RFR(trivial): 8232768: Configuration with --disable-cds --enable-generate-classlist should be reported as an error

[Magnus Ihse Bursie]

On 2019-10-22 09:44, Jie Fu wrote:
> Hi all,
>
> May I get reviews for this one-line change?
>
> JBS:    https://bugs.openjdk.java.net/browse/JDK-8232768
> Webrev: http://cr.openjdk.java.net/~jiefu/8232768/webrev.00/
>
> This kind of configuration, with both --disable-cds and 
> --enable-generate-classlist, will fail.
> The failure is caused by -XX:DumpLoadedClassList [1], which doesn't 
> work as expected when cds is disabled.
>
> It might be better to produce an error.
I'm not sure I agree with this fix. Checking the code, we make an effort 
to determine if we should warn the user:

   # Check if it's likely that it's possible to generate the classlist. 
Depending
   # on exact jvm configuration it could be possible anyway.
   if test "x$ENABLE_CDS" = "xtrue" && 
(HOTSPOT_CHECK_JVM_VARIANT(server) || HOTSPOT_CHECK_JVM_VARIANT(client) 
|| HOTSPOT_CHECK_JVM_FEATURE(cds)); then

Only if the user *explicitly* sets --enable-generate-classlist, the 
warning is printed. The comment clearly states that this might be 
possible anyway, given the circumstances, so I believe a warning, and 
not an error, is the right thing here. The user explicitly requests 
generate-classlist. We warn that it might not work, but we allow it 
since it can work in cases we don't really know about.

If it is the case that generate-classlist *never* will work with CDS 
disabled (this sounds reasonable, but I don't know if this really is the 
case), and you want to catch this, then I suggest you restructure the 
code so you catch the CDS && generate-classlist case separately, and 
give a fatal error for that, and then keep the current check for jvm 
variants/features and let it keep producing a warning for that.

/Magnus
>
> Thanks a lot.
> Best regards,
> Jie
>
> [1] 
> http://hg.openjdk.java.net/jdk/jdk/file/ca620b06b5c9/make/GenerateLinkOptData.gmk#l68
>
>