Functional tests from Swift Package Manager

[uranusjr]

I think I’ve got most things right, but 3/4 of the resolution results are slightly different (each with one discrepancy). Not sure if this is a functional difference between implementations; will need to look into those.

# Kitura
{'https://github.com/IBM-Swift/BlueSocket.git': '0.12.0'} != {'https://github.com/IBM-Swift/BlueSocket.git': '0.12.4'}
{'https://github.com/IBM-Swift/CCurl.git': '0.3.7'} != {'https://github.com/IBM-Swift/CCurl.git': '0.2.3'}

# PerfectHTTPServer
{'https://github.com/PerfectlySoft/PerfectLib.git': '2.0.0'} != {'https://github.com/PerfectlySoft/PerfectLib.git': '2.0.4'}
{'https://github.com/PerfectlySoft/Perfect-Net.git': '2.0.0'} != {'https://github.com/PerfectlySoft/Perfect-Net.git': '2.0.5'}

# SourceKitten
# Passes \o/

# ZewoHTTPServer
{'https://github.com/Zewo/CLibvenice.git': '0.15.0'} != {'https://github.com/Zewo/CLibvenice.git': '0.14.0'}

Assuming both are correct, neither resolver is “superior” IMO since each got newer packages in some cases.

[uranusjr]

For ZewoHTTPServer, all requests to CLibvenice ask for {'0.14.0', '0.15.0'}. CLibvenice has no dependencies. Therefore both 0.14.0 and 0.15.0 are correct, and ResolveLib picks the higher one (as implemented in the provider).

(Note: Four candidates ask for CLibvenice: Venice 0.14.0, File 0.14.0, TCP 0.14.0, and IP 0.14.0.)

The reason Swift Package Manager chooses the lower one may be the Minimal Version Selection principal, which prefers lower versions to higher if both satisfy the constraints.

This should theoretically be implementable by modifying find_matches() to prefer candidates with lower versions (i.e. put them last instead of first). But interestingly this causes the resolver to throw ResolutionImpossible. It is clear we are missing some backtrack possibilities.

[uranusjr]

Merging since I am implementing the tests correctly. The fact they fail is a good thing. The causes should be tracked by subsequent issues.