Re: [gentoo-dev] what to do when an ebuild needs loads of RAM?

[Robin H. Johnson]

On Fri, Nov 19, 2004 at 09:18:36AM +0100, Marius Mauch wrote:
> On Thu, 18 Nov 2004 16:27:07 -0800
> "Robin H. Johnson" <robbat2@g.o> wrote:
>
> > As the requirements for a package can vary based on a number of things
> > or be reasonably fixed, I see a solution like this:
> > An ebuild defines two variables:
> > NEEDED_COMPILE_DISK
> > NEEDED_COMPILE_RAM
> > NEEDED_FINAL_DISK
> > NEEDED_FINAL_RAM
> >
> > which should evaluate to a constant numeric value in Kb of the
> > absolute maximum.
> *If* we do that please lets use Mb as a few Kb won't make a difference
> anyway. That said I don't think this is really an option in general as
> we'd have to consider too many variables, CFLAGS and USE being the major
> ones, to a lesser degree toolchain versions and ARCH. What combination
> should be used to determine the "absolute maximum"?
We don't need a super accurate number, just a ballpark figure.  This
could generally be improvised or just measured.  The diskspace
measurements would be the easiest - just sample the size of the active
tree at various times.

I'd generally expect this should be used only for ebuilds with abnormal
requirements: >1Gb of disk, or >256Mb of RAM (subject to change).

> Also I don't see any specifics how these variables should be used, would
> that be up to the ebuild, an eclass or portage? Too many open questions
> to seriously consider this yet.
I'd throw all the requirement checking stuff into an eclass, but I do
see one thing that would be suitable for portage to have (and it's more
broadly useful):
Ability to do an arbitary calculation at depend/pretend time so that it
can stop the entire emerge process before any packages are
unpacked/compiled/installed.

Eg portage can calc the deptree, and then before printing it out, it
evaluates the testing function for every ebuild in the deptree that has
the function. I already see ebuilds that have something similar in
pkg_setup(), but that runs too late - it should check things before any
installs start.

If portage cached which ebuilds had the testing function this makes the
evaluation of it O(n) with a very small n.

As one place that would definetly use this as well, see PHP packages for
combinations of USE flags that are competely incompatible and automatic
logic to exclude one of the USE flags is not appropriate (currently we
just call die).

-- 
Robin Hugh Johnson
E-Mail     : robbat2@××××××××××××××.net
Home Page  : http://www.orbis-terrarum.net/?l=people.robbat2
ICQ#       : 30269588 or 41961639
GnuPG FP   : 11AC BA4F 4778 E3F6 E4ED  F38E B27B 944E 3488 4E85