Looking at #1119 I noticed that tests use TCP with seemingly the only intention being testing whether the code can handle broken-up packets. This can be achieved without TCP and without artificial delay of 1s considerably slowing down the tests. I've seen an entire library to simulate breaks at random points, I just can't remember how it was called. But it can be done manually too:
// Breaks up reading to force multiple `read` calls when it reaches the end of each part
struct BreakingReader {
// Changing to iterators left as an exercise for the reader
data: Vec<Vec<u8>>,
idx: usize,
pos: usize,
}
impl BreakingReader {
fn new(data: Vec<Vec<u8>>) -> Self {
BreakingReader {
data,
idx: 0,
pos: 0,
}
}
}
impl io::Read for BreakingReader {
fn read(&mut self, buf: &mut [u8]) -> io::Result<usize> {
if self.idx >= self.data.len() {
return Ok(0)
}
// We never let pos to go out of bounds
let part = &self.data[self.idx][self.pos..];
let to_copy = part.len().min(buf.len());
buf[..to_copy].copy_from_slice(&part[..to_copy]);
if to_copy == part.len() {
self.pos = 0;
self.idx += 1;
} else {
self.pos += to_copy;
}
Ok(to_copy)
}
}Then just replace serve_tcp() with BreakingReader::new() and remove the handle/join.
Looking at #1119 I noticed that tests use TCP with seemingly the only intention being testing whether the code can handle broken-up packets. This can be achieved without TCP and without artificial delay of 1s considerably slowing down the tests. I've seen an entire library to simulate breaks at random points, I just can't remember how it was called. But it can be done manually too:
Then just replace
serve_tcp()withBreakingReader::new()and remove thehandle/join.