RFR: 8193513: add support for printing a stack trace on class loading [v2]

[Doug Simon]

On Fri, 30 Jun 2023 19:38:02 GMT, Ioi Lam <iklam at openjdk.org> wrote:

>> Are you saying `log_is_enabled(Info, class, load, cause)` implies `log_is_enabled(Info, class, load)`?
>> Or do I need:
>> 
>>   if (!log_is_enabled(Info, class, load) && !log_is_enabled(Info, class, load, cause)) {
>>     return;
>>   }
>> 
>> to extend the check on [line 3793](https://github.com/openjdk/jdk/blob/e8ff74c7e84ec2440a51fee1b4c45e87332807a0/src/hotspot/share/oops/instanceKlass.cpp#L3793)?
>> 
>> I don't understand the relationship between `class+load+cause` and `class+load` and the comment on `log_is_enabled` doesn't help much.
>
> Usually, the user would need to specify both. Otherwise just having the cause without the `class+load` would be rather useless
> 
> 
> java -Xlog:class+load  -Xlog:class+load+cause 
> 
> 
> Which can be combined like:
> 
> 
> java -Xlog:class+load,class+load+cause 
> 
> 
> or 
> 
> 
> java -Xlog:class+load*=info
> 
> 
> (To see usages of `-Xlog`, you can grep for `[-]Xlog:` in the *.java files under test/hotspot/jtreg)
> 
> 
> In terms of implementing this PR, I would suggest handling the class+load and class+load+cause indecently. Something like
> 
> 
> 
> void InstanceKlass::print_class_load_logging(ClassLoaderData* loader_data,
>                                              const ModuleEntry* module_entry,
>                                              const ClassFileStream* cfs) const {
>   if (ClassListWriter::is_enabled()) {
>     ClassListWriter::write(this, cfs);
>   }
> 
>   log_class_load(...);  // handle -Xlog:class+load 
>                         // (the rest of the old print_class_load_logging() function)
>   log_class_load_cause(...);  // handle -Xlog:class+load
> }
> 
> 
> The output would be something like
> 
> 
> [0.014s][info][class,load] java.lang.Thread$State source: shared objects file
> [0.014s][info][class,load,cause] 	at jdk.internal.misc.VM.toThreadState(java.base/VM.java:336)
> [0.014s][info][class,load,cause] 	at java.lang.Thread.threadState(java.base/Thread.java:2731)
> [0.014s][info][class,load,cause] 	at java.lang.Thread.isTerminated(java.base/Thread.java:2738)
> [0.014s][info][class,load,cause]	at java.lang.Thread.getThreadGroup(java.base/Thread.java:1957)
> ...
> 
> 
> This way, you don't need to repeat the class name, which is already printed by `-Xlog:class+load`

> Also, this is always JavaThread::current() because all the callers have the current JavaThread. No need for the null check or the cast.

Does this mean this existing type check and cast is unnecessary: https://github.com/openjdk/jdk/blob/e8ff74c7e84ec2440a51fee1b4c45e87332807a0/src/hotspot/share/oops/instanceKlass.cpp#L3821-L3823

-------------

PR Review Comment: https://git.openjdk.org/jdk/pull/14553#discussion_r1248265588