Fix and re-enable MSI build

[Boddlnagg]

This supersedes #661, though it does not enable deployment yet.

[Boddlnagg]

I want to see if this works on AppVeyor, but there seems to be a problem with AppVeyor right now ...

[Boddlnagg]

Looks like this works. I will remove the last two commits as soon as this is reviewed.

@brson @Diggsey I have some more changes for the MSI installer ready, but some questions will need to be discussed in the process as to the general direction of the MSI installer.

[bors]

☔ The latest upstream changes (presumably #1230) made this pull request unmergeable. Please resolve the merge conflicts.

[alexcrichton]

Thanks for this @Boddlnagg and sorry for the delay! It looks like this is just a few fixes here and there, so if you'd like to rebase and remove the temporary commits, I can r+ to land this

[Boddlnagg]

@alexcrichton Thanks! I did the rebase.

[alexcrichton]

Should this be removed now that BUILD_MSI is enabled?

[Boddlnagg]

The MSI version is nowhere near finished. I want to enable it on CI just so it won't break. Maybe it would make sense to also set ALLOW_PR: 1 for it.

[alexcrichton]

Ah ok, that makes sense and no worries! My only worry is that this'll conflict with the other i686 builder for the dist artifacts being produced. Enabling the MSI installer just builds more artifacts, right? It shouldn't change the existing binaries?

(if so we can probably just gate on the msi build)

[Boddlnagg]

No artifacts are being produced for MSI currently, see https://github.com/rust-lang-nursery/rustup.rs/blob/ac5bffad7ba126b2313692c3dfce83ec91cb7fea/ci/prepare-deploy-appveyor.ps1#L9-L12

Changing that situation was part of #661, but I didn't include it here. Will do in another PR.

[alexcrichton]

Thanks! Looks like CI is failing though?

[mattico]

Could you use find_vs_version instead? https://docs.rs/gcc/0.3.51/gcc/windows_registry/fn.find_vs_version.html

[Boddlnagg]

Thanks! I hadn't seen that function.

[Boddlnagg]

The CI failure is weird: error: failed to parse lock file at: C:\projects\rustup-rs\Cargo.lock. Running with --verbose gives

Caused by:
  package `gcc 0.3.50 (registry+https://github.com/rust-lang/crates.io-index)` is specified as a dependency, but is missing from the package list

Other crates in the same workspace use gcc 0.3.51 and manually changing the version in the lockfile makes the error disappear (I'm adding that to the PR). But I don't really understand what's going on. Is this expected behavior for cargo?

Edit: Ah, I see, it's missing a checksum line for gcc 0.3.50, I guess that's why it didn't work?

[alexcrichton]

Oh nice! In that case...

@bors: r+

[bors]

📌 Commit eaed698 has been approved by alexcrichton

[bors]

⌛ Testing commit eaed698 with merge c4e7616...

[bors]

☀️ Test successful - status-appveyor, status-travis
Approved by: alexcrichton
Pushing c4e7616 to master...