Linux testing on CI

[janderit]

Note: This PR originally fixed some bugs with NSString on Linux. That has been fixed in #94. Now this PR is for testing SPM support on Linux.

StencilSwiftKit was not usable building on Linux due to two string to NSString coercions. The Linux swift compiler does not recognize the coercion.

Replacing them with NSString(string: ...) allows building projects using StencilSwiftKit on Linux.

[AliSoftware]

Thanks a lot for this PR, this is awesome to have StencilSwiftKit made compatible with Linux 👍

I don't have a Linux machine to test it and our CI isn't configured to test on Linux yet so I'll trust you tested it locally 😉

Can you just please add an entry in the CHANGELOG to mention the change (eg "added Linux compatibility" or something) and credit yourself? Thanks!

As a bonus it would be even better if you could also edit the circleci.yml to add a Linux container to our CI config so that we ensure that we won't introduce a regression on Linux in future versions of the lib 🙂

[djbe]

Totally agree with @AliSoftware, if we add Linux support we should test it on CI.

[djbe]

Yeah so, ruby is missing from that docker image (for linux), so no gem or bundler commands... Anyone have a docker image handy that fits our requirements, or wants to create one?

[djbe]

Right, so I've finally got linux tests working on CI. We should update the job names GitHub expects after the merge.

Note that this introduces LinuxMain.swift and XCTestManifests.swift, both files are autogenerated by swift test --generate-linuxmain. We can keep using this (and should add a Rake task for it), or use Sourcery to generate the needed code.

[djbe]

Oh hey, forgot about this, we merged #94 that fixes a similar issue 😅. I'll rebase this PR to just add the linux support part.

[djbe]

@RahulKatariya With what Swift version did you run the tests on Linux?

There's a bug (swiftlang/swift-corelibs-foundation#1536) that causes our Linux tests to crash when using NSString.standardizingPath (used by Path.normalize()). It's been fixed, but no idea in which Swift version (we're using 4.1.x).

[djbe]

Checked and the fix is neither in the swift 4.1 or 4.2 branches, only in master. @AliSoftware do you think they'd be open to cherry-picking that change into those branches, and would we go about it? Comment on that PR? Open an issue?

[AliSoftware]

For reference, we've pinged some Swift core team members, and @millenomi told us they'll look into merging swiftlang/swift-corelibs-foundation#1536 soon, but that it will not be included in the upcoming release of Swift anyway (too late already). So we'll have to rely on a workaround (at least in the test suite, if a local fix is possible without requiring a dedicated version of PathKit just for that workaround, preferably)

[rahul0x24]

@djbe The path methods were causing the compilation error on Linux, I didn't really use them.

[djbe]

@AliSoftware I'm merging this now as it's "fixed", we can revisit this later if we want, but at least we won't block the PRs in SwiftGen anymore.

[AliSoftware]

I'm ok with merging to avoid waiting.

A comment in the code referencing
swiftlang/swift-corelibs-foundation#1536 would have been nice though, to understand why we do that #if os(Linux) workaround and to be able to quickly check when it's possible to remove the workaround in the future.