Bug #52082 for Math-GMPf: Failing test - Rmpf_eq (with gmp 4.3.2 dev release)

Thu Nov 26 04:22:55 2009 kmx [...] cpan.org - Ticket created

Subject:

Failing test - Rmpf_eq (with gmp 4.3.2 dev release)

Hi, I have just tested Math::GMPf 0.29 with dev release of GMP (4.3.2 candidate) and found a problem with failing test: - t\test1.t - test 18 = Rmpf_eq($q, $r, 96) The interesting is that this test pass with last parameter values <=94 and fail for value 95 and higher. There seem to be some changes in GMP's implementation of function mpf_eq() since GMP 4.3.1 and it may cause this problem. Could you please give me a short feedback if you see it as a GMPlib problem or as a Math::GMPf problem. Thanks. -- kmx

Thu Nov 26 05:50:58 2009 sisyphus [...] cpan.org - Correspondence added

RT-Send-CC:

kmx [...] cpan.org

On Thu Nov 26 04:22:55 2009, KMX wrote: Show quoted text

> Hi, > > I have just tested Math::GMPf 0.29 with dev release of GMP (4.3.2 > candidate) and found a problem with failing test: > - t\test1.t > - test 18 = Rmpf_eq($q, $r, 96) > > The interesting is that this test pass with last parameter values <=94 > and fail for value 95 and higher. > > There seem to be some changes in GMP's implementation of function > mpf_eq() since GMP 4.3.1 and it may cause this problem. > > Could you please give me a short feedback if you see it as a GMPlib > problem or as a Math::GMPf problem. >

Math::GMPf should faithfully copy the behaviour of the gmp library against which it has been built. That is, perl's "Rmpf_eq($q, $r, $bits)" should return the same as C's "mpf_eq(q, r, bits)". If there has been a change to mpf_eq(), then it may be that test 18 needs to be rewritten to accommodate that change - but I probably ought to take a look at the 4.3.2 dev distro you've been using. Where do I get it ? Cheers, Rob

Thu Nov 26 05:50:58 2009 The RT System itself - Status changed from 'new' to 'open'

Thu Nov 26 06:28:10 2009 kmx [...] volny.cz - Correspondence added

Subject:	Re: [rt.cpan.org #52082] Failing test - Rmpf_eq (with gmp 4.3.2 dev release)
Date:	Thu, 26 Nov 2009 12:27:42 +0100
To:	bug-Math-GMPf [...] rt.cpan.org
From:	kmx <kmx [...] volny.cz>

Hi Rob, I am using GMP binaries that are part of mingw-w64 gcc-toolchain binaries provided by sezero (you might remember those :). See enclosed patch that was applied to gmplib. -- kmx Show quoted text

> Math::GMPf should faithfully copy the behaviour of the gmp library > against which it has been built. That is, perl's "Rmpf_eq($q, $r, > $bits)" should return the same as C's "mpf_eq(q, r, bits)". > If there has been a change to mpf_eq(), then it may be that test 18 > needs to be rewritten to accommodate that change - but I probably ought > to take a look at the 4.3.2 dev distro you've been using. Where do I > get it ? > > Cheers, > Rob

Download gmp-4.3.2-dev.diff.zip
application/octet-stream 5.5k

Message body not shown because it is not plain text.

Fri Nov 27 04:50:36 2009 sisyphus [...] cpan.org - Correspondence added

RT-Send-CC:

kmx [...] volny.cz

On Thu Nov 26 06:28:10 2009, kmx@volny.cz wrote: Show quoted text

> Hi Rob, > > I am using GMP binaries that are part of mingw-w64 gcc-toolchain > binaries provided by sezero (you might remember those :).

Indeed I do :-) In test 18, the precision of both $q and $r is 64 - and I'm thinking it's a bit dubious to be testing whether the first 96 bits of two 64- bit numbers are identical or not. Testing that "Rmpf_eq($q, $r, 64)" returns true would seem to be a more sensible test to conduct, and would solve the problem, I believe. Does that sound reasonable ? Cheers, Rob

Fri Nov 27 05:32:00 2009 kmx [...] volny.cz - Correspondence added

Subject:	[SPAM] Re: [rt.cpan.org #52082] Failing test - Rmpf_eq (with gmp 4.3.2 dev release)
Date:	Fri, 27 Nov 2009 11:31:40 +0100
To:	bug-Math-GMPf [...] rt.cpan.org
From:	kmx <kmx [...] volny.cz>

Show quoted text

> In test 18, the precision of both $q and $r is 64 - and I'm thinking > it's a bit dubious to be testing whether the first 96 bits of two 64- > bit numbers are identical or not. Testing that "Rmpf_eq($q, $r, 64)" > returns true would seem to be a more sensible test to conduct, and > would solve the problem, I believe. > > Does that sound reasonable ? >

Yes - "Rmpf_eq($q, $r, 64)" - works. Thanks. -- kmx

Sat Nov 28 03:52:30 2009 sisyphus [...] cpan.org - Correspondence added

RT-Send-CC:

kmx [...] volny.cz

On Fri Nov 27 05:32:00 2009, kmx@volny.cz wrote: Show quoted text

> Yes - "Rmpf_eq($q, $r, 64)" - works.

I'll change it to that in the next release of Math::GMPf (and close this bug report when that "next release" is released). Thanks for raising this. Cheers, Rob

Mon Apr 05 05:29:35 2010 sisyphus [...] cpan.org - Correspondence added

This should be fixed in 0.30. (Meant to close this ticket when 0.30 was released ... but forgot, until now.) Cheers, Rob

Mon Apr 05 05:29:36 2010 sisyphus [...] cpan.org - Status changed from 'open' to 'resolved'