective subdirectory) into account when calculating whether
a job needs rebuilding.
This is a problem, because the modified timestamp of `build` updates
whenever a file is put into it. Effectively, this breaks incremental
builds.
Luckily, GNU(?) Make supports Order-only Dependencies, prefixed with
a pipe (`|`) symbol in the dependency list. That means that the
dependencies are not taken into account for freshness checks, but
will be built before the target (if they are non-fresh).
This commit introduces usage of Order-only Dependencies for the
output directories, which fixes incremental building.
