[1] RFR(XXS): 8197927: Can't set mandatory 'java.vendor.version' property to empty string

[Volker Simonis]

On Wed, Feb 14, 2018 at 4:26 PM,  <mark.reinhold at oracle.com> wrote:
> 2018/2/14 2:20:22 -0800, volker.simonis at gmail.com:
>> can I please get a review for the following tiny fix:
>>
>> http://cr.openjdk.java.net/~simonis/webrevs/2018/8197927/
>> https://bugs.openjdk.java.net/browse/JDK-8197927
>>
>> The new Java 10 specification makes the 'java.vendor.version' property
>> mandatory [1] but the current implementations doesn't allow to set it
>> to an empty string.
>
> This is a bug in the specification, not the implementation.  As I just
> wrote in a comment on 8197927:
>
> JEP 322 expresses the intended behavior: If `--with-vendor-version` is
> not specified at build time then the `java.vendor.version` property has
> no value.  (That's different from the empty string, which is a value.)
> The bug is in the specification, which should be revised so as not to
> require that this property have a value.  Fixing the spec is not
> critical for 10; I suggest downgrading 8197927 to P2 and targeting 11.
>

What you mean by "not require the 'java.vendor.version' property to
have a value"? We can not set a system property to NULL because
System.setProperty() will throw a NPE in that case. So either the
specification requires 'java.vendor.version' as mandatory (as it is
now) in which case it does need to have a value different from NULL.
Or we change the specification such that 'java.vendor.version' isn't a
mandatory property any more. But it is not possible to have
'java.vendor.version' as mandatory property with "no value" as
envisioned by the JEP.

Do you propose to change the specification such that
'java.vendor.version' isn't a mandatory property any more? That would
be fine for me and I agree that we could target this bug to 11 if
that's what you propose. We could actually list "java.vendor.version"
under "Implementation Note:" which contains some additional,
non-mandatory properties.

> - Mark