Re: Smart --with-package=DIR (for absentminded users)

[Ralf Wildenhues]

Hello Thomas,

* Thomas Lavergne wrote on Fri, Jul 06, 2007 at 03:15:58PM CEST:
>
>    My package is an ensemble of C files which are build in a static library 
> libbar.a and associated header bar.h. My library depends on another library 
> (foo) which should be pre-installed on the system.
>
>    At configure time, to make sure I have access to foo, I make 3 tests. 
> First AC_CHECK_HEADER([foo.h]), then 
> AC_CHECK_TYPE([aSymbolIn_foo],,[#include <foo.h>]) and eventually 
> AC_CHECK_LIB([foo],[aMedhodIn_foo]).
>
>    My question: strictly speaking (and IMHO) the last test is too much. The 
> building of a static library does not indeed use any linking. Isn't it a 
> bit too much to ask for LINKING (AC_CHECK_LIB) of a library function at 
> configure time when we won't use the linker at compile time?

But doesn't your package have a couple of test programs that are to be
linked at
  make check

time against (yours and) those libraries, to ensure proper operation of
your library?

> In (almost) real life, we might want to configure, build and install our 
> library although the programming team next door is not yet ok with the 
> implementation of libfoo.a (although we agreed on the interface available 
> in foo.h).

Well, your action-if-not-found for AC_CHECK_LIB need not be erroring
out.  It could be a warning, too.  Or it could be a warning if the user
specified --enable-development-mode.  Or you could just comment out the
test for the time being.

Hope that helps.

Cheers,
Ralf