vapor-community / bits Goto Github PK

In ByteSequence+Conversions.swift.

This behavior seems likely to cause silent errors, as opposed to throwing or returning an optional string.

As an example, I was trying to get the string representation of a BCrypt salt back out of the digest. My first attempt (totally wrong, I should have used Serializer) was this:

try BCrypt.Parser(hashBytes).parseSalt().bytes.makeString()

It returned "" rather than indicating failure.

While trying to debug some performance issues I was having today, I realised the root of my problems was an insanely slow parse in Multipart. Not that the work I was doing was too slow - the parser itself was slow.

I opened up the project in Xcode and added a simple test:

measure {
    let parser = try! Parser(boundary: "MimeBoundary_8923gr836bidfb3i")
    
    var parts: [Part] = []
    parser.onPart = { part in
        parts.append(part)
    }

    parts = []
    try! parser.parse(message)
    XCTAssertEqual(parts.count, 2)
}

For a 332kB message, with two parts (some text and a JPEG image), the implementation using Bytes (i.e. [UInt8]) took 0.422s. Replacing Bytes with Data resulted in the same test executing in 0.0475s. That's nearly a 10x speedup.

Now admittedly Multipart is a small library, and it wasn't zero work to change it over to use Data. Vapor as a whole is enormous, and such a changeover would require care and attention and a lot of work. But I think Bits really can't continue defining its own type for data when that type is so much slower than the native type. IMO, Bits should be a library for operating on Data objects, and Collections/Sequences of UInt8, and should ditch the Bytes type entirely.