RFR: 8014074: Building hotspot with ccache in new build is very slow with empty cache.

[Erik Joelsson]

On 2013-05-27 05:11, David Holmes wrote:
> Hi Erik,
>
> I have mixed feelings on this one. On the one hand I don't like the 
> idea that using/not-using an optimization like ccache, changes the way 
> the build of hotspot is carried out. On the other hand, if this means 
> we will do more builds without precompiled headers then it should help 
> trap errors when there are missing #includes.
I agree, and it feels weird to me too. Precompiled header is also a kind 
of build optimization, though more intrusive than ccache. I was even 
more annoyed at the long build times when the cache didn't give any hits 
though.
>
> Does this have any impact on our actual RE builds?
I do not know if they have ccache installed and available on their 
machines. My guess is that if it is, it's by accident. We could instruct 
them to add --disable-ccache to make sure they don't accidentally pick 
it up. I could also readd and fix the support for precompiled header 
together with ccache and add another option for it.

/Erik
>
> Thanks,
> David
>
> On 24/05/2013 11:22 PM, Erik Joelsson wrote:
>> This patch to configure disables precompiled headers in hotspot when
>> ccache is in use.
>>
>> Background is in this mail (and the bug report):
>> http://mail.openjdk.java.net/pipermail/build-dev/2013-April/008772.html
>>
>> In short, build time is shortened considerably both with an empty and a
>> perfect cache when ccache is in use.
>>
>> Note that this only affects full builds from the root repo and not
>> hotspot only builds.
>>
>> http://cr.openjdk.java.net/~erikj/8014074/webrev.root.01/
>>
>> /Erik