Re: Package optimizations

[Greg Troxel <gdt%lexort.com@localhost>]

Jason Bacon <outpaddling%yahoo.com@localhost> writes:

> On 12/10/18 2:14 PM, Joerg Sonnenberger wrote:
>> On Mon, Dec 10, 2018 at 04:31:24PM +0000, coypu%sdf.org@localhost wrote:
>>> following up:
>>>
>>> there's some discussion about whether we should allow packages to have
>>> their own optimization flags.
>> We generally do not *want* that. The majority of packages has absolutely
>> no valid reason to insist on any specific optimizer flags. Period. There
>> are special cases and those are already just fine. Read: appending
>> certain flags like -ffast-math just works fine. -OMG is not generally
>> better than -O, contrary to the Gentoo Racing Community.
>>
>> Joerg
> How portable do we want binary packages to be?  If a packager inserts
> -mavx2, it may not run on hardware that's just a few years old.
>
> I think if we're going to allow this, there has to be a whitelist of
> allowable flags that are certified for portability to any hardware we
> consider worth supporting.

The general rule is that a package built with defaults should run on the
oldest hardware supported by that version of NetBSD.

So true optimizations are ok, but "assume cpu has feature X" is not.
(Separately from whether pkgsrc's CFLAGS should be respected.)