Does tyCoVarsOfTypesList guarantee a particular order?

[Simon Peyton Jones]

The left to right ordering is guaranteed by RnTypes.extract<blah> functions.
See Note [Ordering of implicit variables]

I don’t think tyCoVarsOfTypesList makes any guarantees.

But it probably *could*.   If you look at FV.hs, you’ll see that it works right-to-left, putting ‘x’ on the front of the accumulating list the first time it encounters it.

So in tyCoVarsOfTypes [a,b]
we push b onto the list, then a.  Result a:b:[]

In in tyCoVarsOfTypes [a,b,a]
we push a on the list (the rightmost occurrence), then b, then we ignore the next occurrence of a.  Result: b:a:[].

You could add new guarantees I suppose.  But they aren’t there right now.

Simon

From: ghc-devs <ghc-devs-bounces at haskell.org> On Behalf Of Ryan Scott
Sent: 05 September 2018 15:08
To: ghc-devs at haskell.org
Subject: Does tyCoVarsOfTypesList guarantee a particular order?

tyCoVarsOfTypesList guarantees that it returns its answer in a deterministic order. For the longest time, I must have assumed that this order was left to right. However, it appears that my assumption was wrong! This can be demonstrated with this program:
    module Main where

    import Name
    import TyCoRep
    import TysPrim
    import Var

    main :: IO ()
    main = do
      putStrLn "(1)"
      print $ map (getOccString . tyVarName)
            $ tyCoVarsOfTypesList
              [TyVarTy alphaTyVar, TyVarTy betaTyVar]

      putStrLn "(2)"
      print $ map (getOccString . tyVarName)
            $ tyCoVarsOfTypesList
              [TyVarTy alphaTyVar, TyVarTy betaTyVar, TyVarTy alphaTyVar]

This gives the following output:

    (1)
    ["a","b"]
    (2)
    ["b","a"]

The first one makes total sense to me. The second, one however, does not. If the free variables of that list were returned in left-to-right order (or even right-to-left order!), then (2) should give the same answer as (1). Instead, it lists "b" _before_ "a", which I find incredibly baffling.

To explain why I care so much about this, we're currently trying to improve Haddock's logic for choosing when to put explicit `forall`s in front of types [1]. Our litmus test is this: if the order in which a user wrote the `forall`d variables differs from the order in which the free variables of the body would normally occur, then have Haddock display an explicit forall. I would have assumed that tyCoVarsOfTypesList [2] would be enough to determine the "normal" order of the free variables, but as the example above proves, this sometimes gives unexpected orderings when there are multiple occurrences of the same variable.

We are currently having to work around this issue [1] by implementing our own custom versions of tyCoFVsOfType and friends that accumulate variables in reverse order (and then reversing the list at the end!) to get the order we expect. This feels incredibly wasteful to me, so I'd like to know if there's a better way. In particular:

1. Is this behavior of tyCoVarsOfTypesList expected?
2. If not, should we change it?

Ryan S.
-----
[1] See https://github.com/haskell/haddock/pull/931<https://na01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fhaskell%2Fhaddock%2Fpull%2F931&data=02%7C01%7Csimonpj%40microsoft.com%7C79995d7103ee4a30066308d6133901b7%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C636717532892126908&sdata=Q%2F2vdira5ae1Y%2FEgXAADmN7D1HZYo1w1d00wlH%2FdDPg%3D&reserved=0>
[2] Actually, I would use tyCoVarsOfTypesWellScoped, but dependency order doesn't come into play in the example I gave above.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.haskell.org/pipermail/ghc-devs/attachments/20180905/93330825/attachment.html>