[Fwd: lilo-22.5..............]

[Auke Kok]

On Sat, 2003-03-08 at 23:01, Ray Maung wrote:
> Thanks for clearing that up.  I guess the devs can't
> test *every* upgrade, so we'll have to settle for
> end-user testing.  I'm pretty sure this was decided
> before I even knew about linux.

FYI, in order to assure you guys out there,

- I was the person who updated lilo to 22.5
- I tested (compiled, ran lilo, rebooted a few OS's) on 2 machines with
different configurations
- I decided that the additions made by the developer of lilo between
22.4.1 and 22.5 were solid and read back the changelog of lilo to
confirm I have seen the changes and that my testing would utilize the
changed code.

therefore:

> I guess the devs can't test *every* upgrade

this is true but was avoided thoroughly before lilo was upgraded.

> so we'll have to settle for end-user testing.

this is not true and certainly not our intent!

> I'm pretty sure this was decided before I even knew about linux

I'm not sure about what you mean 'was decided' here.

I hope I am clear enough about our testing methods, since this is a
reoccurring issue (not everyone will read this message so in a month I
can probably repost this, and maybe that's a good thing) /discussion I
am simply reputting it here briefly:


* all changes to "moonbase" are tested, often (and as often as possible)
by more than one person and often (and as often as possible) on more
than one system.

* "moonbase" only contains "stable" software. This does not mean that it
will build correctly under circumstances. Maintaining 1500 packages is
quite some work, and every change possibly affects all other modules.

example: the upgrade of X to 4.3.0 affects every graphical program and
completely messes up Xft in a lot of modules. We therefore only upgrade
modules if we think everything will work after the upgrade

* changes to "moonbase" are usually tested in 2 passes: 1) a build pass
with manual checks during the process (source location OK? configure
options changed? install ok? remove ok too?) 2) a blind/automatic pass
(recompile 2/3 times, over an existing installation, etc) 

* after testing modules we perform a "functionality" check: reboot, see
if lilo boots and all the options work

concerning lunar/theedge:

* all changes to the "lunar" code have been used by all developers for
days, weeks or sometimes even months. Code changes in "lunar" are tested
in "theedge" before they are put in "lunar"

* all changes to "theedge" are fresh of the press, and usually not
tested thoroughly. 


I hope this clears up your view of our job here ;^)

sofar

-- 
Auke Kok <auke.kok at planet.nl>