about "optimize" files

Pierre Labastie pierre.labastie at neuf.fr
Sat Nov 16 01:29:50 PST 2013


Hi,

Trying to answer everybody in one message, thanks for your inputs!

Le 13/11/2013 21:47, Bruce Dubbs a écrit :
> Pierre Labastie wrote:
>> Hi all,
>>
>> I'd like to change the default optimizations we ship with jhalfs.
>>
>> [...]
> 
> I agree.  The only problem is when we get messages saying that something 
> doesn't work and they neglect to say that they know better than the 
> glibc devs and change the optimization.
> 
I agree, but jhalfs should not be used by those people...
>> for gcc, noOpt is needed, and for grub, I do not know, but it does not
>> hurt...
> 
> It's kinda dumb to change optimization flags on GRUB.  The execution 
> time is negligible and any size changes are not significant.
> 
Agreed.
>> Now, the opt_config file. I'd like to ship it with
>> MAKFLAGS="-j5"
>> (quadricore CPUs are very common these days)
> 
> I'd rather make that a selectable option with -j1 the default.
Agreed. I raised a ticket (#1688) and accepted it. Anybody who wants to
contribute can send comments.
> 
>> and DEF_OPT_MODE=noOpt
>>
>> This is the most likely way people would want to use the file: job
>> parallelization, but otherwise no deviation from the book.

@Thomas:
> I would agree but,  I would keep it noOpt by default as it is a safer
> default, and keeps closer to by the book, if someone really wants to
> optimize glibc they know they will want to and can override at their
> discretion. I've learned from experience over optimization especially
> with the core toolchain can create some very hard to track down bugs.

Sure, but it is exactly what I propose: default to noOpt in opt_config. OTOH,
opt_override is for people who change that default, in which case I propose to
leave them free to not override optimization for all but gcc and grub.

>>   Now, the opt_config file. I'd like to ship it with MAKFLAGS="-j5"
>> (quadricore CPUs are very common these days)
>
> True, their common but your assuming that most are building on current
> hardware, which I don't believe to be the case. I personally have always
> put linux on older boxes, and most systems I have access to are only dual
> core, so -j5 would be a bad default in that case.

Agreed too. We are taking that to the menuconfig interface, with default -j1.

regards
Pierre



More information about the alfs-discuss mailing list