sgodwincs / rtp-rs Goto Github PK

Need to see what common usage is here. Should just allocate room for all 15 CSrcs up front, allocate nothing (equivalent to just using Vec), or somewhere in between?

Ideally, PayloadType would be defined something like:

type PayloadType: TryFrom<u8, Error: Error + Into<HeaderDecodeError<Self>>>;

This can be done now with associated type bound, but it results in a lot of unnecessary where clauses, so this is blocked on implied bounds.

I generally wrap a lot of data types as unit types. But it's quite common to need to expose an actual integer as a range, such as in error details.

Going to block this on at least some const value implementation with LOW < HIGH bounds.

The current setup isn't ideal in my opinion. Ideally, enum variants as types would be supported which would allow a single enum with separate trait implementations per variant. This is particularly useful because it allows a different associated type per variant. With this approach, the header and packets would take an enum variant as a type parameter. Also, an enum is more appropriate as the set of profiles is closed.

Blocked on having enum variants as types (there isn't even an accept RFC yet)

#1 should be done first.

This would remove the Into<HeaderDecodeError<Self> from PayloadTypeDecodeError and instead add a where clause on the trait:

where HeaderDecodeError<Self>: From<<Self::PayloadType as TryFrom<u8>>::Error>

This can be done now, but it results in a lot of unnecessary where clauses, so this is blocked on implied bounds.

I don't have a great understanding of how timestamps work with RTP yet, but it seems that usage is very dependent on either the profile or the payload type being used.

Based on the above, it seems we need to type it based off of the profile being used, but it's also apparently possible to specify it dynamically through non-RTP means.