Add solar.py back to TESTS

[bqpd]

Reverts #33

@mjburton11 It only fails on the Npods=5 solve on those machines, so in hindsight I could have commented that out instead of removing the whole file from TESTS.

[bqpd]

retest this please

[mjburton11]

That's interesting that some machines are not robust enough to handle it

[bqpd]

@galbramc, analysis of this issue has moved to this thread

[galbramc]

What has moved here?

[bqpd]

Oh, discussion from convexengineering/gpkit#1306

Is it macys_VM, reynolds, and windows7x64 that have MOSEK 7? The new QPROP model in solar.py doesn't solve on those three machines. :-/

[galbramc]

Yeah those are the machines with mosek 7. Can you download 7 and give it a try on your machine? Do they still support 7?

[bqpd]

it's still available for download but does not seem to be actively supported

[galbramc]

Your call then if you support 7...

[galbramc]

Don't forget that some people deal with locked down systems. Working with government agencies in particular have lots or restrictions on what they can put on their computer, and upgrading for 7 to 8 may be non-trivial.

[mjburton11]

test this please

[bqpd]

@galbramc after thinking about it, and encountering another MOSEK 7/8 difference on #1315, I think we want to have MOSEK 8 on all machines.

When that's done solar.py should be able to pass tests on those three machines again and we should be able to merge this.

[bqpd]

test this please

[bqpd]

@mjburton11 there appears to be a bug on this branch? it's not passing on any of the machines.

[galbramc]

Can you do anything in your build script to check that people are only using mosek 8 or newer in that case? I don't know when 9 will come out, but I'm sure it will some day.

[galbramc]

I'm going to wait until the bug on this branch is fixed to double confirm that you want the switch to mosek 8 on all machines.