RFR(S): 8005885: enhance PrintCodeCache to print more data

[John Rose]

On Sep 3, 2013, at 4:10 AM, Albert Noll <albert.noll at oracle.com> wrote:

> thanks again for the feedback.
> Here is the new webrev: http://cr.openjdk.java.net/~anoll/8005885/webrev.02/<http://cr.openjdk.java.net/%7Eanoll/8005885/webrev.02/>

Hi Albert.

Please take a look at the usage of 'xtty' in nmethod::print_statistics and other *::print_statistics functions.

Also, look at event-oriented output via xtty in nmethod::log_new_nmethod and Deoptimization::uncommon_trap_inner.

Finally, take a quick look at the interplay between DisplayVMOutput and LogVMOutput.  Specifically, see the logic in ::print_statistics (java.cpp) which uses a FlagSetting of DisplayVMOutput to muzzle stdout while still printing useful information to hotspot.log.

Some observations:

1. Output dumps like this can be more easily worked with if there is some additional markup added to the log file.

2. Event-oriented dumps (which happen in the middle of execution) are especially in need of time stamps and other context information; this is easy to supply via xtty.

3. It is normal to make output to hotspot.log more verbose than output to stdout.  This can help you tune your defaults so that some info (or maybe none) comes to stdout and more info goes to hotspot.log.

I think you can use these patterns to help control the shape and volume of the output.

Thanks!

— John
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/attachments/20130904/397394f6/attachment.html