BBC: Blead breaks DANAJ/Math-Prime-Util-0.73.tar.gz

[hvds]

See http://matrix.cpantesters.org/?dist=Math-Prime-Util+0.73 for examples.

Manual check shows that at the point we merged bignum-0.64 at 9bfb09e the test t/70-rt-bignum.t starts hanging. I'm investigating, and have also reported to danaj/Math-Prime-Util#61 and https://rt.cpan.org/Ticket/Display.html?id=141772

[hvds]

Initial investigation shows that the immediate root cause is this incompatible change in bignum-0.60:

   - The bignum pragma now converts every numeric constant to a Math::BigFloat
     object.

Before v0.60 the docs said:

    All operators (including basic math operations) are overloaded. Integer
    and floating-point constants are created as proper BigInts or BigFloats,
    respectively.

.. and the hanging test in Math::Prime::Util is expecting a Math::BigInt object.

I've asked PJACKLAM to talk to DANAJ to try and resolve it, not sure what else might be affected.

[leonerd]

We stared at this briefly in the PSC call. The breaking unit-test file is:
https://metacpan.org/release/DANAJ/Math-Prime-Util-0.73/source/t/70-rt-bignum.t

It doesn't immediately appear to be "overly-sensitive test is overly-sensitive", so it may be this is real issue here.

[hvds]

@leonerd please see danaj/Math-Prime-Util#61 for the latest state of affairs: there is progress, albeit slowly.

I'm expecting both a change to Math::Prime::Util to fix the test failures with bignum-0.64 and a change to bignum that achieves the original goal of v0.60 in a manner that restores back-compat with pre-0.60 versions (by reversing the changes to bignum.pm and introducing a new bigfloat for the new behaviour). Whether the latter will land in time to be considered for inclusion in perl-5.36 remains to be seen.

If it does not, we could consider backing out the v0.64 upgrade: I expect the bignum family are mainly used for one-liners and throwaway scripts, but there's a smallish risk it is baked into some sysadminy scripts (eg for crypto), where up to now you'd have got away with a use bignum when you only intend to use ints.

The likeliest failure case is probably similar to the MPU test failure, that code like this:

perl -Mbignum -e '$x=1; $x >>= 1, ++$bits while $x'

will loop forever, because $x now gets upgraded to Math::BigFloat instead of Math::BigInt.

[edit: my bignum versions were wrong]

[hvds]

I got the bignum versions wrong in my last comment, I've edited to fix them up.

The incompatible change was introduced in bignum-0.60. We upgraded from 0.59 to 0.63 in 6853e8a (which also updated other modules), and to 0.64 in 9bfb09e.

[hvds]

bignum-0.65 has been released, reversing the incompatibility introduced with v0.60, along with associated updates to Math-BigInt-1.999830 and Math-BigRat-0.2621. I've verified manually that these build and install ok with bleadperl, and with them in place we no longer get a hang in the Math-Prime-Util-0.73 testsuite.

I propose that we should consider bringing those updates into blead to be part of perl-5.36. I'll work on putting together a smoke branch with the relevant updates.

[hvds]

I propose that we should consider bringing those updates into blead to be part of perl-5.36. I'll work on putting together a smoke branch with the relevant updates.

Now created as smoke-me/hv/sync-bignum-20220412.

[hvds]

I propose that we should consider bringing those updates into blead to be part of perl-5.36. I'll work on putting together a smoke branch with the relevant updates.

Now created as smoke-me/hv/sync-bignum-20220412.

Smokes look clean so far : only failures are for Solaris 2.11 which is tripping up in t/porting/utils, I think on utils/zipdetails which is a separate problem.

[demerphq]

On Thu, 14 Apr 2022 at 16:49, Hugo van der Sanden ***@***.***> wrote: I propose that we should consider bringing those updates into blead to be part of perl-5.36. I'll work on putting together a smoke branch with the relevant updates. Now created as smoke-me/hv/sync-bignum-20220412. Smokes look clean so far : only failures are for Solaris 2.11 which is tripping up in t/porting/utils, I think on utils/zipdetails which is a separate problem.

Did you rebase on top of the latest blead? Yves

…

-- perl -Mre=debug -e "/just|another|perl|hacker/"

[hvds]

I propose that we should consider bringing those updates into blead to be part of perl-5.36. I'll work on putting together a smoke branch with the relevant updates.

Now created as smoke-me/hv/sync-bignum-20220412.

Smokes look clean so far : only failures are for Solaris 2.11 which is tripping up in t/porting/utils, I think on utils/zipdetails which is a separate problem.

Did you rebase on top of the latest blead?

No, but I can see on blead smokes that this exact error showed up around the time I made this branch, and has since gone away again. It doesn't seem worth using more smoke capacity just to prove that.