Tracking Issue for seekable io::Take

[wangbj]

Feature gate: #![feature(seek_io_take)]

(feature gate open to change)

This is a tracking issue for adding seek instance for io::Take, see issue #37214.

Why

Imagine there's a smaller components within a larger file, the component have predefined ranges, reading beyond the range should be a hard error. But some fields are variable size encoded, so it's hard to know how many bytes to read until we already decoded/read a large part of it. It is super cumbersome to check every read is within the limit (of the component); however, with io::Take, this is much easier: We only need to create a new io::Take by limiting the upper bound to the component's spec. and just let the user to read what ever needed -- any attempt to read beyond the limit would cause a hard io error.

This is all good, but sometimes we need more powerful io rather than read. For instance, we may need peek and restore stream_position() depends on what data was peek-ed, skip a large chunk of unwanted bytes, or maybe we simply wanted to keep track of the stream_position() for better error handling. This is not possible because io::Take is not Seek.

Public API

Added Seek instance for io::Take<T>, so that we can call io::Seek APIs on io::Take<T>.

pub struct Take<T> {
    inner: T,
    limit: u64,
    cursor: u64,          // <---- new member, current position within `Take`.
}

impl<T: Seek> Seek for Take<T> {
    fn seek(&mut self, pos: SeekFrom) -> Result<u64> {
        // snip
    }
}

This is possible by adding a cursor member to io::Take, together with limit, we have:

• cursor + limit === original_limit (when Take is constructed)
• Seek beyond original_limit => stays at original_limit
• Seek beyond 0 (< 0) => ErrorKind::InvalidInput (EINVAL).
• Read beyond original_limit => io error.
• seek without position changes is honored (This is expected, see BufReader doc).

The proposed method has an advantage over approach in #37214, by not using an extra stream_position() when io::Take is constructed.