Should we aim for a new release in late January?

Johan Tibell johan.tibell at gmail.com
Thu Nov 28 12:32:24 UTC 2013


Hi,

I think it's non trivial due to us not having a nailed down design in
several areas (i.e. it's not just coding work needed). If you look at the
ToDo items at:

   https://ghc.haskell.org/trac/ghc/wiki/Commentary/Packages/MultiInstances

several of the ToDos don't have clear designs yet. For example:

   - instances of packages must install in a different location
      - install directory includes hash?
      - SDM: not done yet. One problem is that we don't know the hash until
      the package is built, but we need to know the install locations earlier
      because we bake them into Paths_foo.hs.
      - Simon and Andres discussed that one option is to let Cabal compute
      its own hash. However, then we'd have two hashes to deal with. Only using
      the Cabal-computed hash isn't an option either according to
Simon, because
      apparently GHC's ABI hash computation is non-deterministic, so
we might end
      up with situations where Cabal's hash is stable, but GHC computes an
      ABI-incompatible version. This is somewhat worrying ...
      - Duncan thinks that we should store both a package identity and a
      package ABI hash. Currently we form the package id from the name, version
      and ABI hash. We should store the ABI hash separately anyway because
      eventually we will want to know it, to know which packages are ABI
      compatible. So Cabal can compute a package Id in advance, however is
      sensible, and the ABI hash is calculated as now, after the build. The
      installation directory follows the package Id.

and

   - GHC: discard conflicting instances during its shadowing phase
      - SDM: GHC will currently do *something* here, but it might end up
      with a result that the user didn't want/expect. One way to improve things
      is to prioritise packages that were installed more recently.
      - Andres suggests that GHC should be much cleverer, and look at the
      actual dependencies of the modules being compiled before deciding which
      packages to enable. This would almost certainly result in more things
      working and possibly less surprising behaviour sometimes, but
Simon thinks
      that (a) it is too hard, (b) if users need this, they should use
Cabal and
      its dependency resolver, which will do a good job, (c) you can often
      resolve problems by adding -package X, and (d) eventually we will
      want a system where users manage separate sessions, so they can set up an
      environment in which the packages they want are available. This has a lot
      in common with cabal-dev and sandboxes, so the mechanisms (and
      concepts) should be shared. (kosmikus: perhaps an alternative is to force
      the user to make an active decision in case of conflicts, i.e.,
to create a
      sandbox that exposes a consistent package set).

-- Johan


On Thu, Nov 28, 2013 at 1:12 PM, Simon Peyton-Jones
<simonpj at microsoft.com>wrote:

>  Is it really that many cycles?  (Compared to all the other great stuff
> the Cabal team are doing.)
>
>
>
> Simon
>
>
>
> *From:* Johan Tibell [mailto:johan.tibell at gmail.com]
> *Sent:* 28 November 2013 12:07
> *To:* Simon Peyton-Jones
> *Cc:* cabal-devel at haskell.org
> *Subject:* Re: Should we aim for a new release in late January?
>
>
>
> Hi Simon,
>
>
>
> I don't think it's being actively worked on. Duncan and I chat about it
> once in a while when we meet in person, but we don't have the cycles to do
> it.
>
>
>
> On Thu, Nov 28, 2013 at 12:41 PM, Simon Peyton-Jones <
> simonpj at microsoft.com> wrote:
>
>  I’m thrilled that there so much Cabal activity!
>
>
>
> Is anyone working on, or interested in, the issue of compiling and
> installing the same package against different dependencies?  (There was a
> Google SoC project about this.)  Cabal sandboxes address the same issue,
> but at some user cost.  It should Just Work.
>
>
>
> Simon
>
>
>
> *From:* cabal-devel [mailto:cabal-devel-bounces at haskell.org] *On Behalf
> Of *Johan Tibell
> *Sent:* 27 November 2013 21:01
> *To:* cabal-devel at haskell.org
> *Subject:* Should we aim for a new release in late January?
>
>
>
> Hi,
>
>
>
> Cabal development is continuing at a furious pace. There are lots of great
> things in master that I'd like to get out to users, such as
>
>
>
>  * relinking avoidance
>
>  * build -j
>
>  * ability to specify exact deps on the command line
>
>  * haskell-suite compiler support
>
>  * bug fixes
>
>
>
> Here's an approximate list of commits (both Cabal and cabal-install) since
> the last release:
>
>
>
>     https://github.com/haskell/cabal/compare/Cabal-v1.18.1.2...master
>
>
>
> It's not entirely accurate as some patches were cherry-picked onto the
> 1.18 branch and thus have different commit IDs.
>
>
>
> -- Johan
>
>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.haskell.org/pipermail/cabal-devel/attachments/20131128/eb1a5340/attachment-0001.html>


More information about the cabal-devel mailing list