Description
Zig Version
0.11.0-dev.2160+49d37e2d1
Steps to Reproduce and Observed Behavior
When attempting to make a GET request to any web page using TLS, the request seems to take 5x longer to read, this can be observed by running zig run -lc example.zig, where example.zig contains the minimal reproduction code below:
const std = @import("std");
pub fn main() !void {
const allocator = std.heap.c_allocator;
var client = std.http.Client{ .allocator = allocator };
var headers = std.http.Client.Request.Headers{};
const url = "https://github.com/MasterQ32/parser-toolkit/archive/master.tar.gz";
var req = try client.request(try std.Uri.parse(url), headers, .{});
const buf = try req.reader().readAllAlloc(allocator, std.math.maxInt(usize));
allocator.free(buf);
}However, it is noticed that when doing the same with a site not using TLS, it does not appear to happen. This also appears to affect the package manager that was implemented in #14265. This can once again be reproduced by using the following minimal reproduction:
.{
.name = "mocha",
.version = "0.1.0",
.dependencies = .{
.ptk = .{
.url = "https://github.com/MasterQ32/parser-toolkit/archive/master.tar.gz",
.hash = "122035a324a6a46bc367ecd79dfb0bf666bda97cb64ab4f5d20e6d7e88972f608d84",
},
},
}
I have eliminated any chances of this being an antivirus or firewall issue, as my system uses Windows' built-in stuff, and I have made no changes to any settings. Around a week ago, however, I noticed this slowdown suddenly start happening, not only on my own devices, but on CI's from multiple websites such as sr.ht, and codeberg.
Expected Behavior
According to hyperfine, downloading the tarball in the reproduction code above takes normally ~300-500ms
/tmp ❯ hyperfine "curl -sL https://github.com/MasterQ32/parser-toolkit/archive/master.tar.gz"
Benchmark 1: curl -sL https://github.com/MasterQ32/parser-toolkit/archive/master.tar.gz
Time (mean ± σ): 362.2 ms ± 118.7 ms [User: 20.2 ms, System: 2.7 ms]
Range (min … max): 277.7 ms … 680.3 ms 10 runs
After compiling the reproduction code to eliminate the chances of it being build-time related, we can see it being nearly 350x slower.
/tmp ❯ hyperfine -r 2 "curl -sL https://github.com/MasterQ32/parser-toolkit/archive/master.tar.gz" "./hello"
Benchmark 1: curl -sL https://github.com/MasterQ32/parser-toolkit/archive/master.tar.gz
Time (mean ± σ): 352.8 ms ± 99.1 ms [User: 21.5 ms, System: 0.0 ms]
Range (min … max): 282.7 ms … 422.9 ms 2 runs
Benchmark 2: ./hello
Time (mean ± σ): 120.566 s ± 0.311 s [User: 0.081 s, System: 0.007 s]
Range (min … max): 120.346 s … 120.787 s 2 runs
Summary
'curl -sL https://github.com/MasterQ32/parser-toolkit/archive/master.tar.gz' ran
341.77 ± 96.02 times faster than './hello'