misc build snags

[Alan Bateman]

On 04/12/2012 14:19, Erik Joelsson wrote:
> Answering what I can from a build-infra perspective.
>
> On 2012-12-03 13:11, Doug Lea wrote:
>>
>> I set up jdk8 builds on linux64, solarisx86 (client and server)
>> and solaris sparc(v9) over the past few days. Here are a few notes.
>> Sorry if any are duplicates of known issues.
>>
>> 1. Defaulting JOBS to #cpus is probably not the best policy.
>> On a 64way linux box, doing this in the jdk build caused
>> enough total threads to be created that it hit resource limits
>> causing hard-to-diagnose problems like dumping OOME messages
>> into generated char encoding files. Maybe you could cap max
>> at some value like 16 unless overridden.
> I can imagine this going bad at times and I've filed bug 8004352 to 
> track this issue.
I've also have problems with the defaults, at both extremes too.

If I read the m4 correctly then it defaults to cores*2, which probably 
explains why building on a dual-core system after configuring with 
--with-num-cores=1 hurts interactive experience more than I expected. I 
also work regularly on Solaris machine with 64 hw threads and I need to 
remember to run configure with --with-num-cores=8 to avoid overloading it.

I've no doubt that there isn't a right answer for all systems but if 
--with-num-cores is specified then it might be better to not double it. 
Also capping it at 16 when not specified would likely help too.

-Alan