I'm seeing dramatically slower runtime performance using byLine and readln when compiling with LDC than with DMD. Has anyone seen this behavior?

Example:

import std.stdio: File, KeepTerminator, writeln;

void main(string[] args) {
    if (args.length != 2) {
        writeln("Provide exactly one filename");
        return;
    }

    auto inputStream = args[1].File();
    size_t lc = 0;
    foreach (line; inputStream.byLine(KeepTerminator.yes)) {
        ++lc;
    }
    writeln(lc);
}

Using a 1.7gb, 37 million line file:
 - ldc2 (-release -O):  37 seconds
 - dmd (-release -O):   4 seconds

Same results with readln rather than byLine.  However, byChunk is fast, even with small chunks like 40 bytes.

This is using:  Mac OS X (10.10.5);  DMD 2.068; LDC 0.16.1

--Jon

LDC 0.16.1 is based on DMD 2.067.1

On Wednesday, 9 December 2015 at 08:43:03 UTC, Kagamin wrote:
> LDC 0.16.1 is based on DMD 2.067.1

Ah, this enhancement:  https://issues.dlang.org/show_bug.cgi?id=11810. Thanks, that's a rather significant optimization.