Introduce a new base module
#474
Conversation
It implements 'Hash' for the provided integer types.
'ParseBytesByRef' now requires all implementing types to be unaligned.
Otherwise, padding bytes wouldn't be accounted for properly, e.g. in
// Has alignment of largest field: 8 bytes
#[repr(C)]
pub struct Foo {
a: u8,
// 7 bytes of padding here
b: u64,
}
The 'derive' can't tell how much padding to use, so it would parse a
'[u8; 9]' as a valid instance of 'Foo'. Every 'repr(C)' would have to
use 'repr(packed)' too.
|
|
||
| //--- Impls for 'Option<T>' | ||
|
|
||
| impl<T: ParseQuestion> ParseQuestion for Option<T> { |
There was a problem hiding this comment.
I don't get this one. Is this to say that the question might not match but it will still fail if no questions are given?
There was a problem hiding this comment.
In any case, it needs docs
There was a problem hiding this comment.
Oh I see it just maps Some over the result? When is None returned?
There was a problem hiding this comment.
None is returned if the message was correctly formatted, but no matching T was found in it (e.g. if T is specifically an A record).
| } | ||
|
|
||
| /// A type that can be constructed by parsing zero or more DNS questions. | ||
| pub trait ParseQuestions: Sized { |
There was a problem hiding this comment.
I'd love a different name for this to differentiate with ParseQuestion. Maybe ParseManyQuestions.
|
|
||
| //--- Parsing from bytes | ||
|
|
||
| impl<'a, N, D> SplitBytes<'a> for Record<N, D> |
There was a problem hiding this comment.
Hmm this is mostly the same as SplitFromMessage can't we implement that one in terms of this?
There was a problem hiding this comment.
Unfortunately, both traits are necessary. SplitFromMessage gets used when parsing DNS messages from the wire. SplitBytes is also used when parsing from unknown record data (e.g. encoded in a zone file), where there isn't a containing Message and name compression is not allowed. I do want to add derives to make some of this easier.
This ended up collecting a lot of small changes as I tried to get
things to compile.
- All the bytes parsing/building traits have been moved to 'wire'.
- The 'wire::ints' module replaces 'U16' and 'U32' from 'zerocopy'.
- '{Parse,Split}BytesByRef' now support parsing from '&mut'.
- Every derive macro is documented under a re-export in 'wire'.
The remaining contents of 'new_base::{build, parse}' might get moved
into a shared 'message' module at some point. We'll see.
Instead of passing the input selection as a range for parsing, the whole message is cut (using 'Message::slice_to()') and only the start is indicated. This ensures that we never cross the end of the range. It also implicitly dictates that compressed names are not allowed to reference future locations in messages. In addition, both the parsing traits now use offsets into the message _contents_ rather than the whole message. They can avoid 'as_bytes()' everywhere and have better guarantees of success. It also ensures the message header can never be selected for parsing.
This introduces a
new_basemodule as a potential candidate for replacingbase. While it has a similar structure (although much is left to be filled out), it has three distinguishing features:It builds the foundation for a
derive-based interface for easily and efficiently parsing (and possibly building) DNS messages.It introduces
RevName, a more efficient type for domain names (where labels are stored in reverse order).There's a lot more work to be done:
Note: this is a new version of #469 with a less problematic branch name.