Fix for issue #156 for gsl_odeiv_system #172
Conversation
|
@hakonhagland wow, this is impressive if it's your first foray into XS/SWIG land! I will see what Travis says about the tests, but I am very interested to merge this |
|
@hakonhagland thanks for fixing 5.8 support! It seems that Travis has lots of PASS except for GSL master branch, which I think could be our test suite potentially being broken on GSL master branch. Not sure if that is your fault, but once we can clear that up, I am happy to merge this awesome PR 👍 |
|
@leto Thanks positive feedback! I am looking at the build failure for the master branch (line #5079) https://travis-ci.org/leto/math--gsl/jobs/597321360#L5079 : it says it seems to me this comes from line #323 in |
|
@hakonhagland you are right! Our code needs to learn about version 2.6, and the associated new functions, so it can do Magic |
|
@hakonhagland would you mind adding a version |
|
@leto Out of curiosity: why do we test against |
|
@leto we test against it so we can detect breakages before a release of GSL comes out. Instead our test suite breaks sooner, giving us time to become compatible for a GSL release. This let's us be prepared, instead of users coming to us and saying "The new GSL release doesn't work with Math::GSL". In theory, So it's not that we "know we don't support it", our goal is to support it and that means test failures exactly when GSL adds/changes stuff, which is more appropriate than much later, from Math::GSL users. We should probably improve our test code to give a more useful error? Any suggestions are welcome. |
|
@hakonhagland the reason this is happening to you is because there has not been a lot of Math::GSL development, so our test suite became a bit old and the new |
|
@hakonhagland I think now maybe this is merge-able or able for you to update for 2.6 and send a new PR |
Implements some typemaps for Math::GSL::ODEIV to fix issue leto#156 for gsl_odeiv_system and gsl_odeiv_evolve_apply().
To compile on perls earlier than 5.18, change the call to av_top_index() with a call to av_len()
mPUSHs was introduced in 5.8.9. Replace with the PUSHs macro to provide backward compatibility.
|
rebased onto current master |
|
Forgot to add |
|
@hakonhagland thank you for going on an Odyssey to fix this!!! It's really exciting stuff. Also, this code really complements my Many numerics people will likely be interested to use your new code and showing the actual performance numbers and providing a "Rosetta stone" to convert Math::GSL programs to Math::GSL::ODEIV programs would be cool. We may be able to do it automatically! The Math::ODE tests are here, they are quite simple: https://github.com/leto/math--ode/tree/master/t Another thing we can do, is compare error estimates across both modules for the same exact equations. You can even vary step size if you want to study that too. I haven't checked but it's likely Math::GSL internally uses RK8 instead of RK4 and so it's should be crazy more accurate and fast. |
|
For what it's worth, I'd like to send my Kudos to both of you for working
thru the perl ODEIV problem. A suggestion would be to include the perl
version of the examples from the GLS lib ODEIV code - I have always learned
a lot from good examples and this is a case where I really need a good
example. Thanks for your efforts.
Tom Morgan
…On Thu, May 14, 2020 at 8:52 PM Duke Leto ***@***.***> wrote:
@hakonhagland <https://github.com/hakonhagland> thank you for going on an
Odyssey to fix this!!! It's really exciting stuff. Also, this code really
complements my Math::ODE module, a pure-Perl module that implements RK4.
I think it would be super interesting to compare what the code looks like
to solve the same equations with both, and also have CPU timings and even
memory use if that is easy.
Many numerics people will likely be interested to use your new code and
showing the actual performance numbers and providing a "Rosetta stone" to
convert Math::GSL programs to Math::GSL::ODEIV programs would be cool. We
may be able to do it automatically!
The Math::ODE tests are here, they are quite simple:
https://github.com/leto/math--ode/tree/master/t
Another thing we can do, is compare error estimates across both modules
for the same exact equations. You can even vary step size if you want to
study that too. I haven't checked but it's likely Math::GSL internally uses
RK8 instead of RK4 and so it's should be crazy more accurate and fast.
—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
<#172 (comment)>, or
unsubscribe
<https://github.com/notifications/unsubscribe-auth/AKSAIXA3IWVAWMZQJLBDVOTRRSG37ANCNFSM4JAHNC7Q>
.
|
|
@morgantomj thanks!! I hope Math::GSL is useful to you and everybody else that uses it. It's definitely the most complex and oldest codebase that I actively maintain! I never knew how long this ride was going to be when I started 😅 |
@leto You are very welcome! I have learnt a lot and it was fun. I think it would be useful to add a wiki page (for developers) explaining the steps needed to upgrade to a new GSL version. It was a trial and error approach for me, and I think the task would have been much easier if there was some kind of a recipe. |
@morgantomj Thank you. |
|
@hakonhagland @morgantomj you both may enjoy this relic of the past I just found: http://leto.net/code/Math-GSL/ And yes, developer docs are absolutely crucial. Math::GSL has one of the most complex build systems of anything that I know of, and definitely one of the most complex build systems in all of CPAN. Fun fact: I started the original Build.PL from https://github.com/rocky/Perl-Device-Cdio but we evolved to be way more complex, because GSL is so complex and ever-changing! |
@leto I sent you an email but seems like I got the wrong email address. The response was "The recipient server did not accept our requests to connect" |
Hello. After a long struggle, I finally managed to implement some typemaps for
Math::GSL::ODEIVto try to fix issue #156 forgsl_odeiv_systemandgsl_odeiv_evolve_apply(). I am still learning XS andswigso please let me know how to improve this :)