Git dependencies

[st3veV]

Git dependencies behavior as discussed in #238 :

• it's now possible to call haxelib install haxelib.json and dependencies from the json will be installed
• dependencies can have version in form: git:<url>#commit-ish... if so git dependency will be installed, this is possible in both json and hxml definitions
• if commit-ish is defined and/or different than one already installed from same json/hxml (ie. because of dependency of dependency) user is prompted to confirm overwriting existing git version

[andyli]

Thanks. Would you take a look at the failed test and try to fix it?

Command: haxe [client_tests.hxml]
Class: tests.TestSemVer ...
Class: tests.TestData .E.......
Class: tests.TestRemoveSymlinks .
Class: tests.TestRemoveSymlinksBroken .
Class: tests.TestHg ...........
Class: tests.TestGit ...........
Class: tests.TestVcsNotFound ....
Class: tests.TestInstall ..
* tests.TestData::testReadDataWithoutCheck()
ERR: exception thrown : Invalid field access : length
Called from StringTools::startsWith line 142
Called from haxelib._Data.Dependencies_Impl_::toArray line 83
Called from haxelib.Data::readData line 339
Called from tests.TestData::testReadDataWithoutCheck line 237
Called from a C function
Called from haxe.unit.TestRunner::runCase line 115
FAILED 42 tests, 1 failed, 41 success
Command exited with 1 in 55s: haxe [client_tests.hxml]

[st3veV]

Fixed.

[ibilon]

Hey, thanks for the contribution.

Could you add a check that vcs dependencies aren't allow when submitting to haxelib?

if commit-ish is defined and/or different than one already installed from same json/hxml (ie. because of dependency of dependency) user is prompted to confirm overwriting existing git version

Not sure I fully get what that means, you can't have two branch of the same repository used as deps by different libs/project?

[st3veV]

Yep, the check makes sense. I'll look at it.

Not sure I fully get what that means, you can't have two branch of the same repository used as deps by different libs/project?

Correct. In the discussion for #238 it was decided to not bother with this and support just one because it already brings quite a lot of value (#238 (comment) and #238 (comment)).

[st3veV]

So I did some digging through the code and testing and there already is a "check"... submit fails with ie.: Error: git:https://github.com/fzzr-/hx.signal.git is not a valid version string because the JSON is being validated. In my opinion the error descriptive enough and submit is failing so it behaves as expected but I'm open to different opinions.

[ibilon]

The error message could be improved but it's still descriptive enough.

@nadako I haven't tried it yet but the design and code are good, what do you think?

[nadako]

lgtm, ship it!