PackageResolver Refactor

[cloudRoutine]

Started working on this in an attempt to address #2030

I wasn't able to get the step function to the point where its tail recursive, but its much closer than it was before and it no longer relies on mutation across iterations.

I think it should be possible by adding a third case to

type private Stage =
    | Outer of conflictState : ConflictState
    | Inner of conflictState : ConflictState

and using it as an argument for step to direct it down one of three pathways for the start of a step, an outer loop, or an inner loop, although when I tried coding that up I couldn't get it to work.

I think this is the block of code that was causing the issues when I took that approach

                    match conflictState.Status with
                    | Resolution.Conflict(_,conflicts,_,_)
                        when
                            (Set.isEmpty conflicts |> not)
                            && nextStep.CurrentResolution.Count > 1
                            && not (conflicts |> Set.exists (fun r ->
                                r = currentRequirement
                                || r.Graph |> List.contains currentRequirement)) ->
                        let flags = StepFlags(flags.Ready,flags.UseUnlisted,flags.HasUnlisted,true,flags.FirstTrial)
                        stepLoop flags (Inner (conflictState))
                    | _ ->
                        stepLoop flags (Inner (conflictState))

but I'm not totally sure.

[forki]

please don't

[forki]

this is really really hard to review- can you please split it in two PRs. one with the resolver change and one with all the cosmetical changes? thanks

[cloudRoutine]

@forki this fine?

[forki]

it seems the new resolver is not resolving correctly tests like ##1254 are red. can you confirm?

[cloudRoutine]

a bug got reintroduced when I redid the commits

[forki]

#1883 install FSharp.Core from Chessie fails!?

[cloudRoutine]

I've been having issues running the integration tests on my machine

I know I can move the dir, but maybe the integration tests don't need to pull in so many netcore packages?

[forki]

yeah these new packages are fucking up everything. MS is clearly out of control here.

[cloudRoutine]

even when I try to run the integration tests on the core3 branch they don't complete

[forki]

do you have VS open at that time? or something else that keeps an handle on the nupkg?

[cloudRoutine]

I had all of that stuff closed at the time. The first time I got those errors I thought that might be the source but when I tried again with everything off it still happened.

I get these errors when I run the tests just in VS too

[forki]

And from fake build?

Am 23.11.2016 16:50 schrieb "Jared Hester" notifications@github.com:

I had all of that stuff closed at the time. The first time I got those
errors I thought that might be the source but when I tried again with
everything off it still happened.

I get these errors when I run the tests just in VS too

https://camo.githubusercontent.com/941f9899d827833319c461cb4c1171afb00f673f/687474703a2f2f692e696d6775722e636f6d2f50717949665a442e706e67

—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
#2039 (comment),
or mute the thread
https://github.com/notifications/unsubscribe-auth/AADgNDMhgrvmLy6PeGchRuyJk5e8gBuDks5rBGDQgaJpZM4K6Nx6
.

[cloudRoutine]

the log in the gist is from the FAKE build

[forki]

Ok can you please try to find out which tool is locking it? Sorry for all the work, but this whole .net standard conversion is a real mess.

[forki]

#1977 seems to work on appveyor. (travis is expected to fail)

[cloudRoutine]

The readme might need a note to install F# 3.0, without it #144 will fail with

But even after I installed F# 3.0

[cloudRoutine]

ok it seems like the error is coming from netstandard getting installed with chessie
when I run the contents of test #1883 in FSI

[cloudRoutine]

A more detailed output

[cloudRoutine]

now i'm pretty sure the bug is here, where I last made a change 🤦
which I'd had as

                    if exploredPackage.Unlisted && not flags.UseUnlisted then
                        tracefn "     unlisted"
                        stepLoop flags (Inner(conflictState))
                    else

before, but that caused an infinite loop printing unlisted

@forki it'd be really useful for whoever has to work on this next, if you or someone else with a deeper understanding of what's going on in this module could add some comments explaining the processes. I'd do it myself, but I don't think I have a firm enough grasp and I wouldn't want to put down something that's wrong

[cloudRoutine]

@forki finally passed on appveyor 🎉

[cloudRoutine]

anything else I need to do for this?

[forki]

thanks so much. let's what the real world thinks about it ;-)