gnucash/doc/README.build-system

88 lines
3.8 KiB
Plaintext
Raw Normal View History

-*-text-*-
===========================================================================
Automake
GnuCash uses automake to handle the build process. Make sure you
understand what automake provides/requires before you add anything
particularly fancy to the Makefile.am files.
Most of the time, a "make" or "make install" will automatically DTRT
and regenerate Makefiles or Makefile.ins, re-run configure, regenerate
configure from configure.in, etc. whenever needed. However, it is
possible, if you make a mistake when working on the Makefile.am's to
leave yourself in a situation where automake won't work because a
Makefile is broken. To fix this, just run:
./autogen.sh
./configure <your usual options>
That should sledgehammer the problem bit back into place.
===========================================================================
Adding files to AC_OUTPUT in configure.in.
Please do not add any non-makefiles to AC_OUTPUT unless you're
*absolutely* sure that it's safe and necessary to do so. If you're
not sure, then please use the "generation with sed by hand approach"
that we use for other non-makefiles like src/scm/bootstrap.scm. See
src/scm/bootstrap.scm.in and src/scm/Makefile.in for details.
The reasoning behind this is that there are often variables that
autoconf uses and replaces via the @@ mechanism that are recursively
defined in terms of other variables. For example, @datadir@ might
expand into $prefix/share, which itself contains an unexpanded
variable. Unless the @datadir@ replacement is performed on a file
that will eventually be processed by make, there's no guarantee that
the variable will *ever* be fully expanded and then your code will
break.
To get around this, we handle all non-makefiles (with a few notable
exceptions) manually with sed. For example, we have
src/scm/bootstrap.scm.in which is processed in src/scm/Makefile.in
according to the following rule:
## We borrow guile's convention and use @-...-@ as the substitution
## brackets here, instead of the usual @...@. This prevents autoconf
## from substituting the values directly into the left-hand sides of
## the sed substitutions. *sigh*
bootstrap.scm: bootstrap.scm.in
rm -f $@.tmp
sed < $@.in > $@.tmp \
-e 's:@-VERSION-@:${VERSION}:g' \
-e 's:@-GNC_CONFIGDIR-@:${GNC_SHAREDIR}:g' \
-e 's:@-GNC_SHAREDIR-@:${GNC_CONFIGDIR}:g'
chmod +x $@.tmp
mv $@.tmp $@
This approach guarantees that the variables referred to will be
properly expanded at the right times.
Note that on OSX, you MUST have the GNU version of sed installed
via fink - the BSD version installed by default will not work with
some of the substitution operators used in the gnucash build.
The only non-Makefiles files that must be handled directly by
AC_OUTPUT are files that refer to variables that, when expanded, may
have makefile-hostile characters like '#' in them like
INCLUDE_LOCALE_H. These may need to be replaced directly in the
relevant file via AC_OUTPUT since going through a makefile would break
the makefile when it interprets the '#' as the beginning of a comment.
You'd have something like this:
FOO = "#include <locale.h>"
If you end up in a situation where you need to refer to both these
makefile-hostile variables and non-makefile hostile variables in the
same file, please restructure things (breaking the file up if
necessary) so that only the makefile hostile variables are in the
files being handled by AC_OUTPUT directly.
===========================================================================
It is not safe to use $prefix in configure.in for anything other than
an unexpanded reference. If the user doesn't specify a --prefix, then
it'll be set to NONE until the end of the configure process.
===========================================================================