On Wed, Jun 8, 2011 at 3:40 AM, "Paweł Hajdan, Jr." <firstname.lastname@example.org>
> On 6/7/11 9:53 PM, Matt Turner wrote:
>> Building 400~ locales is not fun on mips when building stages.
> Do you have some data to quantify "not fun"? How long does it take?
To build glibc
- once for the n32 ABI with only the en_US.UTF-8 locale: 1h 15m
- three times for o32/n32/n64 ABIs with only en_US.UTF-8: 4h 50m
- three times for o32/n32/n64 ABIs with all locales: 6h
I can reconfirm how long it takes just by running locale-gen on all locales,
but according to my numbers it takes 1h 10m to generate all the locales.
The o32/n32/n64 configuration is what I've been planning to ship as a
compromise between shipping only a single ABI and a set of stages for every
ABI, but each ABI adds significant time, and n32 replaces o32 so I'll just
plan to drop o32.
So roughly the time required to build all the locales is equal to the time
to build glibc for a single ABI.
>> No user has a need for more than some small subset of the total
>> available locales.
>> I filed bug  to request the ability to select locales in catalyst
>> spec files, but no responses after six months -- which is totally
>> typical of catalyst bugs.
>> I commented in bug  suggesting that we perhaps reduce the default
>> locale.gen to only 'en_US.UTF-8 UTF-8' or some other limited subset.
> It makes sense to me, especially for "exotic" arches. It seems that
> locale-gen is cheap anyway on x86 and amd64, where it probably matters