Change the mapping in Prefixes to Prefix -> Namespace instead of
Namespace -> Prefix. This allows us to not have duplicate prefixes
anymore, but requires us to store the prefix on Element. This prefix is
only taken as a hint anyway and used only when coming from the reader.
This commits also partially removes the possibility to add prefixes
when creating an Element via `Element::new`, `Element::builder` or
`Element::bare`. Proper errors should be added in the following commits.
Signed-off-by: Maxime “pep” Buquet <pep@bouah.net>
I would have liked to handle all of this in `Element::new` only, but I
also have to do it in `Element::builder` unfortunately because then
element builder then pushes prefixes it gathered itself.
Signed-off-by: Maxime “pep” Buquet <pep@bouah.net>
Below is what I think I did.
A few changes:
- Change prefixes to be something less important in the API.
- Rework the Element struct to force a namespace. In XMPP everything is
namespaced.
- Remove parent ref on what was previously NamespaceSet and is now
Prefixes.
More specifically this means `Element::new` has changed to require
`Element`'s new new properties as parameters. `Element::builder` and
`Element::bare` now require a namespace unconditionally.
`Element::prefix` has been removed.
This new API is based on the fact that prefixes are non-essential
(really just an implementation detail) and shouldn't be visible to the
user. It is possible nonetheless to set custom prefixes for
compatibility reasons with `ElementBuilder::prefix`. **A prefix is
firstly mapped to a namespace, and then attached to an element**, there
cannot be a prefix without a namespace.
Prefix inheritance is used if possible but for the case with no
prefix ("xmlns") to be reused, we only check custom prefixes declared on
the tag itself and not ascendants. If it's already used then we generate
prefixes (ns0, ns1, ..) checking on what has been declared on all
ascendants (plus of course those already set on the current tag).
Example API:
```rust
let mut elem = ElementBuilder("stream", "http://etherx.jabber.org/streams")
.prefix(Some(String::from("stream")), "http://etherx.jabber.org/streams)
.prefix(None, "jabber:client")
.attr(..)
.build();
assert_eq!(elem.ns(), String::from("http://etherx.jabber.org/streams"));
```
See also the few tests added in src/tests.
TODO: Fix inconsistencies wrt. "prefix:name" format provided as a name
when creating an Element with `Element::new` or `Element::bare`.
`Element::builder` already handles this as it should, splitting name and
prefix.
TODO: Change `Element::name` method to `Element::local_name` to make it
more explicit.
Signed-off-by: Maxime “pep” Buquet <pep@bouah.net>
Although it is still possible to create such elements, this is not a a
case that should happen in XMPP. Changing to API to prevent the creation
of these elements is next on the list.
Signed-off-by: Maxime “pep” Buquet <pep@bouah.net>
And install it as the container doesn't have it. Fortunately it's rather
quick (few seconds).
The previous nightly container apparently dropped rustfmt, which is
understandable as not all tools are available all the time. That's on
us.
Signed-off-by: Maxime “pep” Buquet <pep@bouah.net>
This bug was introduced by 2e97f4de2e, to
fix another bug where the parser would choke on whitespace.
The bug would manifest whenever a stanza was sent in different parts,
for example:
<< "<message "
<< "type='chat><body>foo</body></message>"
Would produce the following once parsed:
`<messagetype='chat'><body>foo</body></messagetype='chat'>`
This commit ensures this doesn't happen anymore (by not trimming
whitespaces before feeding the parser), and also ensures that
whitespaces are now handled at the correct layer.
The removal of xmpp_codec::test_lone_whitespace only happens because I'm
not sure if it's supposed to be here anymore. Maybe it should be at a
different layer? Or written differently?
Signed-off-by: Maxime “pep” Buquet <pep@bouah.net>
