Optimize EBML encoding

[catamorphism]

"optionally perform 'relaxations' on end_tag to more efficiently encode sizes; this is a fixed point iteration" (from std::ebml)

[pnkfelix]

(This note doesn't really belong here, but it was what got me on the subject during via bug triage.)

By the way, I don't know which EBML specification we claim to adhere to, but the link at the top of ebml.rs is to the v1.0 RFC for EBML (dating from 2004), which is not only typo-ridden and (IMO) difficult to read, but also, most importantly, way way out of date, as described on this matroska-devel email from September 2012.

Unfortunately, that email post did not actually provide a recommendation for which spec to use.

• Here's one option: http://ebml.sourceforge.net/specs/
• This pdf seems like it might provide an accurate discussion of EBML; it was last updated in 2009, which is five years better than the v1.0 RFC. But it also is broader than EBML, since it covers the whole Matroska spec.

(Or maybe it does not matter what variant of the spec we use; the more recent specs seem quite tied to Matroska specifics, while the v1.0 spec explicitly says it should not be used for an implementation.)

[pnkfelix]

visiting for bug triage, email from 2013-08-26.

nothing to add, though I still wonder about my spec question from two months ago.

[nikomatsakis]

I tend to think there is no reason to use EBML at all. I would prefer to just shift to a very compact encoding.

[pnkfelix]

Accepted for P-low.

[esummers]

EBML (and similar formats: msgpack, protocol buffers, etc) are nice, but if there is a move away from EBML it might be interesting to look at something similar to what Cap'n Proto uses. Everything is aligned and unpacked (in some cases maybe just a memcpy) and a second pass uses a simple compression algorithm to remove the zeros. http://kentonv.github.io/capnproto/encoding.html

[steveklabnik]

triage: removing libs, since we aren't exposing this publicly.

[lattice0]

What is std::ebml?

[steveklabnik]

@LucasZanella this bug is from 2012. It's something that existed then, but does not exist now.

[lattice0]

@LucasZanella this bug is from 2012. It's something that existed then, but does not exist now.

I was searching for ebml libs on Rust. Was this one?