Home » forums » LtU Forum

Compile-time constraint solver?

C's type system can be seen as a constraint checker; Haskell's as a constraint solver. If you had a general purpose constraint solver at compile-time what could you use it for other than type checking?

By George Morrison at 2013-06-13 17:31 | LtU Forum | previous forum topic | next forum topic | other blogs | 4700 reads

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.

Linking.

Linking.

By Z-Bo at Thu, 2013-06-13 19:46 | login or register to post comments

But that's link-time!

But that's link-time, not compile-time!

(I kid).

By George Morrison at Sat, 2013-06-15 10:50 | login or register to post comments

See my PPOPP 2013 paper:

See my PPOPP 2013 paper: automatic parallelization of declarative graphics :)

By lmeyerov at Thu, 2013-06-13 20:04 | login or register to post comments

I can't see any reference to this paper online.

I can't see any reference to this paper online. Is there a preprint available anywhere?

By George Morrison at Sat, 2013-06-15 10:55 | login or register to post comments

Here is the ACM DL link:

Here is the ACM DL link: http://dl.acm.org/citation.cfm?id=2442535&CFID=226715040&CFTOKEN=37515098.

Unfortunately, you need an ACM DL membership to download it (or download from a University).

By Sean McDirmid at Sat, 2013-06-15 23:33 | login or register to post comments

Here you go

PDF here. The relevant bit here is that I use Prolog to search for a decomposition of an attribute grammar's static dependency graph into a nice schedule that is a sequence of parallel tree traversal template instantiations.

The axiomatic/constraint formulation is nice in several ways:

As the system designer, it simplifies adding new templates to the system. Furthermore, it enables techniques such as autotuning for free. Looking forward, I think it can enable building smart debugging tools with a robust basis. (Imagine a future where, instead of one-off CHI/UIST papers, ideas can scale to all of CSS.)

As the grammar (visualization) author, I have nice control over the parallelization. I can rely on the fully automatic approach, or for parts I care about, put in manual yet declarative&orthogonal parallelization constraints. This latter 'schedule sketching' feature became a life saver when we were testing parallelization hypotheses and trying to share parallel code with one another. Ultimately, we didn't need the full power of Prolog at compile-time to control the schedule, but it still provides a nice embedding language as Prolog provided a relatively clean syntax, the ability to name subterms, etc.

By lmeyerov at Sun, 2013-06-16 00:35 | login or register to post comments

Interesting

Thanks for that, it reminds me of a thought I had a while back: seeing a web browser as a set of compilers / translators. On the input with HTML, CSS, Javascript etc. and the output translating from an internal representation to a sequence of graphics operations to render the page.

By George Morrison at Sun, 2013-06-16 10:18 | login or register to post comments

Tons of optimizations

Polymorphic specialization analysis.
Alias analysis.
Flow analysis.
Escape analysis.

By Ben L. Titzer at Thu, 2013-06-13 20:27 | login or register to post comments

Instruction ordering

When I came up with this idea, the first thing I thought of was instruction ordering.

By George Morrison at Sat, 2013-06-15 10:52 | login or register to post comments

Indeed: instruction order = parallel schedule

My usage is essentially a fancy version of that -- parallel scheduling.

By lmeyerov at Sun, 2013-06-16 00:42 | login or register to post comments

Version resolution

Selecting which version of a shared dependency to use based on the version constraints placed on it. This is usually implemented by a package manager, but you could push it directly into the language.

By munificent at Thu, 2013-06-13 21:23 | login or register to post comments