Use context instead of nested its to fix breaking changes in Crystal .27.1

[Jens0512]

As of Crystal .27.1 nested it's are no longer allowed (crystal-lang/crystal#7297), and it raises at runtime. The breaking change affects 6 of our it's, hence the specs are broken.

This PR simply changes the parent it's to describe's which fixes the problem.

[Jens0512]

The CI is broken: No command 'crystal' found. 👀

Seems we're not the only ones with failing crystal on travis: https://travis-ci.community/t/apt-get-install-crystal-0-27-1-failed/2093

[faultyserver]

Seems good to me. Will merge once they get the CI image updated. I suppose also updating shard.yml and .tool-versions to 0.27.1 would be good to do here as well.

That said I still don't understand why they've been so adamant to not allow users to specify the version of Crystal they're using. There's only latest and nightly, and even nightly has almost never worked.

[Jens0512]

Fixed, according to bcardiff of the Crystal core team.

[faultyserver]

Looks like it's still failing because of the changes to integer overflow handling.

[Jens0512]

That's weird, the overflow behavior should be disabled unless -D preview_overflow is passed to the compiler.

From the PR merging the overflows:

In order to avoid breaking changes and to allow a preview of this feature, the overflow behavior is only enabled if compiled with -D preview_overflow.

[faultyserver]

Very. I haven't pulled and run it myself to see what exactly is causing it, it might just be a stack recursion overflow or something else.

In any case, I would imagine it's reproducible.

[Jens0512]

It works perfectly well for me: asciineema recording (Finally a reason to try asciinema!).

The OverflowError on the CI is raised during the compilation of spec/all_spec.cr, which means the error is with the compiler. But I fail to make sense of how it fails on the CI, but not on my machine. I've got a slightly different version: Crystal 0.27.1 (2019-02-02), while the CI says it has Crystal 0.27.1 [64137d045] (2019-01-30), but I dunno if that has much to say...

[Jens0512]

The CI 64137d045 crystal version ref is exactly the 0.27.1 release. The only thing that I can imagine makes sense is that the CI has some kind of limitation on its resources, which makes the build fail. However I fail to see how it would raise an OverflowError, I mean, it's an integer overflow, right? The int max and min values are the same on the CI and my computer are the same, as both are x64 systems, AFAIK crystal doesn't even support 32-bit architechtures. Alternatively, might this be a bug in the compiler?

[Jens0512]

Yep, it's the compiler.

[Jens0512]

There, now we just wait for .27.2. Shouldn't be long.