Arrays in a C struct

I noticed in reading the code that the Julia struct

struct ReadStatVariable
    readstat_types_t::Cint
    index::Cint
    name::Ptr{UInt8}
    format::Ptr{UInt8}
    label::Ptr{UInt8}
    label_set::Ptr{Void}
    offset::Int64
    width::Csize_t
    user_width::Csize_t
    missingness::ReadStatMissingness 
end

does not align with the C declaration in readstat,h

typedef struct readstat_variable_s {
    readstat_type_t         type;
    int                     index;
    char                    name[256];
    char                    format[256];
    char                    label[1024];
    readstat_label_set_t   *label_set;
    off_t                   offset;
    size_t                  storage_width;
    size_t                  user_width;
    readstat_missingness_t  missingness;
    readstat_measure_t      measure;
    readstat_alignment_t    alignment;
    int                     display_width;
} readstat_variable_t;

The problem is that the fixed-size char arrays, like name do not correspond to Ptr{UInt8} or Cstring in Julia. If you want the structs to align in memory you need to insert a Julia struct that has the correct width. The Clang package at one time did this by creating a Julia struct with 256 UInt8 fields and inserting that as the name field. However, that approach is rather tedious.

In this case there is no need to define the Julia struct because the memory is allocated in the C code and passed by reference. I can prepare a PR if you wish.

Yes, that would be great!

Just one warning: we use a quite old version of the readstat lib. I've started work to get us up to speed with the latest version by using BinaryBuilder and compiling a new version, but that is not done yet. I'm not sure whether there are any changes to these types upstream, but just something to keep in mind.

I looked at the main repo and it seems that they have changed that struct quite a few times so if we really want to match it in Julia, we'd have to be careful with checking the version of the linked library. However, there are quite a few getter functions defined so it is safer and simpler to just rely on them instead of trying to match the struct in Julia.

So that maybe suggests that we don't even need to define these structs at all, but can just pass a pointer around and then use the getter functions to access things, right? That seems an all around safer approach.

@davidanthoff

@davidanthoff Yes. I think it will be safer overall to use the getter functions and not try to emulate the C structs. I am working on a PR.

I was wrong about passing readstat_value_t as an opaque pointer. It is passed by value in the C code so there must be a Julia struct like ReadStatValue, which must have the right size, as you have ensured.

Well, I didn't write any of this code ;) This julia package was original created by Evan Miller, who is the creator for the ReadStat C library. He is no longer interested in maintaining the julia package, so he transferred the repo over to me because I had helped out getting it to run on Windows a while back. So really he deserves the credit for any of this here.

@davidanthoff

@davidanthoff Would you be willing to transfer ownership to the JuliaData group?

I'm happy to keep maintaining it. I like the model where someone is on the hook if things don't work, and for now I'm happy to be that person ;) If I run out of time to do that, we could move it over.

Navigation Menu

Repository navigation

Arrays in a C struct #33

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Arrays in a C struct #33

Description

Activity

davidanthoff commented on Jan 13, 2018

andreasnoack commented on Jan 16, 2018

davidanthoff commented on Jan 16, 2018

dmbates commented on Jan 16, 2018

dmbates commented on Jan 16, 2018

davidanthoff commented on Jan 16, 2018

dmbates commented on Jan 16, 2018

davidanthoff commented on Jan 16, 2018

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions