As part of the
libgcj project (55), I had to incorporate the
zip program into our source tree. Since this particular program is only used in one part of the build, and since this program was already fairly portable, I decided to take a quick-and-dirty approach to autoconfiscation.
First I read through the `README' and `install.doc' files to see how
zip is ordinarily built. From there I learned that
zip came with a `Makefile' used to build all Unix ports (and, for the initial autoconfiscation, Unix was all I was interested in), so I read that. This file indicated that
zip had few configurability options.
ifnames on the sources, both Unix and generic, confirmed that the
zip sources were mostly self-configuring , using system-specific `#defines' ---a practice which we recommend against; however for a quicky-and-dirty port it is not worth cleaning up:
| || |
$ ifnames *.[ch] unix/*.[ch] grep ^__ head __386BSD__ unix/unix.c __CYGWIN32__ unix/osdep.h __CYGWIN__ unix/osdep.h __DATE__ unix/unix.c zipcloak.c zipnote.c zipsplit.c __DEBUG_ALLOC__ zip.c __ELF__ unix/unix.c __EMX__ fileio.c ttyio.h util.c zip.c __FreeBSD__ unix/unix.c __G ttyio.h __GNUC__ unix/unix.c zipcloak.c zipnote.c zipsplit.c
Based on this information I wrote my initial `configure.in' , which is the one still in use today:
| || |
AC_INIT(ziperr.h) AM_INIT_AUTOMAKE(zip, 2.1) AM_MAINTAINER_MODE AC_PROG_CC AC_HEADER_DIRENT AC_DEFINE(UNIX) AC_LINK_FILES(unix/unix.c, unix.c) AC_OUTPUT(Makefile)
The one mysterious part of this `configure.in' is the define of the `UNIX' preprocessor macro. This define came directly from
zip 's `unix/Makefile' file;
zip uses this define to enable certain Unix-specific pieces of code.
In this particular situation, I lucked out.
zip was unusually easy to autoconficate. Typically more actual checks are required in `configure.in' , and more than a single iteration is required to get a workable configuration system.
From `unix/Makefile' I also learned which files were expected to be built in order to produce the
zip executable. This information let me write my `Makefile.am' :
| || |
## Process this file with automake to create Makefile.in. ## NOTE: this file doesn't really try to be complete. In particular ## `make dist' won't work at all. We're just aiming to get the ## program built. We also don't bother trying to assemble code, or ## anything like that. AUTOMAKE_OPTIONS = no-dependencies INCLUDES = -I$(srcdir)/unix bin_PROGRAMS = zip zip_SOURCES = zip.c zipfile.c zipup.c fileio.c util.c globals.c \ crypt.c ttyio.c unix.c crc32.c crctab.c deflate.c trees.c bits.c ## This isn't really correct, but we don't care. $(zip_OBJECTS) : zip.h ziperr.h tailor.h unix/osdep.h crypt.h \ revision.h ttyio.h unix/zipup.h
This file provides a good look at some of the tradeoffs involved. In my case, I didn't care about full correctness of the resulting `Makefile.am' -- I wasn't planning to maintain the project, I just wanted it to build in my particular set of environments.
So, I sacrificed `dist' capability to make my work easier. Also, I decided to disable dependency tracking and instead make all the resulting object files depend on all the headers in the project. This approach is inefficient, but in my situation perfectly reasonable, as I wasn't planning to do any actual development on this package -- I was simply looking to make it build so that it could be used to build the parts of the package I was actually hacking.