is ccache working effectively?

Erik Joelsson erik.joelsson at oracle.com
Wed Nov 21 16:19:08 UTC 2018


On 2018-11-21 04:16, Maurizio Cimadamore wrote:
>
> On 20/11/2018 11:49, Magnus Ihse Bursie wrote:
>> On 2018-11-09 11:09, Maurizio Cimadamore wrote:
>>> That does the trick, thanks.
>> If ccache and PCH is so bad in combination, maybe we should not allow 
>> it? Like turning off PCH by default of --enacle-ccache is given? Or 
>> at least *warn* that this combination is no good?
>
> That sounds like a good idea
>
I know it has worked in the past, even if the effectiveness does go 
down, it shouldn't be a net loss. There is something weird going on here 
that would need some investigation but I'm pretty sure it can be fixed.

/Erik

> Maurizio
>
>>
>> /Magnus
>>>
>>> I take a very small hit in the cold cache case, but then I get huge 
>>> boost once the cache is hot.
>>>
>>> cache hit (direct)                  1629
>>> cache hit (preprocessed)             115
>>> cache miss                             1
>>> cache hit rate                     99.94 %
>>> cleanups performed                     0
>>> files in cache                      5121
>>> cache size                         317.6 MB
>>> max cache size                       5.0 GB
>>>
>>> And, in my laptop the clean/build time went down from 11 mins to 3 :-)
>>>
>>> Maurizio
>>>
>>> On 09/11/2018 00:24, Erik Joelsson wrote:
>>>> You can try --disable-precompiled-headers and see if that helps. We 
>>>> have had special considerations for combining these features in the 
>>>> past, but it's certainly a source of trouble for ccache.
>>>>
>>>> /Erik
>>>>
>>>>
>>>> On 2018-11-08 16:20, Maurizio Cimadamore wrote:
>>>>>
>>>>> On 08/11/2018 20:31, Erik Joelsson wrote:
>>>>>> I think we need more details here to figure it out, but it 
>>>>>> doesn't surprise me that hotspot is the part that's failing. 
>>>>>> Building hotspot is way more complex than the rest. 
>>>>>
>>>>> Digging more into the command lines of the various HS files, they 
>>>>> include many generated sources and, also this folder:
>>>>>
>>>>> build/<CONF>/hotspot/variant-server/libjvm/objs/precompiled
>>>>>
>>>>> Now, I don't know exactly how ccache deals with PCH but I notices 
>>>>> that every time that the files in this folder are regenerated by 
>>>>> the build (I can easily do that by rerunning the same command line 
>>>>> that is stored inside the , a ccache miss follows.
>>>>>
>>>>> Maurizio
>>>>>
>>>>
>>


More information about the build-dev mailing list