Peter,
On 2005 Feb 22 , at 17.55, Peter W. Draper wrote:
>> I can't think why it would just barf for Alpha, but I can think why it
>> would change. I don't recall how Steve generates the dependencies for
>> the component-by-component build, but the dependencies for make world
>> are expressed to make, and it works out a build order for itself.
>> Changes in the set of dependencies could well result in a change of
>> build order which is equivalent for the expressed dependencies, but
>> happen not to be equivalent for the _actual_ dependencies. Does that
>> make any sense?
>
> [...]
> However, for any fixed set of dependencies expressed in a series of
> component.xml files, using the same logic (==autobuild?) you would
> expect
> the same outcome, wouldn't you. We're clearly not seeing this.
I wasn't sure if we were or not. Since we don't know exactly when the
nightly-builds run or the repository is rsync'd -- they're not at a
particular time, nor is the start time written in the build log -- it
could be that we're seeing platform-specific behaviour only because
different platforms happen to have slightly different dependencies in
the generated Makefile.in.
Alternatively, and possibly more plausibly, since the transformation
from dependencies to build order is done by make, the make
implementations on Tru64 (bsdmake?) and Debian might simply be
producing a different build order from the make (GNUmake) on RH and OS
X.
> The simplest way to check all dependencies for all time would be to use
> the Makefile system to cold-build each component from its manifest
> dependencies (which I admit would probably take a week on some of the
> machines!). When things go wrong that's how I check it (if a quick peer
> doesn't do it).
That's true, but it probably would take a week! We must be pretty
close to a correct set of dependencies, now, and the rate of change of
dependencies edits is slowing down.
Norman
--
----------------------------------------------------------------------
Norman Gray : Physics & Astronomy, Glasgow University, UK
http://www.astro.gla.ac.uk/users/norman/ : www.starlink.ac.uk
|