[opentheory-users] Fwd: Re: cannot build opentheory

[Joe Hurd]

Hi Rob,

Thanks for digging into the problem - I'm happy there is a short-term
workaround and a long-term plan to fix the underlying problem.

Cheers,

Joe

On Fri, Aug 17, 2012 at 8:15 AM, Rob Arthan <rda at lemma-one.com> wrote:
> Here is an update with a work-around from David Matthews:
>
> I've taken a quick look at this but I'm going to leave it for the moment.
> It definitely looks like a bug in the inling of recursive functions and I'm
> aware of another bug in that area that caused a crash.  I need to spend some
> time working through the code and probably rewriting a chunk of it but I
> don't want to start on this just now.  I'd like to release the current
> version with all the storage management changes in it fairly soon.
>
> As a work-around, setting PolyML.Compiler.maxInlineSize to something smaller
> seems to avoid the problem.
>
>
> Regards,
>
> Rob.
>
> On 17 Aug 2012, at 10:32, Ramana Kumar wrote:
>
> (forwarding to the list)
>
> ---------- Forwarded message ----------
> From: "Rob Arthan" <rda at lemma-one.com>
> Date: Aug 17, 2012 8:25 AM
> Subject: Re: cannot build opentheory
> To: "Ramana Kumar" <ramana.kumar at gmail.com>
>
> Ramana,
>
> Thanks. I think this is the same problem but I think it might not have been
> explained properly to David Matthews. The problem is clearly an unintended
> loop at compile time and if you make minor adjustments to the source it
> compiles. I have pinpointed the exact function binding that causes the
> problem and have sent a slightly cut-down version of the source to David.
>
> Regards,
>
> Rob.
>
> On 16 Aug 2012, at 15:20, Ramana Kumar wrote:
>
>> Rob, the forwarded conversations below from last year may be of
>> interest for your Poly/ML problem. I thought they were in the list
>> archive, but apparently not (until now).
>>
>> ---------- Forwarded message ----------
>> From: Ramana Kumar <ramana.kumar at gmail.com>
>> Date: Mon, Sep 19, 2011 at 11:47 AM
>> Subject: Re: cannot build opentheory
>> To: David Matthews <David.Matthews at prolingua.co.uk>
>> Cc: Joe Hurd <joe at gilith.com>
>>
>>
>> On Mon, Sep 19, 2011 at 4:34 PM, David Matthews
>> <David.Matthews at prolingua.co.uk> wrote:
>>> An infinite loop could be a problem with the compiler but I have seen
>>> this
>>> with storage allocator.  Is this on 32-bit or 64-bit?  You could try a
>>> larger value of -H.  This sets the initial heap size and also controls
>>> the
>>> step-size when it needs to increase the heap.  The debug settings are
>>> quite
>>> limited in 5.4.1.  I've extended them quite a bit in SVN HEAD.
>>
>> I am using SVN r1324 at the moment. I am on 64-bit. I'll happily try
>> any debug or other options you think would be helpful. In fact I'll
>> try a larger -H1000 now... it still appears to be stopping at the same
>> point. (I'll leave it running for another 10 minutes...)
>>
>>
>>>
>>> David
>>>
>>> On 19/09/2011 16:24, Ramana Kumar wrote:
>>>>
>>>> On Mon, Sep 19, 2011 at 3:26 PM, David Matthews
>>>> <David.Matthews at prolingua.co.uk>  wrote:
>>>>>
>>>>> It may be a problem with the garbage-collector especially if you're
>>>>> using
>>>>> the SVN version.  Try giving an explicit heap size (e.g. -H500) on the
>>>>> poly
>>>>> line and see if that makes a difference.
>>>>
>>>> Adding -H500 appears not to make any difference. Is that supposed to
>>>> limit the heap size, or just set its initial size? Are the debug
>>>> options listed anywhere?
>>>>
>>>>>
>>>>> David
>>>>>
>>>>> On 16/09/2011 12:39, Ramana Kumar wrote:
>>>>>>
>>>>>> After seeing your timings I wanted to try polyml again, and
>>>>>> unfortunately it still appears to go into an infinite loop while
>>>>>> building opentheory.
>>>>>> This could be a bug in polyml. In fact, I you say it definitely is,
>>>>>> since you can get it to build with a different version?
>>>>>> I am using version 5.4.1 (svn r1324).
>>>>>> It goes into the loop when trying to interpret the code starting at
>>>>>> line 56548 in bin/polyml/selftest.sml (the definition of ppTerm in the
>>>>>> Haskell structure).
>>>>>> Do you have any ideas on how to isolate the problem? Maybe I should
>>>>>> also leave it running for longer to see if it's really a loop (I've
>>>>>> only left it for about 15 minutes - how long does it take to build
>>>>>> with your working version of poly?)
>>>>>> I am CC'ing Dave Matthews in case he has any questions or ideas..
>>>>>>
>>>>>> On Thu, Sep 15, 2011 at 11:09 PM, Ramana Kumar<ramana.kumar at gmail.com>
>>>>>> wrote:
>>>>>>>
>>>>>>> On Thu, Sep 15, 2011 at 11:01 PM, Joe Hurd<joe at gilith.com>    wrote:
>>>>>>>>
>>>>>>>> Hi Ramana,
>>>>>>>>
>>>>>>>>>> With polyml, gilith/basic appears to fail one of it selftests and
>>>>>>>>>> building opentheory appears to go into an infinite loop.
>>>>>>>>>
>>>>>>>>> Poly/ML seems to be misbehaving in ways I don't understand right
>>>>>>>>> now.
>>>>>>>>> I'm investigating this, and will let you know what I find out.
>>>>>>>>
>>>>>>>> I installed the Hombrew
>>>>>>>>
>>>>>>>> http://mxcl.github.com/homebrew/
>>>>>>>>
>>>>>>>> version of Poly/ML for my Mac, and basic and opentheory seem to work
>>>>>>>> fine now. I was seeing the same behaviour as you before I did that,
>>>>>>>> so
>>>>>>>> perhaps there's a dodgy version of Poly/ML in circulation?
>>>>>>>
>>>>>>> Maybe. I usually use a pretty bleeding edge poly too (i.e. svn rather
>>>>>>> than a release, although I don't svn up that often).
>>>>>>>
>>>>>>>>
>>>>>>>> Is this problem blocking you?
>>>>>>>
>>>>>>> I have both mosml and mlton too so it's not a major problem.
>>>>>>>
>>>>>>>>
>>>>>>>> Cheers,
>>>>>>>>
>>>>>>>> Joe
>>>>>>>>
>>>>>>>
>>>>>
>>>
>
> _______________________________________________
> opentheory-users mailing list
> opentheory-users at gilith.com
> http://www.gilith.com/opentheory/mailing-list
>
>
>
> _______________________________________________
> opentheory-users mailing list
> opentheory-users at gilith.com
> http://www.gilith.com/opentheory/mailing-list
>