Re: Making camlibs and iolibs relocatable

On Tue, 28 Sep 2021 13:41:05 +0100
Jim Easterbrook <[email protected]> wrote:

> I've recently added "binary wheels" to my Python interface to 
> libgphoto2. These include pre-built copies of libgphoto2 v2.5.27 and
> are much quicker and easier for a user to install as they don't have
> to install libgphoto2 (development headers) and compile the interface
> files.
> 
> One problem I encountered was the hard coding of absolute paths into 
> where libgphoto2 looks for its camlibs and iolibs. Luckily the
> CAMLIBS and IOLIBS environment variables solve that problem, but I'd
> rather they weren't labelled as for "broken/test installations".

I have just changed the wording for IOLIBS and CAMLIBS to: "You only
need this with non-standard installations."

If it helps you in any way, we now (i.e. in 2.5.27.1 in git since commit
94d9912e from 2021-10-12) a way to build libgphoto2 and libgphoto2_port
such that the default location it looks for camlibs and iolibs can be
set at configure time:

     configure \
         camlibdir=/where/to/install/the/camlibs \
         DEFAULT_CAMLIBS=/where/to/find/the/camlibs \
         iolibdir=/where/to/install/the/iolibs \
         DEFAULT_IOLIBS=/where/to/find/the/iolibs

The defaults should be sane:

  * If camlibdir= is not set at configure time, use a default location
    based on ${libdir} to find out where to install the camlibs.

  * If DEFAULT_CAMLIBS is not set at configure time, use the value of
    ${camlibdir} to find out where to look for the camlibs.

  * Analogous behaviour for iolibs.

What we do not have at this time is a way to change the driver lookup
location at runtime with an API function call. That has been proposed
in https://github.com/gphoto/libgphoto2/pull/534 but I have
not thought much about it so far. I certainly do not want to hurry
changes to the public API and ABI as those tend to stick for a long
time.

Uli