Ricardo Abreu

I don't know, I fear that total link time may grow too much when changing something that affects many tests. I believe linking is where most time is spent when...

> Agreed. Having separate executables will incur extra link time. However, if we introduce an intermediate dynamic library that bundles all the static ones, with the use of [--whole-archive](https://linux.die.net/man/1/ld) (or...

FWIW, 5 years ago I tried to [randomize tests](https://github.com/canonical/multipass/pull/1399) and found that they couldn't handle it. Unfortunately it didn't cross the priority threshold and 3 years later it remained open,...

Hey @levkropp, while this is hot in your terminal, what happens if you touch `utils.cpp` rather than `test_utils.cpp`? Just object libraries vs current main with mold would be enough.

OK, I think this has become moot really, at least as far as the original goal of speeding up iteration time. Let's discuss how to take it from here in...

Not a review, but I am on board with the idea. It makes it more consistent and predictable. :+1:

Hey @xmkg, > We rely on the main/primary/default (or whatever you would call it) interface for Multipass's basic operation We won't for long though, right? With the vsock/agent we could...

All that said, I don't feel super strongly about this. Perhaps there are other aspects that I could be missing?

Thanks for the PR @artemvlas! We're a little tied up ATM, but we'll review when we get a chance.

Hi @artemvlas, I think that'a about the best we can do without changing the return type.