Make it possible to bisect nqp and MoarVM

[AlexDaniel]

Not sure how this should be implemented exactly, but it would be nice to have.

[AlexDaniel]

MoarVM builds are on their way (due to #119)!

[AlexDaniel]

02:23:07 <MasterDuke> you know what would be nice? nqp bisectable
02:24:14 <MasterDuke> bisect all the things
02:55:40 <AlexDaniel> MasterDuke: I think it can be easily integrated into bisectable
02:56:02 <AlexDaniel> most subs now accept a backend parameter
02:58:01 <AlexDaniel> MasterDuke: for that we'd have to build nqp also
02:58:52 <AlexDaniel> which I guess can be easily done by extending build.p6 a little bit
02:59:05 <AlexDaniel> it already can work with both MoarVM and Rakudo
02:59:10 <AlexDaniel> adding nqp shouldn't be that hard…
02:59:11 <MasterDuke> yeah, shouldn't be too bad
02:59:28 <AlexDaniel> sounds like a LHF actually
03:03:40 <MasterDuke> i would think so
03:05:32 <AlexDaniel> interestingly, nqp also supports --git-reference
03:05:50 <AlexDaniel> I wonder if the same reference can be used by rakudo and nqp at the same time
03:05:53 <AlexDaniel> possibly not
03:06:39 <AlexDaniel> so references should be moved to another directory I guess…
03:06:45 <MasterDuke> isn't the argument to that just a path that has the right named folders in it?
03:06:49 <AlexDaniel> well, repos for references
03:06:53 <AlexDaniel> yes
03:07:19 <AlexDaniel> but I'm not sure what it does
03:07:37 <AlexDaniel> it does pull in these folders, I guess, but anything else?

[AlexDaniel]

After a long discussion (@timotimo++, @MasterDuke17++, @Skarsnik++) here's what I came up with:

1. We can build moar-nqp-rakudo one by one during the bisect process, but it's faster to build everything that may be required before you start bisecting (because we have MoarVM builds already available, and everything else does not seem to become faster if you throw extra cores at it, so parallelize instead).
2. This means that you're guaranteed to wait for the amount of time it takes to build one rakudo, which is a few minutes. If that's the case, why can't we prepare builds for all bump commits in existence? Well, we can.

The only question that's left is what has to be done if we have extra nqp commits between the bumps. Right now I only have this simple case in mind.

[AlexDaniel]

<AlexDaniel> maybe it shouldn't attempt anything smart?
<AlexDaniel> just go through all nqp builds one by one and see what happens
<AlexDaniel> I mean, all nqp commits
<AlexDaniel> so rakudo is on the bump commit, and then we just go through every nqp commit with whatever moar version is declared in nqp
<AlexDaniel> So it is pure nqp bisect first
<AlexDaniel> then it'll get back to the user with results, probably saying that more than one commit is at fault (or that stuff is unbuildable or whatever)
<AlexDaniel> but that's alright

[AlexDaniel]

Eh, not exactly sure why this had LHF tag. It's not too easy.