Re: conditionally compiling C++ code

[Sander Niemeijer]

On dinsdag, apr 29, 2003, at 18:04 Europe/Amsterdam, Patrick Welche 
wrote:

> On Tue, Apr 29, 2003 at 02:31:54PM +0200, Sander Niemeijer wrote:
> > I was just wondering, but aren't the --with-PACKAGE/--enable-FEATURE
> > options of configure meant for this?
> >
> > I have a package that can optionally make use of the HDF4 library
> > (which, if included, provides extra functionality to import/export 
> > data
> > in HDF4 format). I dealt with this in my configure.ac as follows:
> >
> > ---
> > AC_ARG_WITH([hdf4],
> >   [AC_HELP_STRING([--with-hdf4],[build HDF4 support into this
> > package])],
> >   [ac_cv_with_hdf4=$withval],
> >   [AC_CACHE_CHECK([use HDF4],ac_cv_with_hdf4,ac_cv_with_hdf4=no)])
> >
> > if test $ac_cv_with_hdf4 = yes ; then
> >   ... <Check for HDF4 libs and headers via environment variables that
> > specify their location and set HAVE_HDF4 definition and
> > ac_cv_have_hdf4> ...
> >   if test $ac_cv_have_hdf4 = no ; then
> >     AC_MSG_ERR([HDF4 libraries and/or header files are not found. Try
> > setting the appropriate environment variables.])
> >   fi
> > fi
> > ---
> > This approach forces you to explicitly say that you want to make use 
> > of
> > the HDF4 functionality and if so it will give an error if it could not
> > find the HDF4 libs/headers.
>
> Exactly, and I was suggesting that part should become part of say the
> AC_CHECK_LIB macro, so you wouldn't have to write it (yet again) in 
> your
> configure.ac.

In my opinion this wouldn't be a good idea. The HDF4 example above is 
maybe a good example to illustrate this. For HDF4 you would need four 
libraries (-lmfhdf -ldf -lz -ljpeg). In this case it it is much more 
logical to check whether we need to find and include all four libraries 
with one configure option, instead of four options for each of the 
libraries. This also hides the 'configure user' from the inter library 
dependencies. And as the --with-PACKAGE and --enable-FEATURE names 
already say, the options should be given in terms of packages or 
features (and not specific files), which is a good idea in my opinion.

Furthermore, in the case of my example there is a HAVE_HDF4 #define 
created that includes/excludes specific parts of my package's source 
code that deals with HDF4 import/export. Although it is possible to 
introduce similar HAVE_<libname> #defines for libraries, I think it is 
a bad idea to have conditional code based on the availability of a 
certain library. Within source code you should only have conditional 
parts based on feature usage (should we include HDF4 import/export 
functionality yes/no. It is in general a bad idea to derive whether to 
include a feature from the availability of a single library in my 
opinion (it is usually also linked to the availability of one or more 
header files for instance). It should always be the other way around), 
declarations (including the proper header files), and 
system/platform/compiler specifics (stuff like WORDS_BIGENDIAN).
As a result, if you leave out the HAVE_<libname> possibility, then you 
either don't need the library at all (in which case you don't include 
any checks for it and don't link to it) or you /do/ need it, in which 
case you should generate an error if it is not found (otherwise the 
compiler will do the complaining for you at link time). So the 
functionality you are proposing is only useful if it would generate a 
HAVE_<libname> #define to determine whether to include a certain 
feature of your package (which is a bad idea as I explained before).

Regards,
Sander