Windows: Win32 >= 2.12.0.0 (for GHC > 9.0.1) breaks foundation

[Release-Candidate]

A version of Win32 > 2.10.0.0 is needed to support GHC 9.0.2 (2.10.0.0 works with GHC 9.0.1), see: https://github.com/commercialhaskell/stackage/issues/6400

Win32 changed to explicit exports with version 2.12.0.0 https://github.com/haskell/win32/releases/tag/v2.12.0.0 https://github.com/haskell/win32/commit/fbc49f8bb694a276da49a70166bed1e6a85b9cf4#diff-3c294c5dd8b170586cad01c72d6ffd35225608811de64b538677cfde2c0b7d64 which breaks the usage of the C bindings.

See https://github.com/haskell-foundation/foundation/issues/557 from the 14th of January (2022)

Solution

• Use an upper bound of Win32 < 2.12 for foundation >= 0.0.10 https://github.com/haskell-foundation/foundation/issues/557#issuecomment-1013088994

• use the exported functions without c_ Pull Request: https://github.com/haskell-foundation/foundation/pull/558

Error message

> Foundation\Time\StopWatch.hs:73:9: error:
>     * Variable not in scope:
>         c_QueryPerformanceCounter :: Ptr a0 -> IO a1
>     * Perhaps you meant `queryPerformanceCounter' (imported from System.Win32.Time)
>    |
> 73 |         c_QueryPerformanceCounter (castPtr p `ptrPlus` 8)
>    |         ^^^^^^^^^^^^^^^^^^^^^^^^^
>

[andreasabel]

* Use an upper bound of `Win32 < 2.12` for foundation >= 0.0.10

We can easily help with making the necessary revisions on hackage.

• use the exported functions without c_

This requires modification of the code, thus either a NMU (non-maintainer upload) or a package takeover. (Please google the relevant procedures.)

[sjakobi]

It would be nice if a Windows user could check which foundation versions need to be revised. So far the assumption that v0.0.10 and up require revisions is based on looking at the code – it would be good to have more certainty.

[phadej]

This requires modification of the code, thus either a NMU (non-maintainer upload) or a package takeover. (Please google the relevant procedures.)

Given that there is takeover of cryptonite and memory in flight, I think foundation and basement will either dropped from the dependency tree or taken over as well. (I assume that's the motivation to make foundation compile/work).

EDIT: a link https://mail.haskell.org/pipermail/libraries/2022-January/031582.html

So I'd suggest that we, Hackage Trustees, stay away from engaging into NMU. (Also that change doesn't sound like very simple one).

[sjakobi]

I have performed the revisions for foundation-0.0.10 and up, see e.g. https://hackage.haskell.org/package/foundation-0.0.10/revisions/.

The build failure in v0.0.26.1 and the usage of the same c_QueryPerformanceCounter function in v0.0.10 and up seemed like sufficient evidence that these revisions are necessary.

I did not check whether foundation < 0.0.10 is compatible with Win32 >= 2.12.

[juhp]

I am not Windows user myself but feel for them... in the short term it would be great if https://github.com/haskell-foundation/foundation/pull/558 could be merge and released IMO.

[gbaz]

Vincent has informed the HF board that he's still around, does not authorize a package takeover, or any trustee actions on his packages, and will make releases when he wants to. Trustees are therefore unable to act in this situation. The job of trustees is to help maintainers and at times to supplement when maintainers fall away or are unresponsive. When there is an author and maintainer who says they are active, we are not in a position to judge if they are genuinely active or not, nor is it in the remit of trustees to act against maintainers wishes in terms of NMUs, etc.

In such a situation, the ecosystem will have to find another mechanism to adapt.

[phadej]

or any trustee actions on his packages

This is very sad, as it includes revisions as well. I don't see any way forward other then migrating to something else™.

[andreasabel]

Vincent's message is here: https://groups.google.com/a/haskell.foundation/g/board/c/68K8jlYChFQ/m/sWj_s96HBQAJ?pli=1