I'm concerned about the affine-types extension and its impact on ghc and haskell at large

Carter Schonwald carter.schonwald at gmail.com
Sat Feb 8 00:37:04 UTC 2020



As current maintainer of vector, and a member of the CLC, both of which
roles really should be AFFIRMATIONAL stakeholders in this, I only see costs
and concerns.

As someone who's spent the past few years doing a LOT of modelling and
prototyping around marrying linear logic, formal methods, and functional
programming in an applied industrial setting, i should be an Affirmational
stakeholder. yet again I am not.

theres very real costs in the complexity and scope of impact that impact
EVERY single user and stakeholder of ghc and haskell. And I do not see any
concrete population that benefits.

cale even makes a very constructive and articulate point

> I don't know how much my opinion matters at this point, but I'd really
like to see the non-toy real-world use cases of this before I can even
consider thinking that it would be a good idea to merge to the main
compiler. It has such a huge potential impact on so many people in the
Haskell community:
>     * Library maintainers who start getting PRs that make "improvements"
to linearity while making their libraries harder to maintain because their
interface becomes more rigid, and harder to understand because the types
are more complicated.
>     * Beginners, or simply ordinary users of the language who have to pay
the mental overhead of living with the linear types and polymorphism
spreading everywhere as types are adjusted to make terms more usable in
places where one is concerned with linearity.
>     * Commercial users of the language who pay for the additional time
taken by all their employees waiting for the compiler to run, regardless of
whether or not they're using the extension. If Carter's 6-7% slowdown is
real even in cases where one doesn't care about the extension, I can
imagine wanting to make a fork without the extension. The compiler is
already 2 orders of magnitude slower than I'd like. If that weren't the
case, maybe 6-7% wouldn't be a huge deal. While ghci is often helpful at
shortening the feedback loop, it's not always a viable solution.
> But really, I just want to know how anyone would put this to practical
use in a real setting -- it needs to be _really_ compelling to make up for
the social cost. It can't be that hard for Tweag, or someone enthusiastic,
to live on a fork until they have a decent case study to show the world, so
we can say "okay, that's actually really cool, maybe we actually want that
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.haskell.org/pipermail/libraries/attachments/20200207/9a2c8990/attachment.html>

More information about the Libraries mailing list