Request for review (M / Verifier Error messages)
karen.kinnear at oracle.com
Wed Aug 1 13:29:34 PDT 2012
First - thank you for doing this and for attaching the detailed output information,
which I think will be very useful.
Looks good and I like the way you structured this.
1. stackMapFrame.hpp lines 230-231: is that comment still accurate?
and stackMapFrame.cpp lines 225-226 - are these still true?
Would there be any value in printing out what the flags are? Or is that
would it be hard to rename print to print_on since you now pass it a stream?
Also end of print, you call tty->print_cr, don't you want to use str not tty?
On Aug 1, 2012, at 1:36 PM, Keith McGuigan wrote:
> On 7/26/2012 6:02 PM, Keith McGuigan wrote:
>> Would appreciate any review of the following code:
>> This code adds additional information to VerifyError messages to make it
>> easier to diagnose bytecode problems in the field. It may be a full (or
>> partial) solution for JEP 136 (http://openjdk.java.net/jeps/136).
>> The basic idea is to create an ErrorContext object when a verification
>> error occurs and populate it with context information so that things
>> like the current frame, stackmaps, and bytecodes can be included in the
>> VerifyError's message. In addition a diagnostic flag,
>> -XX:+VerifyVerbose is added which prints out the details of the
>> verification as it occurs.
>> The included test case has a 'testcases.jar' file that contains
>> hand-modified classfiles that trigger VerifyErrors in each possible
>> location in the verifier. Some cases appeared unreachable so those
>> cases are missing. I'm attaching the output of the test so you may see
>> the resulting error message format.
>> - Keith
More information about the hotspot-runtime-dev