expose lilv_free() with features, for windows compatibility

[x42]

Various functions e.g. abstract_path() make the caller responsible for freeing the returned value with free().

On Windows this is not an option. Memory allocated by a library must be free()ed by that library/module

[x42]

I was too quick.

A plugin does not link against liblilv and cannot call lilv_free() directly. A plugin receives abstract_path(), absolute_path() via the LV2_State_Map_Pathfeature. Is free()` needs to be part of the provided feature.

[drobilla]

See https://gitlab.com/lv2/lv2/tree/free-path and 055e645

[drobilla]

@x42 I've updated this on current master, and it seems fine as far as the (relatively limited) tests go. Did you have a specific motivating case it would be good to check before merging this stuff?

[x42]

All state restore methods that use map_path->abstract_path(). I've #idef'ed out the free() call in many plugins like

https://github.com/x42/convoLV2/blob/ff291a3d5e60ddbcf9458c3ad24d4a06c8252acb/lv2.c#L473-L475

[drobilla]

I don't have a setup at the moment that can build stuff outside lv2kit and friends on Windows. Let me rephrase: is there a specific motivating case you would like to check before this stuff gets merged? :)

[drobilla]

API-wise, as is so often the case, I really regret LV2 not having the ability to just jam stuff on the end of structs, getting a feature for this is pretty annoying, but... given reality, this seems fine to me.

[x42]

What is the envisaged way to test at compile-time if LV2_STATE__freePath is available?

[drobilla]

I didn't think about that. The LV2 version I guess?

[drobilla]

You could also just keep the fallback ifdeffed out as you did there I suppose.

[x42]

is the URI guarenteed to be a #define? if so

#ifdef LV2_STATE__freePath
...
#endif 

might work

[drobilla]

Same as anything else, really. I guess having an LV2_VERSION declared in the header would be handy, but it's annoying to have generated stuff in the headers, and that's what build systems are for...

Anyway, I guess I can stick this stuff on master and it can get hammered out a bit in practice in Ardour before release since IIRC you aren't using releases anyway.

[x42]

You could also just keep the fallback ifdeffed out as you did there I suppose.

I will do that for a while, at least until debian/stable picked up new LV2 headers.
Also #ifdef LV2_STATE__freePath looks like a reasonable test.

As far as I'm concerned, this issue can be closed. Thanks!

[drobilla]

Addressed in ac562ca