[Python-Dev] zlib not compiled by default

Jeremy Hylton jeremy@beopen.com
Fri, 14 Jul 2000 11:36:00 -0400 (EDT)

>>>>> "GvR" == Guido van Rossum <guido@beopen.com> writes:

  >> I *don't* think we need to maintain any such thing in the CVS
  >> tree -- only in the .tar.gz we let the world download.

  GvR> Agreed.

  GvR> I have to add that I'm +0 on this idea -- it would seem that if
  GvR> we distribute a sumo RPM distribution we'd help way more
  GvR> people.

I think it's significantly harder to create a .tar.gz file with a
configure script and a makefile that builds Python and all the
third-party libraries it depends on.  Does the zlib source belong in
the .tar.gz? How about OpenSSL? Tcl/Tk? libc? 

There are at least two problems with putting anything other than
Python in a .tar.gz that we distribute.  First, the user who grabs it
may already have some of the third-party extensions installed.
Second, we have a version-skew nightmare because each library is
updated at its own pace.

This is why Andrew is still correct!

>>>>> "AMK" == Andrew Kuchling <akuchlin@mems-exchange.org> writes:

  AMK> <broken-record>The correct fix is to make it easy to
  AMK> automatically download & install packages automatically;
  AMK> distributions then become simple lists.  "You want to do Web
  AMK> development?  Grab Cookie.py, wizard.py, DocumentTemplate, ..."
  AMK> "Numeric work?  NumPy, PhysicalValue, Snow,
  AMK> ..."</broken-record>

The distutils plus platform-specific package management like RPM and
Debian packages are the right solution.  We can produce a Python
package that indicates its dependence on readline, Tcl/Tk, or anything
else.  We can also provide a collection of Python packages via
distutils that addresses the needs of particular audiences -- Web
slingers, numerical analysts, etc.