RFR: CODETOOLS-7902936 Stray VM warnings in forked PrintPropertiesMain lead to hard to debug XML parse exception

[Aleksey Shipilev]

On Fri, 21 May 2021 02:27:09 GMT, Jason Zaugg <jzaugg at openjdk.org> wrote:

> - Recover from stray output before the XML prelude
>   - Report the output in the exception otherwise
>   - Silence VM warnings in PrintPropertiesMain when JEP-158 (-Xlog:) is available (JRE 9+)

I see little sense in trying to recover here. The stray VM output can come in the middle of that XML dump too. I would prefer to cover the "hard to debug" part only: dump the proper diagnostics and then fail hard. We can probably print the helpful message how to shun the VM warnings, if user is so inclined?

jmh-core/src/main/java/org/openjdk/jmh/runner/Runner.java line 839:

> 837:         command.add("-XX:+IgnoreUnrecognizedVMOptions");
> 838:         command.add("-Xlog:all=error");
> 839: 

I think pushing `-Xlog:` to unsuspecting VMs is asking for trouble, even with `IgnoreUnrecognizedVMOptions`. Let's not do this.

jmh-core/src/main/java/org/openjdk/jmh/util/Utils.java line 595:

> 593:             try {
> 594:                 out.loadFromXML(new ByteArrayInputStream(baos.toByteArray()));
> 595:             } catch (InvalidPropertiesFormatException ex) {

Move this `catch` block to the existing `try-catch` block. Make the whole thing failing hard without trying to recover. Print the output contents in the exception.

jmh-core/src/main/java/org/openjdk/jmh/util/Utils.java line 597:

> 595:             } catch (InvalidPropertiesFormatException ex) {
> 596:                 // Maybe some VM output has preceded the XML content?
> 597:                 String output = new String(baos.toByteArray(), Charset.defaultCharset());

I suspect that since we are reading from the console, we need to use `Utils.guessConsoleCharset()` instead of trusting the default one.

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

Changes requested by shade (Committer).

PR: https://git.openjdk.java.net/jmh/pull/41