RFR(M) 8209385: CDS runtime classpath checking is too strict when only classes from the system modules are archived
calvin.cheung at oracle.com
Fri Aug 17 21:31:52 UTC 2018
Thanks for your quick review, Ioi.
I'll push the fix soon and will let you know.
On 8/17/18, 2:21 PM, Ioi Lam wrote:
> Hi Calvin,
> The changes look good. I'll wait for your push, before pushing my
> filemap.hpp cleanup (JDK-8209657)
> - Ioi
> On 8/17/18 2:14 PM, Jiangli Zhou wrote:
>> Looks good! Thanks for fixing this issue.
>> On 8/17/18 2:06 PM, Calvin Cheung wrote:
>>> bug: https://bugs.openjdk.java.net/browse/JDK-8209385
>>> webrev: http://cr.openjdk.java.net/~ccheung/8209385/webrev.00/
>>> If a path such as -cp points to an non-empty directory is specified
>>> during CDS dump time but is not being referenced for class loading,
>>> we should not exit with the error "Cannot have non-empty directory
>>> in paths".
>>> The fix is to skip checking the path entries which are not being
>>> referenced during CDS dump time. Adding the _max_used_path_index
>>> field in ClassLoaderExt so that the unreferenced path entries won't
>>> be checked during dump time. The _max_used_path_index will also be
>>> stored in the CDS archive header so that during run time, the
>>> unreferenced path entries will not be checked.
>>> Testing: hs-tier 1,2,3.
More information about the hotspot-runtime-dev