Alternative to using array constructor for writing array to JSON? about json-fortran

[jacobwilliams]

Yes, there is a way to do it. Stand by for example...

[kmanalo]

@jacobwilliams thanks, looking forward to it!

[jacobwilliams]

OK, this might be what you need. The idea is you have to create an empty array object, and then add the strings one at a time. Then you add this array to your main json structure. There is no need to put the allocatable string into a temp string. Code snippet below.

type(json_value),pointer :: array  !temp for creating the array

call json_create_array(array, 'char_array')  !create an empty array

!add each element (note: name is not used)
do i=1,size(derived_type_array)
  call json_add(array, '', derived_type_array(i)%str)  ![this is json_value_add_string]
end do

call json_add(outp, array)  !add the array to the output [this is json_value_add_member]
nullify(array) !don't need this anymore

Note: I haven't tried to compile a working example, so if you can't get it to work, let me know. I think this should work. Also: be aware that Gfortran still has some bugs related to the allocatable character strings... I've got several workarounds in json-fortran to account for them, but it's possible you could be seeing one of them in your code..not sure.

[kmanalo]

@jacobwilliams I will give this approach a shot and let you know how it goes.

[zbeekman]

I wonder if it would be worth the hassle to add some kind of direct ragged edge string array API… I can’t quite picture what it would look like at the moment, without publicizing a new ragged edge string array derived type, which is likely beyond the scope of this project.

[jacobwilliams]

I think it would have to look like @kmanalo's string type above. Then we would have json_get, etc. routines for arrays of this type. I actually thought about something like this before, but decided against it. Perhaps one day, Fortran will let us have arrays of allocatable strings...

[zbeekman]

yeah exactly what I was thinking… probably best to leave it alone for now.

btw, auto-deploy almost done :-)

[kmanalo]

I tried your suggestion and it worked out very well.

Thanks, this was helpful in a non-string context also, in working with double precision arrays. I had a similar need to avoid the array constructor there also.

Also, I'm a dummy, this may be more of a semantic issue: every time I read the comment "don't need this anymore" in the suffix to nullify(), I totally thought that meant that one didn't need to use nullify anymore. I get it now.

[jacobwilliams]

Ha ha! Yeah, I was just saying that the pointer is no longer needed once the data it is pointing to has been added to the outp structure. I nullify it just to be safe (for example, you might want to reuse it later on). Note that the json_destroy routine deallocates the memory and destroys the linked list, which is not appropriate here for array (since the data itself is now part of outp), but would be appropriate to call for outp, when you are finished with it, in order to avoid a memory leak.

Happy to help!

[zbeekman]

Speaking of memory leaks, I wonder what it would take to get a reference counting scheme going, for better automatic memory management… it seems that there is a rather large burden on the end user to ensure that they cleanup after themselves to avoid memory leaks, but at the same time ensure that they don’t create any dangling pointers.

[jacobwilliams]

For the json_file class it might be fairly easy I think (we could even try to use a final method...although last time I tried that it was buggy). However, if you are using json_value pointers to build your own structure, I'm not sure how that would work...it could be tricky. It may be possible...but I've not thought in detail how that would work with Fortran pointers.