building gpgme for Android dies on install

[Hans-Christoph Steiner]

On 08/16/2012 02:36 PM, Hans-Christoph Steiner wrote:
> 
> On 08/16/2012 03:31 AM, Marcus Brinkmann wrote:
>> On 08/16/2012 03:35 AM, Hans-Christoph Steiner wrote:
>>>
>>> I believe I found the source of the trouble. The various *-config
>>> scripts all return values based on the assumption that everything will
>>> be installed before being used.  Since we're cross-compiling, that's not
>>> the case.
>>
>> pkgconfig and its imitations (such as our config scripts) share this
>> problem with libtool.
>>
>>> So now I'm trying to force things by using sed to replace prefix= in the
>>> various *-config scripts, but I haven't find a good time in the process
>>> to do that, and a good pattern to guarantee the results needed...
>>
>> Don't forget the .la files :)  That's actually the canonical way to do
>> it, I think (and yes, it's a PITA).  For GPGME and friends, you can also
>> just build and install for the cross-compilation build paths, and
>> use the --with-gpg and --with-gpgsm etc. options to specify the
>> hard-coded target file names of the crypto engines.  The result is
>> relocatable except for the .la and -config files, which need to be fixed
>> up anyway.  (So, if you want to develop on build and target, you
>> absolutely need two copies of those files, no matter what you do).
> 
> Ah yes!  That was the trick to get everything to build.  I set --prefix=
> on all of the libs to what I had been using for  $(DESTDIR)$(prefix) and
> the whole build worked.  I'll test it now.  The only question I have is
> that setting --prefix= in gnupg will alter where gpg-agent puts its
> socket file, no?

And changing prefix= to the whole thing on 'gnupg' means it looks in the
wrong place for dirmngr and friends. Looks like I'll need a sed hack
after all.

.hc