RFR 8012326: Deadlock occurs when Charset.availableCharsets() is called by several threads at the same time
xueming.shen at oracle.com
Wed May 8 19:15:14 UTC 2013
On 05/08/2013 11:58 AM, Alan Bateman wrote:
> On 04/05/2013 20:05, Xueming Shen wrote:
>> Another sync need is for the "init()", in which it may update the aliasMap,
>> classMap and aliasNameMap via charset() and deleteCharset() if those
>> special properties are defined. There are two sources for the charset()/
>> deleteCharset(), one is from the constructor, one is from the init(), given
>> ExtendedCharsets is now singleton and get initialized at class init, these
>> should be no race concern between these two sources, so no need to
>> have any sync block inside charset() and deleteCharset(), the only thing
>> need to defend is inside init(), and all three public methods invoke the
>> init() at the beginning of the method body.
> One question on this. What would be the downside be doing all the initialization in ExtendedCharsets's constructor instead of two phases? Aside from availableCharsets then I would expect that all other usages would require the mappings to be setup.
ExtendedCharsets might be loaded/looked up for the default charset during
vm-boot, in which the system properties might not have been initialized yet.
So the "init()" need to be invoked again later, after boot, after we have
those system properties initialized (if specified).
More information about the core-libs-dev