Detect platform identifier with init script

[Theodus]

Detect platform using a .platform config file created by the init script. This also cleans up much of the code for the show command and adds tests for the platform string parsing.

With this change, supplying --platform=musl on musl-based Linux distros like Alpine should no longer be required as "musl-based" should be detected when ponyup is installed.

If ponyup is upgraded from an earlier version without using the init script, then this change will have no impact and --platform=musl will continue to be needed.

[SeanTAllen]

There are a couple of issues right now:

• Detect isn't working on macOS.
• bootstrap test isn't installing in the usual /root/.local/ on Linux. the install location isn't in the PATH so stable can't be found.

[SeanTAllen]

Rather than true/false for glibc, can this be:

check for glibc: true/false
check for musl: true/false

If neither comes up, we error out?

[SeanTAllen]

@Theodus are you thinking with this that we have 2 ponyup packages for Linux, one for musl, one for gnu and folks download the one that is appropriate for their distro?

[SeanTAllen]

@Theodus I'm having reservations about this.

The problem:

We need to link to libSSL. If we also link to Glibc, then we have to dynamically link the SSL lib. The statically linked binaries are very nice. Having a single ponyup is very nice. I think that linking against Glibc might be worse than having musl users have to give their platform. Rustup defaults to Glibc and requires you to explicitly request musl.

I like the bootstrap testing changes that have been added to this PR. I think we should definitely keep them in some form (they already found a bug).

Ideas:

• Leave as-is with having to supply platform to get musl version of Pony
• On Linux, in the ponyup directory, store some configuration for the platform to use. If config isn't set, then we prompt the user to supply it unless --platform was explicitly provided. If provided, we could save the selected platform and use it as the default in the future. Given that ponyup-init.sh is a shell script, we could try to get cc -dumpmachine info at the time ponyup is installed and write out the configuration if possible then.
• Shell out to cc -dumpmachine until such time as we don't use cc for linking.

[Theodus]

@SeanTAllen I agree with your concerns. I'll remove the libc detection bits from this PR and add go forward with expecting the user to provide the libc in --platform.

[Theodus]

Ok, you do it

[SeanTAllen]

@Theodus ok, the temporary -musl bootstrap test is in, once this is merged, it can be removed via PR after the next nightly to verify everything fully works.

[SeanTAllen]

@Theodus I updated the PR comment with additional info. If there is anything else you think should be added to it, please do and we can use that comment as the commit comment when we squash and merge.

[Theodus]

Looks good to me. I should probably remove the bit about needing --platform on musl from the README

[SeanTAllen]

@Theodus ya and the README update for ponyc, I'll update that as well. And we can merge that after the next ponyup release (which I plan to do before the end of the year).

[SeanTAllen]

@Theodus should the bootstrap test be running make test or just make. I'm not sure what we get out of doing make test. I'm thinking we should only do make.

[Theodus]

Just make seems fine to me

[SeanTAllen]

@Theodus are these failures at this point because not all the tools have macOS releases yet? The error messages are really obscure at this point, we should definitely improve them. It's unclear to me what the failures are.

[Theodus]

They are a result of the packages not being available, yes

[SeanTAllen]

@Theodus what do you think of breaking the tests apart so that release and nightly are different? that would be a bit more clear.

it appears that release tests are hardcoded to specific versions. what is the idea with that? is it recent releases, any releases, all releases? what is your thought for those.

[Theodus]

Only the select test has hard coded versions. The sync tests are a port from the shell script and uses the 2 most recent nightly and release builds.

[SeanTAllen]

What about ifdefing the tests that will fail on macOS for now and create an issue to update them once they will pass by removing the ifdef and updating the versions?

[Theodus]

Good idea, I always forget that pony has ifdefs

[Theodus]

All CI tests are passing now except for the macOS bootstrap (which we expect to fail)

[SeanTAllen]

Merged. After tonight's nightly we should be able to remove the temporary musl script.

@Theodus can you open an issue to remove the ifdefs in the future once they are no longer needed?