On 7/26/07, Alan W. Irwin <[EMAIL PROTECTED]> wrote:
On 2007-07-26 12:32-0400 Brandon Van Every wrote:
>
> So why will the experimental testing method be widely used? And when
> it is used, why will people report their results?
This is standard fare in most free software releases these days. To name
just a few major projects off the top of my head, the Linux kernel, Debian,
KDE, and GNOME projects do this,
These have a critical mass of tweakerheads that CMake does not have.
People who think it's important to recompile their kernel and their
libraries for their specific CPU and so forth.
and there are huge numbers of minor
projects (such as PLplot) that do this as well.
That would be more comparable to CMake's current popularity and scope.
The way this works is a given software package puts out a testing release,
and the cutting-edge types who are attracted by the new features in the test
release, test it, report bugs, etc. Most software users are not
cutting-edge types and don't bother with testing releases and apparently you
are part of that majority. :-)
You can reasonably expect a build system engineer to value stability
over the bleeding edge. I wager you'll find that true of people in
the CMake community.
Nevertheless, the testing release model
normally works well because there are a substantial minority that do like to
be cutting edge. For example, with PLplot our testing releases have
substantial popularity judging by their download rate statistics, and we do
get valuable feedback from such early-adopter users. Since we value that
feedback we make it extremely easy for users to try testing distributions of
PLplot, and I call on KitWare to do the same with the modules.
I see a difference: PLplot is an end product, not an underlying
configuration tool affecting many applications and libraries. Who
tweaks Autoconf? Perhaps a survey of the release methodologies of
other major configuration tools is in order, i.e. Ant, SCons.
I think it is very important that any experimental releases have no
effect on official CMake installations at all. The end user should
have to make a conscious choice to allow experimental stuff to
operate. Configuration options of the form
USE_EXPERIMENTAL_MODULE_MODULENAME might do the trick. They'd be OFF
by default. Of course, this is a fairly conservative approach and
will keep the testing from being widespread. But I think for a build
system, conservative has to be the official default. Otherwise
CMake's reputation for building things reliably is jeopardized.
Cheers,
Brandon Van Every
_______________________________________________
CMake mailing list
[email protected]
http://www.cmake.org/mailman/listinfo/cmake
