On Fri, 10 Dec 1999, Andrew M. Kuchling wrote:
M.-A. Lemburg writes:
There were issues with zlib 1.0.4 and later ones. Also, many Linux distributions don't have the zlib header files installed.
For example, on RH6.0, zlib.h and zlib.a are in zlib-devel.XXX.rpm, and zlib.XXX.rpm only contains libz.so. On the other hand, anyone who's compiling Python should really have the various -devel RPMs
Exactly. The distro's *have* the headers -- it all depends on what you installed. I happen to have the headers on my system (because I installed zlib-devel, as AMK mentions).
installed. I'd argue against including it, because it might cause odd versioning problems. For example, what if I have PIL compiled against zlib1.1.2 (zlib is used for writing PNGs) and the Python binary includes zlib1.1.3? There might be hard-to-debug problems caused by calling the wrong symbol.
I totally agree.
... Just received Guido's email suggesting skipping compression in archives; not a bad idea. You'd use less CPU, but might do more I/O because you're reading more sectors off disk. There probably isn't much need for compression when the archive is on-disk; Java needed it because of applets.
There are all kinds of things that we can do here. Consider mmap'ing the archive into a shared memory segment, used by all the Python processes on the system... woo! :-) IMO, the standard distro can use zip files, and just bail if they are compressed, but Python cannot load zlib. Obvious failure with an obvious remedy. No big deal. As Guido also mentions, an installer can just bring along zlib if they want to use a compressed archive. i.e. their choice. Cheers, -g -- Greg Stein, http://www.lyra.org/