std::make_pair

Cython does not include definitions for std::make_pair

The blog post https://molpopgen.github.io/fwdpy/docs/examples/cython/WritingExtensions includes the following example

%%cython --cplus --compile-args=-std=c++11 -I $fwdpy_includes -I $fwdpp_includes -l sequence -l gsl -l gslcblas

from fwdpy.fwdpy cimport *
from libcpp.utility cimport pair
import numpy as np

ctypedef vector[double] vd
ctypedef pair[vd,vd] pvdvd

#Annoyingly, Cython currently does not 
#expose std::make_pair, so we will 
#do it here ourselves!
#If we don't have std::make_pair,
#we end up making extra temporary copies
#of our return values in memory.  That's 
#not cool, as we're doing this because
#we care about efficiency!
cdef extern from "<utility>" namespace "std" nogil:
    pair[T,U] make_pair[T,U](T&,U&)

cdef pvdvd freq_esize_cpp(const singlepop_t * pop):
    cdef vd freq,esize
    cdef double twoN = 2.0*float(pop.N)
    cdef size_t i = 0
    cdef size_t nm=0
    for i in range(pop.mcounts.size()):
        if pop.mcounts[i]>0:
            if pop.mutations[i].neutral is False:
                freq.push_back(float(pop.mcounts[i])/twoN)
                #s records the effect size/selection coefficient
                esize.push_back(pop.mutations[i].s)
    #For some reason, we need to provide casts
    #so that Cython can get the types right for the call to
    #make_pair:
    return make_pair(<vd>freq,<vd>esize)

def freq_esize(Spop pop):
    return freq_esize_cpp(pop.pop.get())

Is this the “best practice” way? If so, would it make sense to include the make_pair definition in Includes/libcpp/?

[robertwb]

Yes, for sure. PRs welcome.

On Fri, Nov 9, 2018, 5:17 AM Leonard Lausen <[e-mail] wrote:

Cython does not include definitions for std::make_pair

The blog post https://molpopgen.github.io/fwdpy/docs/examples/cython/WritingExtensions includes the following example

%%cython --cplus --compile-args=-std=c++11 -I $fwdpy_includes -I $fwdpp_includes -l sequence -l gsl -l gslcblas

from fwdpy.fwdpy cimport * from libcpp.utility cimport pair import numpy as np

ctypedef vector[double] vd ctypedef pair[vd,vd] pvdvd

#Annoyingly, Cython currently does not #expose std::make_pair, so we will #do it here ourselves! #If we don’t have std::make_pair, #we end up making extra temporary copies #of our return values in memory. That’s #not cool, as we’re doing this because #we care about efficiency! cdef extern from “<utility>” namespace “std” nogil: pair[T,U] make_pairT,U

cdef pvdvd freq_esize_cpp(const singlepop_t * pop): cdef vd freq,esize cdef double twoN = 2.0*float(pop.N) cdef size_t i = 0 cdef size_t nm=0 for i in range(pop.mcounts.size()): if pop.mcounts[i]>0: if pop.mutations[i].neutral is False: freq.push_back(float(pop.mcounts[i])/twoN) #s records the effect size/selection coefficient esize.push_back(pop.mutations[i].s) #For some reason, we need to provide casts #so that Cython can get the types right for the call to #make_pair: return make_pair(<vd>freq,<vd>esize)

def freq_esize(Spop pop): return freq_esize_cpp(pop.pop.get())

Is this the “best practice” way? If so, would it make sense to include the make_pair definition in Includes/libcpp/?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/cython/cython/issues/2706, or mute the thread https://github.com/notifications/unsubscribe-auth/AAdqgdEq4NjnD9Gt0IV6ctDBGN0Z3vb4ks5utQHAgaJpZM4YV702 .

[CrimsonVex]

Fixed by changing make_pair( 1, 2 ) to pair[ int, int ]( 1, 2 )