On 9/9/14, 12:39 AM, Alan Bateman wrote:
It's not clear to me that this is the right place to address this issue. Clearly the changes to JRELocaleProviderAdapter to use ServiceLoader have upset the ordering that things are initialized for this error case, but it looks to me that it's always been very fragile. I think we need stack back a bit and examine the stack trace to see where might be a more appropriate place to address this, it may be that the security code needs to be more tolerant of attempts at recursive initialization.
It's an inherent issue where some init code issues locale sensitive services, such as in this case, *Formatter. So this could happen not only in Security class, but could be anywhere. So I think we still need the change proposed here in order for avoiding regressions. On error cases, the diagnosability will be addressed with the issue with 8057075.
Naoto
