Skip to content

HTTPS clone URL

Subversion checkout URL

You can clone with
or
.
Download ZIP

Loading…

deferred callback execution #6

Closed
tadam opened this Issue · 5 comments

2 participants

@tadam

I tried to use Math::GSL::Min and here is a problem with gsl_min_fminimizer_iterate() mentioned in KNOWN_BUGS.

This problem arises from typemap for gsl_function. It assumes that callback executes when we pass it to some gsl_* function.
But in Math::GSL::Min (and at least in Math::GSL::Root) we have a deferred callback execution: some function like gsl_min_fminimizer_set_with_values()takes callback as a param and after this some function like gsl_min_fminimizer_iterate() uses this callback.

I made a primitive patch for this typemap ( https://gist.github.com/1089862 ). It makes 2 things:
1) we dynamically create a stuct gsl_function_perl * pointer
2) we remove typemap(freearg) gsl_function * because our pointers must live after.
Commented tests in t/Min.t and t/Roots.t works after this patch (and others too).

But after this we have 2 problems with memory:
1) we should check that such function as gsl_min_fminimizer_free() or undef'ing minimizer really frees memory. If it doesn't we should fix it somehow.
2) if we use some gsl_* functions with immediate callback execution (for example gsl_integration_qagiu()) we should call some *_free() function anyway.

I will explore this problem deeply (my knowledge of XS isn't good). But comment please if you have any ideas.

@leto
Owner

If you turn that gist into a branch, I will test it out on my machine.

In general, if your patch makes more tests pass, then I am all for it.

To help with memory leaks, we should probably add an author test that runs our test suite under Valgrind. I think there are some CPAN modules that help with that.

@tadam

#28 is a pull request that solves this problem for a case when we have gsl_function as input parameter and make a deferred callback execution (exactly two cases: Math::GSL::Roots and Math::GSL::Min).

Sorry for additional issue ( #28 ) created together with pull request.

For another cases (such as a gsl_function_fdf) we can make separate isssues and resolve these issues relying on this issue.

@tadam

I have added comments for the pull request.

@tadam

Sorry, actually I placed comments in my commit, not in the pull request:
tadam@2254916#comments

@leto
Owner

Fixed in #28.

@leto leto closed this
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Something went wrong with that request. Please try again.