Re: configure wrongly caches results

[Ralf Wildenhues]

Hi Patrice, Stepan,

* Stepan Kasal wrote on Tue, Sep 27, 2005 at 10:23:05AM CEST:
> On Mon, Sep 26, 2005 at 12:51:53AM +0200, Patrice Dumas wrote:
> > So it seems that there is a kind of caching that cannot be disabled and
> > that is preventing such tests to work.
> 
> Indeed, you can switch off the cache _file_, so that it is not read at
> the beginning and written at the end, but you cannot switch off the
> "internal" caching of autoconf.
> 
> > checking for netcdf.h... no
> > checking for netcdf.h... (cached) no
> 
> ... while you would expect:
> 
> checking for netcdf.h... no
> checking for netcdf.h... yes
> 
> which really looks weird, so you added an echo to get:
> 
> checking for netcdf.h... no
> checking for netcdf.h... found with libnc-dap
> yes
> 
> Anyway, the internal cache is a limitation of current autoconf.
> 
> I think there is a very natural workaround/fix:
> 
> AC_CHECK_HEADERS([hdf2/netcdf.h])
> AC_CHECK_HEADERS([libnc-dap/netcdf.h])

Actually, in this special case, I'd require the user to pass
  CPPFLAGS=-I/usr/include/hdf2
or
  CPPFLAGS=-I/usr/include/libnc-dap

and appropriate LDFLAGS.  Why?  This looks like a nonstandard scenario.
netcdf.h might be below /opt/hdf2, or most anywhere else (it's somewhere
different on this system), and just picking the first one the script
finds might just lead to breakage unless you also ensure you find the
library that fits to this one.

See pkgconfig for a part-solution of this problem.

Cheers,
Ralf