Thorough testing about zld

[rmaz]

I think passing the bundled unit tests as well as doing a couple of performance comparisons would be a great start.

link the "signal", "storage", and "XUL" files used in the benchmarks, then ensure they're byte-for-byte the same as ones linked from a previous version

Worth comparing to the comparable system linker for byte compatibility? Or would the layout differ somehow?

[michaeleisel]

Yeah, perf comparisons also. The system linker differs due to optimizations around hashing and unstable sorting, but those optimizations can be disabled by using -DREPRO, in which case it should (last I checked) build same as the linker.

[michaeleisel]

I'm noticing non-determinism even when using the same linker. Getting rid of that non-determinism may be a larger project for later, so for now I think we can leave binary reproducibility alone (or at least only test a subset of the binary for equality). As for how to reintegrate the unit tests in a way that's easy for people to use, I think I'll look into putting an XCTest wrapper around all of the tests that actually work, and finding a way to ignore any auto-generated files. Then we can start adding other tests as part of the same XCTest target, e.g. for the 3 linking invocations that I've focused on. The tests will all be put back in and ready by end of next week, no need to worry :) I'm also happy to help if you want to get this done sooner and do the work yourself.

[michaeleisel]

Also, there are a lot of interactions with the filesystem in these tests, hopefully that's not a problem for any CI processes

[michaeleisel]

Closing for now