问题
This is a follow up to a question I asked here... How do I allocate input arrays with f2py?
I've reworked the question in response to comments. The original question was unclear and too lengthy. See comment 1 below.
I am trying to use f2py to wrap various fortran subroutines/functions for reading the data output from a variety of atomic physics codes written in fortran 77. I am using the Anaconda python distribution (python 3.4 64 bit) on Ubuntu 14.04 (64 bit) with the gnu compiler suite (gcc, gfortran, etc.)
I am having issues trying to preallocate the arrays in python. My original question dealt with allocating a character array. Turns out I had the wrong shape. I've fixed that (I think), and am now dealing with a segmentation fault. I've compiled the fortran subroutine with f2py using the --debug-capi flag to enable verbose error reporting.
Here is the relevant portion of the terminal output I get when I run my python script.
debug-capi:Python C/API function adf04_2py.xxdata_04(iunit,titled,iz,iz0,iz1,bwno,npl,bwnoa,lbseta,prtwta,cprta,il,qdorb,lqdorb,qdn,iorb,ia,cstrga,isa,ila,xja,wa,cpla,npla,ipla,zpla,nv,scef,itran,maxlev,tcode,i1a,i2a,aval,scom,beth,iadftyp,lprn,lcpl,lorb,lbeth,letyp,lptyp,lrtyp,lhtyp,lityp,lstyp,lltyp,itieactn,ltied,ndlev=len(ia),ndtrn=shape(tcode,0),ndmet=len(bwnoa),ndqdn=len(qdn),nvmax=len(scef))
debug-capi:int lltyp=:input,required,scalar
debug-capi:lltyp=0
####### omitted ##########
debug-capi:int ndmet=len(bwnoa):input,optional,scalar
debug-capi:ndmet=1
debug-capi:Checking `len(bwnoa)>=ndmet'
debug-capi:int ndqdn=len(qdn):input,optional,scalar
debug-capi:ndqdn=1
debug-capi:Checking `len(qdn)>=ndqdn'
debug-capi:int ndlev=len(ia):input,optional,scalar
debug-capi:ndlev=1
debug-capi:Checking `len(ia)>=ndlev'
debug-capi:int lqdorb=:input,required,array,dims((ndqdn*(ndqdn+1))/2|(ndqdn*(ndqdn+1))/2)
debug-capi:string cprta=:input,required,array,string array,dims(ndmet|ndmet,9|9)
debug-capi:double prtwta=:input,required,array,dims(ndmet|ndmet)
debug-capi:double beth=:input,required,array,dims(ndtrn|ndtrn)
debug-capi:string cstrga=:input,required,array,string array,dims(ndlev|ndlev,(*)|(*))
debug-capi:double zpla=:input,required,array,dims(ndmet|ndmet,ndlev|ndlev)
debug-capi:int ltied=:input,required,array,dims(ndlev|ndlev)
debug-capi:int npla=:input,required,array,dims(ndlev|ndlev)
debug-capi:int i2a=:input,required,array,dims(ndtrn|ndtrn)
debug-capi:int lbseta=:input,required,array,dims(ndmet|ndmet)
debug-capi:int i1a=:input,required,array,dims(ndtrn|ndtrn)
debug-capi:int ila=:input,required,array,dims(ndlev|ndlev)
debug-capi:double scom=:input,required,array,dims(nvmax|nvmax,ndtrn|ndtrn)
debug-capi:double qdorb=:input,required,array,dims((ndqdn*(ndqdn+1))/2|(ndqdn*(ndqdn+1))/2)
debug-capi:string cpla=:input,required,array,string array,dims(ndlev|ndlev,1|1)
debug-capi:int isa=:input,required,array,dims(ndlev|ndlev)
debug-capi:double wa=:input,required,array,dims(ndlev|ndlev)
debug-capi:int ipla=:input,required,array,dims(ndmet|ndmet,ndlev|ndlev)
debug-capi:double xja=:input,required,array,dims(ndlev|ndlev)
debug-capi:Fortran subroutine `xxdata_04(&iunit,&ndlev,&ndtrn,&ndmet,&ndqdn,&nvmax,titled,&iz,&iz0,&iz1,&bwno,&npl,bwnoa,lbseta,prtwta,cprta,&il,qdorb,lqdorb,qdn,&iorb,ia,cstrga,isa,ila,xja,wa,cpla,npla,ipla,zpla,&nv,scef,&itran,&maxlev,tcode,i1a,i2a,aval,scom,beth,&iadftyp,&lprn,&lcpl,&lorb,&lbeth,&letyp,&lptyp,&lrtyp,&lhtyp,&lityp,&lstyp,&lltyp,&itieactn,ltied,slen(titled),flen(cstrga))'
Segmentation fault (core dumped)
My first question is this. Is it possible to tell from the debug output exactly what segfaulted? I don't see how.
As a second question, if you look at the beginning of the debug output, you will see something that looks like a function call, which I believe is python calling the function from the shared object. At the end of the debug output you see something almost identical, but is labelled as a 'fortran sub-routine'. The variables in the fortran subroutine are not in the same order as the python function. The shared object was automatically created by f2py from the original fortran code. Any one know why these variables are in a different order?
Thanks in advance
回答1:
No sure if I'm supposed to be answering my own questions like this. If this is against protocol then I beg forgiveness from the stack exchange powers that be.
All the information IS (so far) contained in the debug output. After allocating the 'scef' variabale as a numpy array of the proper size/shape/type the segfault went away.
I am still getting further errors, but so far going through the terminal output using --debug-capi and carefully resizing anything that is not the correct shape/size/type seems to be fixing them.
If any one else is experiencing these issues, I hope this helps.
来源：https://stackoverflow.com/questions/35322345/segmentation-faults-using-f2py