Externals don't work in plugin (only in standalone)

[FlachyJoe]

Hi,
I tried to add the purr-data external path to PlugData ( /opt/purr-data/lib/pd-l2ork/extra ) and C coded libraries fail to load.
Adding
[iemlib/init] : undefined symbol s_float
[iem_ambi/ambi_decode] : undefined symbol s_list
[lyonpotpourri/bv_play~] : undefined symbol garray_class

NB: These types are defined in Libraries/pure-data/src/m_pd.h

[timothyschoen]

Do these libraries load on vanilla pd? PlugData has the same core as pd-vanilla, it's possible that some externals for purr-data don't work because they use a different pd core.

[FlachyJoe]

They load in vanilla.

[timothyschoen]

Then they should load in PlugData, I'll check it out!

[timothyschoen]

This problem is caused by the fact that the externals are compiled without the PDINSTANCE and PDTHREAD flag, while PlugData needs to be compiled with those flags. I need to compile with PDINSTANCE because we need to be able to use multiple instances to make it work as a plugin. Camomile suffers from the same issue, there's unfortunately not much I can do about it.

I'll leave the issue open, so people can see that this doesn't work yet.

[timothyschoen]

Externals now work in the standalone! I'll have to do some more testing but it's looking good.

For the plugin, support for externals is still a problem because we need to have multiple instance support.

PS: For M1 Mac users: remember to use PlugData in Rosetta if you want support for x64 externals.

[FlachyJoe]

Hi,
I still have the error in Standalone :

CMakeCache.txt

[timothyschoen]

Looks like I made a mistake in the cmakefile, I'll fix it right now.

[timothyschoen]

Can reproduce that it doesn't work on linux

[timothyschoen]

It works now! Turns out I needed to pass the -export-dynamic linker flag on Linux to get this kind of backwards linking to work.

[timothyschoen]

I don't know if you're building from source, but if you're not I'll create a release soon when I'm done with some other bug fixes!

[timothyschoen]

I see that windows has the same problem (unfortunately not with the same solution), so I'll look into that as well. My bad for assuming that it'd just work on all platforms, oops!

[FlachyJoe]

I confirm it works on linux. Thank you !

[timothyschoen]

I may have found a solution...

If we define PD_THIS with a function, get_this() which is defined in an .c file instead of a header, we can make any new externals that are compiled work both with and without PDINSTANCE.

I still need to investigate this further, but it would mean that externals only need to be recompiled against the new header in order to work, without having to change any source code at all. If it works, I hope the pd devs will accept this change. It would also mean that Camomile could start supporting this (also by just recompiling!).

Because the PD_THIS external in existing externals already points to the &pd_maininstance symbol (which still exists), it shouldn't break anything.

EDIT: Okay, so I think the only problem would be that newly built externals would stop working in old pd versions. So maybe this is not a perfect solution after all... It's an interesting discussion. It might be possible to add some kind of fallback mechanism for this.

EDIT 2: I've managed to make externals built with new the PDINSTANCE system work in an unmodified pd-vanilla! Basically, you can just add a backup definition for these functions into your external, since dlopen won't overwrite these when it finds a duplicate. This can be done automatically in m_pd.h.

[alfonso73]

Edit 2 almost deserve a "wont fix" tag removal from the issue :-D
Great stuff tho! Finger crossed