Arf... Indeed that was my problem. I didn't define the exported function manually at linking time (using /export with the cl) but directly in the C file using PD_INTERNAL because the approach is similar to the one presented in the MSCV documentation to create dll (https://msdn.microsoft.com/en-us/library/ms235636.aspx) and it worked pretty well until now... 

So to solve my problem, I have two solution: using my own EXPORT definition or generating dynamically the name of the function to export with CMake. Thanks a lot! 
PS: I still don't understand why it worked for some symbols but not for other...  

Thanks a lot !


2018-05-23 15:24 GMT+02:00 Christof Ressi <christof.ressi@gmx.at>:
so I hope you're not putting EXTERN in front of your external's setup function and defining PD_INTERNAL yourself... that wouldn't be a good idea :-)

> Gesendet: Mittwoch, 23. Mai 2018 um 15:15 Uhr
> Von: "Christof Ressi" <christof.ressi@gmx.at>
> An: "Pierre Guillot" <guillotpierre6@gmail.com>
> Cc: pd-dev <pd-dev@lists.iem.at>
> Betreff: Re: [PD-dev] pd.lib & msvc - missing symbols
>
> > and so **PD_INTERNAL** must also be used if we want to export properly the function `setup_class()` using **EXTERN**.
>
> with 'setup_class()' you mean the setup function of your external? it is exported with the /export:testobj_setup option to the linker.
>
> when you build an external, PD_INTERNAL must *not* be defined because you want to *import* the API functions from Pd.
>
> otoh, PD_INTERNAL must be defined if you build the pd.dll with MSVC, because there you want to *export* the API functions.
>  
>
> Gesendet: Mittwoch, 23. Mai 2018 um 14:09 Uhr
> Von: "Pierre Guillot" <guillotpierre6@gmail.com>
> An: "Christof Ressi" <christof.ressi@gmx.at>
> Cc: pd-dev <pd-dev@lists.iem.at>
> Betreff: Re: Re: [PD-dev] pd.lib & msvc - missing symbols
>
> Hi Christof,
>  
> I'm not a specialist of Windows so I can miss something important. 
> How do you specify the target machine? Are you sure you compiled against the 64bit version of pd.lib? 
> Because I didn't check the symbols of the 32 bit distribution, perhaps they are available. 
>  
> Another question: I assumed that **_WIN32** must be defined (isn't it?) and so **PD_INTERNAL** must also be used if we want to export properly the function `setup_class()` using **EXTERN**. 
> I see that you don't define any of these constants, are they useless in your case? And why? 
>  
>  
> 2018-05-23 13:40 GMT+02:00 Christof Ressi <christof.ressi@gmx.at[mailto:christof.ressi@gmx.at]>:this is odd. I made a simple Pd external which uses s_list and s_signal with MSVC in two versions:
> one links against Miller's pd.lib, the other links against the pd.lib I get when building from source with MinGW and autotools.
> Both versions compile, link and load just fine with either Pd version.
>
> I'm using the Developer command prompt, these are my commands:
>
> > cl /W3 /DNT /DPD /nologo /I. /IPDSOURCEPATH /c testobj.c
> > link /dll /export:testobj_setup testobj.obj oldnames.lib kernel32.lib PDBINPATH\pd.lib
>
> Christof
>  
>
> Gesendet: Mittwoch, 23. Mai 2018 um 12:01 Uhr
> Von: "Pierre Guillot" <guillotpierre6@gmail.com[mailto:guillotpierre6@gmail.com]>
> An: "IOhannes m zmoelnig" <zmoelnig@iem.at[mailto:zmoelnig@iem.at]>
> Cc: pd-dev <pd-dev@lists.iem.at[mailto:pd-dev@lists.iem.at]>
> Betreff: Re: [PD-dev] pd.lib & msvc - missing symbols
>
> > however, it would be nice if the shipped pd.lib (which uses the MinGW
> > toolchain and doesn't have the MSVC tools) would contain all the
> > required symbols.
>  
> Yes, what seems strange is that s_list and s_signal were defined in the shipped pd.def but not in pd.lib...
>  
> > otoh, i do think that those global variables (`e.g. `s_float`) shouldn't
> > be used in the first place: use the `gensym()` equivalents.
>  
>
> Why? 
> I used these variable because I assumed that they are more efficient than using `gensym()` (and as they are defined EXTERN I also assumed that they can be used). 
> But yes perhaps a better solution would be to generate and store the symbols in the new method of the object using `gensym()` (and it will be the same but for the sustainability of old codes, these symbols must be included).
>  
> Thanks
>  
> 2018-05-23 11:38 GMT+02:00 IOhannes m zmoelnig <zmoelnig@iem.at[mailto:zmoelnig@iem.at][mailto:zmoelnig@iem.at[mailto:zmoelnig@iem.at]]>:On 2018-05-23 11:29, Pierre Guillot wrote:
> > But there are missing symbols: on Windows
> > 64bit I can't use s_list and s_signal for example. I solved the issue by
> > creating my own pd.lib and pd.def from pd.dll. So if you encounter the same
> > issue, you can simply do
> >
> >> dumpbin /EXPORTS pd.dll > pd.exports
> >> lib /def:pd.def /out:pd.lib
>
> the problem with this is that it requires MSVC tools
> now, the target audience are MSVC-users (who else needs that pd.lib...),
> so they already have it.
> however, it would be nice if the shipped pd.lib (which uses the MinGW
> toolchain and doesn't have the MSVC tools) would contain all the
> required symbols.
>
> otoh, i do think that those global variables (`e.g. `s_float`) shouldn't
> be used in the first place: use the `gensym()` equivalents.
>
> gasdmr
> IOhannes
>
> > I hope it could help some of you. Perhaps we could add a documentation
> > somewhere about it... (I don't know if it useful...) I can help if needed.
>
>
>
>  
> _______________________________________________
> Pd-dev mailing list
> Pd-dev@lists.iem.at[mailto:Pd-dev@lists.iem.at][mailto:Pd-dev@lists.iem.at[mailto:Pd-dev@lists.iem.at]]
> https://lists.puredata.info/listinfo/pd-dev[https://lists.puredata.info/listinfo/pd-dev]
>  _______________________________________________ Pd-dev mailing list Pd-dev@lists.iem.at[mailto:Pd-dev@lists.iem.at] https://lists.puredata.info/listinfo/pd-dev[https://lists.puredata.info/listinfo/pd-dev]
>
> _______________________________________________
> Pd-dev mailing list
> Pd-dev@lists.iem.at
> https://lists.puredata.info/listinfo/pd-dev
>