RFR 8199756 : Simplify language, country, script, and variant property initialization
david.holmes at oracle.com
Mon Mar 19 22:38:18 UTC 2018
On 19/03/2018 11:55 PM, Roger Riggs wrote:
> Hi Alan,
> The original changeset  is most easily viewed in the jdk-8u repo.
> The comment and placement of the removes of the properties and the code
> to re-set them
> around the call to the JVM_InitProperties (which sets properties from the
> -D arguments among others), seem to indicate that the previous property
> values are irrelevant.
Not irrelevant: defaults.
> The remaining question is whether there are any side effects or usage of
> four property values between the time they are initially set and when
> they are removed.
> The code in System.initProperties is straightforward and the bulk of the
> is simply setting property values, with one interspersed call to
> initialize the sun.jnu_encoding.
> From all appearances there are no side effects or uses of the initially
> set values of those four properties.
I would have expected these values to possibly have an impact on
InitalizeEncoding and subsequent use of PUTPROP_ForPlatformNString. Else
why set them early then unset them again? Or maybe they impact any error
messages that might arise in this stage of initialization?
I expect any issues here to be extremely subtle - as Alan indicated. If
there is any path to Java code that may try to read and use these
properties then they need to be set.
And if these are truly unnecessary then doesn't that imply that placing
them in sprops via GetJavaProperties in the first place is also unnecessary?
Has this been extensively tested on a non-English Windows system? And do
we even have tests for various initialization failures?
> Thanks, Roger
>  https://bugs.openjdk.java.net/browse/JDK-4700857
> RFE: separating user locale and user interface locale
> On 3/17/18 5:46 AM, Alan Bateman wrote:
>> On 16/03/2018 20:32, Roger Riggs wrote:
>>> Please review a small simplification of the initialization of System
>>> properties for
>>> language, country, script, and variant. Some steps for initializing
>>> them are unnecessary.
>>> The tests pass; a careful review would be appreciated so as to avoid
>> There are subtle interactions with how system properties are set in
>> the VM and on the command line which will take a bit effort to
>> research in order to review this and be satisfied that the change is
>> safe. Have you dug into the ancient history and bugs to see why it was
>> done this way?
More information about the core-libs-dev