Re: Howto override the argument names of certain class member functions

On 27/06/2020 14:41, William S Fulton wrote:
>
>
> On Sat, 27 Jun 2020 at 12:13, Terry Barnaby <[email protected] 
> <mailto:[email protected]>> wrote:
>
>     Hi William,
>
>     Thanks for the reply. Notes below in the text.
>
>     Terry
>     On 27/06/2020 11:52, William S Fulton wrote:
>>     Hi Terry
>>
>>
>>     On Sat, 27 Jun 2020 at 10:42, Terry Barnaby <[email protected]
>>     <mailto:[email protected]>> wrote:
>>
>>         Hi William,
>>
>>         I have a large class library. There are multiple functions
>>         that have a, say, "(Channel& chan)" argument. In most cases
>>         these are used to return values, but in some cases the
>>         functions modify the contents of the object who's reference
>>         is passed. So I cannot generically use the %template
>>
>>
>>     I think you meant %apply.
>
>     Yes, sorry, I meant %apply.
>
>
>>
>>         system as that only applies to the type and names of
>>         arguments passes, there is not ability to say a particular
>>         function is just using that argument as an output return. A
>>         lot of this libraries code is auto generated from a IDL and I
>>         generate a set of modified *.h files with the argument names
>>         changed to OUTPUT at the appropriate places but quite a few
>>         class member functions are not like that.
>>
>>     So I think you are saying you want to %apply these typemaps for
>>     parameters with the same name and type, but only on a select few
>>     functions and you don't want to modify the header files. The best
>>     solution SWIG has is to use %apply and %clear as Rui alluded to.
>>     This will work to some degree as the %apply will apply for all
>>     the functions in a header file you have include with %include.
>>     You can then use %clear after the %include. If you want to
>>     selectively apply the typemaps to parameters with the same type
>>     and name in one header file, then the next best solution is to
>>     use %ignore and then add the functions back in with %extend,
>>     renaming the parameters to something else. You imply that you
>>     only have a few of these that you don't want the typemaps to
>>     apply to, so this might be workable for you.
>
>     Unfortunately this is a large library and within header files
>     there are multiple functions which I would need to do this on, so
>     it can't be done using %apply and %clear unless a re-write the
>     header file into SWIG. I could use %ignore and %extent but I think
>     this would get quite messy especially as most of the functions are
>     members of classes. I will have a play with that.
>
> Use a macro which does the %ignore and %extend in a simple call. 
> That'll be easy to use unless you have a huge number and variety of 
> function signatures.
>
>
>>
>>     Looking at the original post, you have this in a header file:
>>
>>       class DataFile {
>>        public:
>>            Error setInfo(const Info& info);    // Input only reference
>>            Error getInfo(Info& info);        // Output only reference
>>        };
>>
>>     the types are NOT the same as one is const and the other is not
>>     const. If you look at the typemap rules, you'll see that if you
>>     have a typemap for const Info&, it will not apply to non-const
>>     Info&. However a non-const Info& typemap it wil apply to const
>>     Info&. So if you want to consistently apply typemaps to const and
>>     non-const, then solution is easy, something like:
>>
>>     %typemap(argout) const Info& OUTPUT "" // does nothing
>>     (effectively clears the typemap)
>>     %typemap(argout) Info& OUTPUT "/* some code for argout*/"
>>     %apply Info& OUTPUT { Info& info };
>>     %apply const Info& OUTPUT { const Info& info };
>>
>>     The main reason SWIG does not have function and parameter name
>>     combinations for typemaps is there is no C++ scoping syntax to
>>     get at the scope of a parameter name from outside of the
>>     function. I can elaborate more if anyone is really interested.
>
>     That was a bit of a simplification, and as you say didn't show the
>     issue! In effect there are multiple functions like getInfo. I do
>     have many functions that use the "const Type&" which as you say
>     the %typemap, %apply system works well on.
>
> Perhaps your API needs to be improved. I can't see why you would want 
> to use the OUTPUT typemaps on a non-const reference type. Are you 
> needing to have OUTPUT typemaps for non-const references?
>
>     Well I would have thought it would be possible in SWIG even if at
>     a simple pre-processing regular expression string replacement
>     level which is how I have effectively handled it now ? I would
>     have thought though that SWIG could do this internally it must
>     know: <Namespace>::<class>::functionName::[Arg Type/name list]
>     from its C++ parsing to nodes ?
>
> Something like that would indeed be possible and would be a powerful 
> extension to SWIG. I think if there was valid C++ syntax and scoping 
> for specifing a parameter, it would have been implemented ages ago. I 
> can think of various approaches to providing this matching capability. 
> If you'd like to discuss and contribute to an implementation of it, 
> please raise a Github issue.
>
> Regards
> William

Ok, will try and find the time to have a look at the SWIG code a bit 
more to try and understand how it works to see if I can see a way to do 
this.

_______________________________________________
Swig-user mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/swig-user