Rust encourages users to use BufReader to accelerate small I/O operations on Read implementations that have high per-operation overhead. BufReader's Seek implementation however has a performance footgun: it always dumps the internal buffer which potentially makes performance even worse than using no buffering at all.
The discussion of this issue in https://github.com/rust-lang/rust/issues/31100 eventually resulted in the addition of BufReader::seek_relative. That method enables seeking on a BufReader without dumping the internal buffer.
Unfortunately, it isn't available to generic code parameterized on R: Read + Seek.
Motivating examples or use cases
A natural approach to writing an image decoding library is to take a R: Read + Seek which enables the user to provide any type implementing Read and Seek.
The decoding process itself will generally involve many small read_exact calls, along with some seeks. Depending on the format, seeks may be to absolute positions in the file, or relative seeks past certain sections (or some combination). Nonetheless, by tracking stream position, the absolute seeks can be done as relative seeks instead.
This presents two performance pitfalls:
If the user passes an un-buffered File object the read_exact calls will be slow.
If they pass a BufReader<File> seeking will constantly be dumping the buffer.
Rust users are generally aware about (1), and library documentation can reinforce the proper solution. However, there isn't a clear way to avoid (2) because BufReader::seek_relative isn't callable given a generic impl Seek.
Solution sketch
Add a seek_relative method to the Seek trait with the same signature as the method on BufReader.
Alternatives
Have libraries always wrap their arguments in BufReader. But that would mean the generic code would sometimes make BufReader<Cursor<&[u8]>> and similar.
Deprecate BufReader in favor of a BufReader2 or similar that had reasonable semantics for seeking.
This issue contains an API change proposal (or ACP) and is part of the libs-api team feature lifecycle. Once this issue is filed, the libs-api team will review open proposals as capability becomes available. Current response times do not have a clear estimate, but may be up to several months.
Possible responses
The libs team may respond in various different ways. First, the team will consider the problem (this doesn't require any concrete solution or alternatives to have been proposed):
We think this problem seems worth solving, and the standard library might be the right place to solve it.
We think that this probably doesn't belong in the standard library.
Second, if there's a concrete solution:
We think this specific solution looks roughly right, approved, you or someone else should implement this. (Further review will still happen on the subsequent implementation PR.)
We're not sure this is the right solution, and the alternatives or other materials don't give us enough information to be sure about that. Here are some questions we have that aren't answered, or rough ideas about alternatives we'd want to see discussed.
Proposal
Problem statement
Rust encourages users to use
BufReader
to accelerate small I/O operations onRead
implementations that have high per-operation overhead.BufReader
'sSeek
implementation however has a performance footgun: it always dumps the internal buffer which potentially makes performance even worse than using no buffering at all.The discussion of this issue in https://github.com/rust-lang/rust/issues/31100 eventually resulted in the addition of
BufReader::seek_relative
. That method enables seeking on aBufReader
without dumping the internal buffer.Unfortunately, it isn't available to generic code parameterized on
R: Read + Seek
.Motivating examples or use cases
A natural approach to writing an image decoding library is to take a
R: Read + Seek
which enables the user to provide any type implementingRead
andSeek
.The decoding process itself will generally involve many small
read_exact
calls, along with someseek
s. Depending on the format, seeks may be to absolute positions in the file, or relative seeks past certain sections (or some combination). Nonetheless, by tracking stream position, the absolute seeks can be done as relative seeks instead.This presents two performance pitfalls:
File
object theread_exact
calls will be slow.BufReader<File>
seeking will constantly be dumping the buffer.Rust users are generally aware about (1), and library documentation can reinforce the proper solution. However, there isn't a clear way to avoid (2) because
BufReader::seek_relative
isn't callable given a genericimpl Seek
.Solution sketch
Add a
seek_relative
method to theSeek
trait with the same signature as the method onBufReader
.Alternatives
BufReader
. But that would mean the generic code would sometimes makeBufReader<Cursor<&[u8]>>
and similar.BufReader
in favor of aBufReader2
or similar that had reasonable semantics for seeking.Links and related work
What happens now?
This issue contains an API change proposal (or ACP) and is part of the libs-api team feature lifecycle. Once this issue is filed, the libs-api team will review open proposals as capability becomes available. Current response times do not have a clear estimate, but may be up to several months.
Possible responses
The libs team may respond in various different ways. First, the team will consider the problem (this doesn't require any concrete solution or alternatives to have been proposed):
Second, if there's a concrete solution: