Add <event>::borrow() and some minor stuff
#416
Conversation
…c links Compiler will check existence of the intradoc links. Cross-references in the Deserializer couldn't be converted to intradoc links, because trait is implemented for `&mut` and rustdoc cannot resolve that links
Why? This would be a bit unfortunate from a usability perspective, let's say you wanted to write a XML formatter and you just wanted to shuffle events from the reader directly over to a writer with a minimum amount of processing. |
Codecov Report
@@ Coverage Diff @@
## master #416 +/- ##
==========================================
- Coverage 49.60% 49.24% -0.37%
==========================================
Files 22 22
Lines 13967 13984 +17
==========================================
- Hits 6928 6886 -42
- Misses 7039 7098 +59
Flags with carried forward coverage won't be shown. Click here to find out more.
Continue to review full report at Codecov.
|
|
To solve lifetime issues, for example #332. Currently reader events can hypothetically contain an owned data, but in reality they never have. That, however, has negative implications on lifetimes. I'll plan to make a reader event to always borrow (from buffer for buffered reader or from input for a borrowing reader). Reader event will implement an You can wait with your opinion until I'll finish that work and make a PR for it |
src/de/mod.rs
Outdated
| Start(BytesStart::borrowed_name(b"inner")), | ||
| End(BytesEnd::borrowed(b"inner")), | ||
| End(BytesEnd::borrowed(b"inner")), | ||
| Start(QName(b"inner").into()), |
There was a problem hiding this comment.
I'm mildly negative about this kind of usage because it makes the API less consistent and IMO harder to follow.
That's not to say that the implementation shouldn't be there, because it makes sense to have the implementation, but in practice when used like this it flows awkwardly when some tags use QName and others use BytesText::from_*, BytesCdata::from_*, etc. Before you only needed to look at the second "column", afterwards you have to jump back and forth between the first column and the middlle to track what is happening.
Not a blocker, just FWIW
There was a problem hiding this comment.
I've made this change in order to reduce diff in one of following commits, that splits common Event into reader::Event and writer::Event. reader::Event will not have explicit public methods to construct a value (this is unnecessary for the user, if user need an event he could use writer::Event, reader events is read-only), so it will impossible to construct it in the doctests. I can move that change there
|
@Mingun Would you mind splitting the 7 "less interesting" commits off into a different PR, keeping 7280390 and b55aac6 here, and then then continuing your work on this PR? I feel like it would be helpful to evaluate them alongside the additional changes you have planned. The rest are totally straightforwards. |
|
I think I'll just strip out controversial commits from this PR -- it is simpler. |
Mingun
changed the title
From<QName> for Start and End events and more<event>::borrow() and some minor stuff
This PR makes some refactoring that I do on my path to the implementing namespaces support.
Currently I need to fix lifetime issues and I plan split reader and writer events into separate structs, which requires some preparing work. In order to not make that PR too big I made this PR with this work.
I fixed some intradoc links, move tests to appropriate place and introduced two improvements:
BytesStartandBytesEndstructs now implementsFrom<QName>that allows to create them easely (useful for writing)Writer::write_eventnow consumes event using anIntotrait.AsReftrait, used before, does not allow to accept two unrelated structs whichreader::Eventandwriter::Eventwill be. On the contrary,writer::Eventwill implementFrom<reader::Event>which would allow to write reader events as before. In order to not lose the ownership, theEvent::borrow()method is introduced, as a counterpart to theEvent::into_owned()method, soWriter::write_eventcould consume only borrowed event