Re: libusb and libusb-compat: conflict

["Mike (mwester)" <mwester@dls.net>]

Koen Kooi wrote:
...
> Some background: libusb1 is the rewrite of libusb which brings up
> goodness as better performance and more powersaving, but its ABI is
> incompatible with libusb. That is why the libusb people create
> libusb-compat, it's a drop-in replacement for libusb.
> 
> When I added it to OE I made sure the runtime situation worked and
> changed all packages to libusb-compat. It seems I missed a few.

Grep comes up with a bunch, but probably a lot of those are just older
recipes -- dfu-util and openocd are definitely broken, though.

> It should be safe to change everything over to libusb-compat, unless
> your favourite apps abuses private libusb API (as gnuradio does).

I encountered simple linker failures (symbols not found).  I'll get rid
of libusb out of staging entirely and see if dfu-util and openocd will
link correctly against libusb-compat.  I can test dfu-util, but I hope
someone else can test openocd.

> And there's no such thing as CONFLICTS in OE, only RCONFLICTS.

Bummer.  That would mean that we would need to switch to
"virtual/libusb" I guess.  That seems a lot of effort for something
that's obsolete -- it might be better to warn people if libusb is being
built instead of libusb-compat.  I wonder if this might be a use case
for a "WARNING" metadata item in a bitbake recipe.

I will do nothing with any metadata then except attempt to build
dfu-util and openocd with libusb-compat.

> regards,
> 
> Koen

Mike (mwester)