Some fixes for v1.0

[simonschoelly]

See #65
I fixed some things to make this work on Julia 1.0 . Unfortunately Compose.jl is not yet ready and produces a stack overflow when trying to print a Compose.Context to the console and there also seem to be some other problems when running the tests for GraphPlot.jl.

What doesn't work:

using LightGraphs, GraphPlot
g = CompleteGraph(3)
gplot(g)

what works:

using LightGraphs, GraphPlot, Cairo
g = CompleteGraph(3)
draw(PDF("test.pdf", 16cm,16cm), gplot(g))

and

using LightGraphs, GraphPlot
g = CompleteGraph(3)
gplothtml(g)

So this also seems to fix #54

Unfortunately, I also had to add Arpack.jl to the requirements. And I took the liberty of bumping the version requirement to 0.7.

[simonschoelly]

Strangely, Github somehow shows a commit for a different pr (from a different branch) in the commit tab for this pr.

[simonschoelly]

The problems with Compose are now fixed and we just have to wait for them to tag a new version. In the meantime, adding Compose#master should make this pr already work, for those who need it.

There was a strange problem on my computer. Apparently it is not enough to specify VisualRegressionTests 0.2.2 in tests/REQUIRE but it also has to be specified in REQUIRE, even though it is only used for testing. We will have to check if this is also the case on travis or if this is just a problem with my computer.

[jpfairbanks]

do we need this? Can we use the [email protected] LightGraphs.LinAlg.eigs?

[simonschoelly]

I don't think we need that. I therefore changes the code so that it uses LightGraphs.LinAlg.eigs.

[jpfairbanks]

you will need to specify the which=LR() for this eigs call.
and using ArnoldiMethods in order for this to work

[simonschoelly]

Ah I see, I though I thought LightGraphs.LinAlg.eigs was just an identical replacement for the old eigs functions. Any reason why I should use LightGraphs.LinAlg.eigs instead of partialschur directly?

[simonschoelly]

Apparently not all lines of the code are covered by tests and therefore there were a lot of errors that slipped by me. The situation should be better now, but we need more tests.
See #67

[simonschoelly]

Compose has tagged a new version. There where some strange errors, that prevented the tests from passing on Travis, but I found some solutions for them and made some notes in the comments.