RFR: 8168073: Speed up URI creation during module bootstrap

[Claes Redestad]

Hi Paul,

On 2016-10-17 19:39, Paul Sandoz wrote:
> Hi Claes,
>
> This looks good.

Thanks!

>
> Did you consider adding asserts to the package private constructor?

No, might be reasonable. Do you insist? :-)

/Claes

>
> Paul.
>
>> On 17 Oct 2016, at 04:17, Claes Redestad <claes.redestad at oracle.com> wrote:
>>
>> Hi,
>>
>> one partial cause for startup regressions due to jigsaw is related to creating
>> URIs for the location of each module.
>>
>> By providing a package-private constructor we can avoid the time to scan and
>> validate the URI, which takes a little time (executes ~80K bytecodes) but also
>> pushes various methods over compilation thresholds during early startup,
>> which more notably interferes with perceived startup.
>>
>> http://cr.openjdk.java.net/~redestad/8168073/webrev.01/
>> https://bugs.openjdk.java.net/browse/JDK-8168073
>>
>> Notes:
>>
>> When looking at this I discovered that the existing methods in JavaNetAccess
>> is no longer in use, and keeping the initialization in URLClassLoader forces
>> added classloading and costly initialization no longer needed during normal
>> bootstrap. Moving JavaNetAccess to URI seems like the natural choice.
>>
>> By exploiting the lazyness of URI.toString(), this patch also saves a few Kbs of
>> retained heap.
>>
>> Thanks!
>>
>> /Claes
>