[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: efficient use of auto-builder machines (was Re: Bug#43787: changed title, and remade the proposed change)



On Tue, Sep 07, 1999 at 06:12:07PM -0400, Ben Collins wrote:
> Umm, what about libraries that purposely compile -dbg packages? This
> is a silly idea, it's not a good idea for the autobuilders to muck
> around with the way the package is meant to be built.

Good point.

Of course, this is solvable -- for example, build those packages in a
different autobuilder environment (perhaps make the cc cover sensitive
to an environmental variable).

This would make the autobuilders a bit more complicated -- but we are
talking about optimizing the autobuilders, so that's probably ok.  Also,
this sort of implementation would yield a far greater speed improvement
over the short run than your current policy proposal would.

In the long run, we need a decent source dependency system, which
integrates well with the build environment.  [I'm imagining something
along the lines of bsd's .include <bsd.whatever.mk> designed for
debian/rules, and a source dependency system that lets you specify
compiler versions or library versions independently from the target
architecture.]

If such a thing were designed/implemented/built/whatever, and if the
package maintainers for the major build tools thought it was well put
together, I'd support a policy change so that all new or editted packages
would (at some point) be required to use/support this new system.

But what you're doing right now: optimizing a hack...  That's more
likely to get in the way of the long term solution than anything else.
[If we had a decent source dependency, etc. design I'm sure what you're
proposing would fit right in -- with a few minor changes.  But if we
go ahead with the optimization when we're still at hack implementation
stage we're asking for backwards compatability problems.]

Some of your advice about the developer's options is good -- but that's
more material for the packaging manual than for policy.

And, of course, your original concept (that -g should be optional)
was good -- but you'd already won that one before you even started.

-- 
Raul


Reply to: