darcs patch: Fix kind inference + test (and 6 more)

Samuel Bronson

4 Apr 2008 4 Apr '08

5:48 a.m.

Mon Mar 31 07:28:01 EDT 2008 Samuel Bronson * Fix kind inference + test Also, comment out the case in kindOf that makes things look fine when they aren't -- unless of course * isn't the right kind Mon Mar 31 11:18:39 EDT 2008 Samuel Bronson * Work around type inference bug relating to main (I didn't do it!) Fri Mar 21 00:29:22 EDT 2008 Samuel Bronson * Split TArrows where appropriate Thu Mar 20 16:42:02 EDT 2008 Samuel Bronson * Actually define (->)... Tue Mar 18 18:43:45 EDT 2008 Samuel Bronson * Add Control.Monad.Instances to base package Tue Mar 18 18:56:40 EDT 2008 Samuel Bronson * Add Control.Category Thu Mar 20 17:33:26 EDT 2008 Samuel Bronson * Make Control.Category actually not import Prelude...

Attachments:

fix-kind-inference-_-test.dpatch (text/x-darcs-patch — 18.8 KB)

Show replies by date

Samuel Bronson

4 Apr 4 Apr

6:17 a.m.

On Thu, Apr 3, 2008 at 7:18 PM, Samuel Bronson wrote:

...

Mon Mar 31 07:28:01 EDT 2008 Samuel Bronson * Fix kind inference + test Also, comment out the case in kindOf that makes things look fine when they aren't -- unless of course * isn't the right kind

This one makes the kind inference actually run when FrontEnd.KindInfer.hsTypeToType, and makes kindOf stop reporting * when the name in question is not in the environment -- signifying that kind inference has definitely not seen it. The former behaviour was useless, serving only to make things appear to work when they in fact did not, merely because * happened to be the correct kind in most cases.

...

Mon Mar 31 11:18:39 EDT 2008 Samuel Bronson * Work around type inference bug relating to main (I didn't do it!)

This one just makes the test case for the previous patch actually run.

...

Fri Mar 21 00:29:22 EDT 2008 Samuel Bronson * Split TArrows where appropriate

This one breaks apart TArrows when they must be matched with other things in type inference.

...

Thu Mar 20 16:42:02 EDT 2008 Samuel Bronson * Actually define (->)...

This establishes the kind of (->) as ?? -> ? -> *, as copied from GHCi output for :kind (->).

...

Tue Mar 18 18:43:45 EDT 2008 Samuel Bronson * Add Control.Monad.Instances to base package

Tue Mar 18 18:56:40 EDT 2008 Samuel Bronson * Add Control.Category

Thu Mar 20 17:33:26 EDT 2008 Samuel Bronson * Make Control.Category actually not import Prelude...

The first two are needed by the still-failing testcase; the third one just makes it build faster by side-stepping Prelude. I seem to have left some patches out this time, though...

Samuel Bronson

6:31 a.m.

...

I seem to have left some patches out this time, though...

Okay, this one has the final patch -- this one mostly replaces hard-coded TAp constructors with tAp invocations -- tAp returns a TArrow instead of a TAp in the appropriate case.

John Meacham

6:36 a.m.

On Thu, Apr 03, 2008 at 07:47:28PM -0500, Samuel Bronson wrote:

...

On Thu, Apr 3, 2008 at 7:18 PM, Samuel Bronson wrote:

...
Mon Mar 31 07:28:01 EDT 2008 Samuel Bronson * Fix kind inference + test Also, comment out the case in kindOf that makes things look fine when they aren't -- unless of course * isn't the right kind

This one makes the kind inference actually run when FrontEnd.KindInfer.hsTypeToType, and makes kindOf stop reporting * when the name in question is not in the environment -- signifying that kind inference has definitely not seen it. The former behaviour was useless, serving only to make things appear to work when they in fact did not, merely because * happened to be the correct kind in most cases.

This was due to the kind inference algorithm specified in the report, any unconstrained kinds are supposed to be defaulted to *. This may not be necessary anymore though as I think I added an explicit defaulting step. In any case, It wasn't put there by accident or a hack even if it is no longer needed. John -- John Meacham - ⑆repetae.net⑆john⑈

Samuel Bronson

7:01 a.m.

On Thu, Apr 3, 2008 at 9:06 PM, John Meacham wrote:

...

...
This one makes the kind inference actually run when FrontEnd.KindInfer.hsTypeToType, and makes kindOf stop reporting * when the name in question is not in the environment -- signifying that kind inference has definitely not seen it. The former behaviour was useless, serving only to make things appear to work when they in fact did not, merely because * happened to be the correct kind in most cases.

This was due to the kind inference algorithm specified in the report, any unconstrained kinds are supposed to be defaulted to *. This may not be necessary anymore though as I think I added an explicit defaulting step. In any case, It wasn't put there by accident or a hack even if it is no longer needed.

Well, that's what I thought it must be an attempt at. But in fact there is other code that does a much better job -- or at least, it does it in a way that provides less false comfort. The other code only defaults kinds for things that have made it into the kind environment. Anyway, the rule of thumb for compilers and RTSs is "crash ASAP", so...

John Meacham

7:20 a.m.

On Thu, Apr 03, 2008 at 09:31:33PM -0400, Samuel Bronson wrote:

...

Well, that's what I thought it must be an attempt at. But in fact there is other code that does a much better job -- or at least, it does it in a way that provides less false comfort. The other code only defaults kinds for things that have made it into the kind environment. Anyway, the rule of thumb for compilers and RTSs is "crash ASAP", so...

Yeah, the original kind checker was out of the THIH in haskell paper, where only simple kinds were allowed so just defaulting to * was a straightforward way to do it. When I added the kind constraint stuff, it became necessary to be a bit smarter in the defaulting stage so that code was no longer needed. You will still find remenants of the old THIH kind checker and type checker around, though I have rewritten both a couple times since then, some code constructs and terminology has a way of sticking around. John -- John Meacham - ⑆repetae.net⑆john⑈

6490

Age (days ago)

6490

Last active (days ago)

List overview

Download

5 comments

2 participants

participants (2)

John Meacham
Samuel Bronson