Upgrading Stack to Cabal 2.2

Tue Feb 20 17:45:31 UTC 2018

I'm sorry, I'm not quite sure I understand your recommendation. Are you
saying that I should ideally replace all usages of `License` in the Stack
codebase with `Either SPDX.License License`? That _should_ be possible, the
only questions I'd have are:

1. We additionally grab license info from the GHC package dump. Would there
be any problem parsing that license field into the old License data type
and storing as Right?
2. If we're going to have to treat this as arbitrary text anyway, is there
any reason not to represent it as `newtype TextualLicense = TextualLicense
Text` or similar, and convert immediately with `display`?

On Tue, Feb 20, 2018 at 6:12 PM, Oleg Grenrus <oleg.grenrus at iki.fi> wrote:

> Hi Michael,
>
> thanks for your comments!
>
> - The allBuildInfo change is
> https://github.com/haskell/cabal/commit/8fc10320a5dc4898927c84ad6a2dce
> 7965ef30db,
> I agree with Herbert on this. New `allBuildInfo` implementation is
> correct given the name. There was even a TODO to make that change.
> `reallyAllBuildInfo` would been silly. I also didn't felt ok to change
> the type to `Traversal` (we have lenses, please try out them too and
> tell if something is missing!). We'll highlight the change in the
> changelog, as it's easy to miss.
>
> - The lack of SPDX changelog entry is my bad. It was a series of
> patches, and the changelog + manual update is still not done. I try to
> summarise:
>     - `cabal-version: 2.2` files use SPDX expression syntax for
> `license: ` field. PackageDescription's licenseRaw will be Left expr,
> expr :: Distribution.SPDX.License.License
>     - `cabal-version: 2.0` files still use old `License` syntax and
> type, licenseRaw is `Right l`, l :: Distribution.License.License
>     - I recommend treating the field as opaque. You can `either display
> display` to show it
>        - Next best choice is to use `SPDX.License` as in that direction
> conversion is lossless (licenseToSPDX), but have to workaround lack of
> e.g. `OtherLicense` as a concept (IIRC it's converted to
> LicenseRef:OtherLicense which is valid SPDX-License-Id). This might be
> necessary if you plan to read (human readable) text back.
>
> Oleg
>
> On 20.02.2018 15:47, Michael Snoyman wrote:
> > Hi all,
> >
> > Oleg mentioned to me on Reddit that a 2.2 branch has been cut for
> > Cabal, and recommended we try to upgrade Stack to it. I'm sharing
> > information here on that process in case it's useful to others, either
> > as feedback on the API changes, or help for others going through
> > similar upgrades. If anyone would rather that I do or do not post such
> > reports in the future, let me know.
> >
> > I've opened a PR for this change[1], which currently is a single
> > commit[2]. As you can see from the change to stack.yaml[3], I needed
> > to refer to the commit of Cabal in question, and turn on `allow-newer`
> > for some packages (cabal-doctest and http-api-data) with restrictive
> > upper bounds on Cabal. Both of these packages compiled without issue.
> >
> > No change was necessary to Stack's Setup.hs file.
> >
> > Most of the changes so far were compile-time errors, which is a Good
> > Thing. Changes I had to make:
> >
> > * The build type is no longer a Maybe field. (As a user: this is a
> > nice change, no longer a need to guess about what a Nothing value means.)
> > * There are now two License types, the original one and the SPDX
> > variant. I don't understand what this change is about, some further
> > explanation in the docs or the ChangeLog would definitely be
> > appreciated. But hacking my way through the types seems to have worked.
> > * Parsing a package description now works on a ByteString instead of a
> > String. This allowed me to remove some UTF8 conversion and
> > BOM-stripping code, which is very nice.
> > * The new parsing API exposes the cabal file version when that
> > prevented the parse (at least, that's my understanding). The changes
> > to accommodate the new API were relatively trivial, so another +1 here.
> > * Also: getting file position information for warnings and errors is
> > _very_ nice. +1
> >
> > I haven't done extensive testing yet, but I've so far only found one
> > bug due to runtime changes: the behavior of allBuildInfo has changed
> > to now include non-buildable components. This resulted in some
> > confusion until I reviewed the changelog more closely. If I could make
> > a request, it would be that, in the future, new runtime behavior come
> > with a new function name, to convert runtime errors into compile time
> > errors. This was _not_ a particularly difficult issue to work around
> > though, in large part thanks to the changelog.
> >
> > I'll continue testing this branch of Stack. Our plans are to merge
> > this to master, and release Stack 1.7.0 close to the Cabal 2.2 release
> > date.
> >
> > Michael
> >
> > [1] https://github.com/commercialhaskell/stack/pull/3878/files
> > [2]
> > https://github.com/commercialhaskell/stack/pull/3878/commits/
> a101341d04213d6dd8e0cf16d2f2fef8e7ed5cd5
> > [3]
> > https://github.com/commercialhaskell/stack/pull/3878/commits/
> a101341d04213d6dd8e0cf16d2f2fef8e7ed5cd5#diff-
> fafd0cdcd559a7b124cc61c29413fb54
> >
> >
> > _______________________________________________
> > cabal-devel mailing list
> > cabal-devel at haskell.org
> > http://mail.haskell.org/cgi-bin/mailman/listinfo/cabal-devel
>
>
>
> _______________________________________________
> cabal-devel mailing list
> cabal-devel at haskell.org
> http://mail.haskell.org/cgi-bin/mailman/listinfo/cabal-devel
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.haskell.org/pipermail/cabal-devel/attachments/20180220/48b089b5/attachment-0001.html>