Inaccurate version bound

[hvr]

As can be seen at https://matrix.hackage.haskell.org/package/transformers-lift@1549836919 transformers-lift is not compatible with transformers >= 0.5.6 due to module name clashes.

I've already revised

• https://hackage.haskell.org/package/transformers-lift-0.2.0.0/revisions/
• https://hackage.haskell.org/package/transformers-lift-0.2.0.1/revisions/

accordingly.

The resulting build reports can be inspected at

https://matrix.hackage.haskell.org/package/transformers-lift@1549837448

[int-index]

I checked and I can't find any module name that would clash. What is the exact build failure?

Could it be caused by the writer-cps-transformers dependency? The Control.Monad.Trans.Writer.CPS module was copied from there to the transformers package in 0.5.6.

If that's the case, I think writer-cps-transformers must be revised, not transformers-lift.

[int-index]

@hvr I figured out how to read the logs, and indeed, the issue is due to writer-cps-transformers:

src/Control/Monad/Trans/Lift/Local.hs:30:18:
    Ambiguous module name `Control.Monad.Trans.Writer.CPS':
      it was found in multiple packages:
      transformers-0.5.6.2 writer-cps-transformers-0.1.1.4

Please undo your changes to my correct version bounds.

[int-index]

Okay, come to think of it, if two packages expose equally named modules, maybe it is downstream's issue to deal with. I tried to adapt my code to transformers >= 0.5.6 but it does not expose .Internal modules for Control.Monad.Trans.Writer.CPS and Control.Monad.Trans.RWS.CPS, so that turned out impossible.

[hvr]

@int-index hrm... yeah, this one can in theory have multiple ways to be resolved.

But it's not writer-cps-transformers's "fault" that another package suddenly decided to name-clash its modules (i.e. there's no requirement whatsoever that packages have unique module names over Hackage or even just among packages that may be used together by a consumer). In this particular case I believe the best course of action is to use {-# LANGUAGE PackageImports #-} and disambiguate that you indeed want the CPS modules to be taken from writer-cps-transformers

---

Otoh, I just realised due to minad/writer-cps-transformers#7 that @minad seems to intend to actually turn writer-cps-transformers into a compat package (which is also a bit tricky as it needs to ensure that the API version of writer-cps-transformers accurately reflects the API exposed for transformers older and newer than 0.5.6.0 (iow, depending on writer-cps-transformers == x.y.z doesn't result in some conditional API)... or things will become extremely messy to fixup for Hackage Trustees :-) )

[minad]

@hvr indeed. The writer-cps-* packages will turn into compatibility packages soon.

[hvr]

@minad fyi, recently with the network/network-bsd split, we had a related situation which took some time to figure out and it also shows the trickiness involved (hinted at in the package description): http://hackage.haskell.org/package/network-bsd

...applied to the writer-cps-transformers case it'd require to carefully track the future transformers releases to make sure that everytime the API reexported through writer-cps-transformers might change, a new release of writer-cps-transformers with the appropriate version bump is performed.

[int-index]

The thing I can't figure out is how to deal with .Internal modules. I need them for some functionality, but transformers doesn't seem to expose an unsafe interface to the CPS modules.

[hvr]

@int-index if transformers doesn't expose things you need, I'm afraid you're out of luck -- unless some dirty unsafeCoerce games are played :-)

[minad]

Transformers should expose those modules. Can you open an issue please?

[int-index]

@minad Alright, I opened an issue there https://hub.darcs.net/ross/transformers/issue/67