chore: Replace ctor with std::sync::Once

[Marwes]

A std::sync::Once should effectively do the same thing as ctor, only difference is an extra check before failures whether the initialization has already been done.

Haven't tested this (on windows) but it ought to be fine. If nothing else maybe someone on windows can use this PR to test or just have faith that it is ok :)

[tommilligan]

This looks amazing, thank you! I agree that the Once pattern should do the same job and is far less magic than ctor, given that everyone knows about once_cell and I had to look up what ctor does.

As this is a Windows specific feature, I agree there needs to be some Windows QA before this is changed. Ideally I'd like to setup some representative CI that actually verifies this feature, but I'd settle for this being tried out on a Windows machine.

I'd actually be interested to see if this has any compilation speed impact on non-windows platforms, not that I think it should.

[sjackman]

@tommilligan Do you have a moment to review this PR?

[tommilligan]

@sjackman the code looks good to me - from my view this is blocked on someone testing on Windows system and checking it still works.

If anyone can verify this branch and post a screenshot of the example running in powershell or something similar, that's good enough for me.

[dbanty]

Anyone have suggestion on how to test this? I thought, from the description of output_vt100, that I would not see any colors in PowerShell or Command Prompt if init was not called. However, I ran cargo run --example pretty_assertion in both PowerShell and Command Prompt and I get pretty colors even with output_vt100 completely removed 😅.

This is on Windows 10, is this only required for older versions of Windows?

[tommilligan]

@dbanty Thank you for testing this.

After a quick dive into the git history, it appears this dependency was added by the dependency's author (Phundrak) in a commit back in 2019. The dependency itself is very low star count, and the download count is almost all due to it's inclusion in pretty_assertions.

Therefore, I'm going to conclude it's probably no longer necessary/not doing anything useful, and remove it in the next minor version. If anyone screams, we can reinstate it after getting details of the failure environment for testing.

Thanks all for your contributions to investigating this!

[tommilligan]

Closing in favour of #118

[tommilligan]

For reference, the original PR/discussion of why this was initially added: #28

The original states that ANSI codes where not supported by default in powershell, which appears to no longer be the case.

[tommilligan]

ctor has been removed in v1.4.0