On 08/10/2009 05:17 PM, Bill McGonigle wrote:
On 08/07/2009 02:54 AM, Rahul Sundaram wrote:Pointing it out on a review and restoring to calling the packages bad quality if people don't follow your controversial recommendation isn't going to scale at all.This is a good perspective, Ralf. Putting the same energy into individual reviews won't have as amplified an impact as convincing the packaging committee of problems.
I am member of the FPC, but ... I have failed to convince the FPC so far.
They are very easy to demonstrate. Commonly known cases are building gcc, binutils, gdb, firefox etc.I understand the theoretical value of a deterministic package build - I'm not aware of specific examples of where non-determinism has caused problems in Fedora, though I can imagine some.
Other cases are pretty easy to find. Actually, probably almost any non-trivial, complex package has such issues.
It's only the fact that most packages are trivial autotool-wise and the fact that autotools-changes often are subtile, which lets people who are not intimate with the autotools (erroroniously) believe it's safe to run autotools during builts.
Gathering evidence of breakage may cause a change of opinion. Having a practical alternative is probably required as well.
The practical alternative is very simple: Run the autotools on the system you are testing on, create diffs from them and to apply them during builds.
I am applying this approach to several of my Fedora packages (some of which I know to suffer from such issues, e.g. Coin2), fixed some packages (owned by others) this way, which had failed during the F11-mass-rebuild, exactly because of such issues.
Ralf