RFR: 8275677: Test fails due to truncated hs_err file

[Thomas Stuefe]

On Fri, 24 Nov 2023 04:52:20 GMT, David Holmes <dholmes at openjdk.org> wrote:

> This failure has not reproduced in mainline for a long while but as a preventative measure, at the suggestion of @JesperIRL , I've made a slight change to the register printing so that we always print the raw value before we start trying to figure out what it points to. That may shed some light on any future truncation.
> 
> Testing tiers 1-3 and manual inspection of hs_err files produced by tests
> 
> Thanks

Looks good, minus small nit.

I would rename this issue, or do the patch under a different issue, since the patch does not fix the truncation problem but makes print_location more resilient.

src/hotspot/share/runtime/os.cpp line 1183:

> 1181:   // with debugging things if we fail to decode the location.
> 1182:   st->print(INTPTR_FORMAT ": ", x);
> 1183:   st->flush();

Nit: flush is not necessary, and strictly speaking should not be done by this function, since it depends on under which conditions the printing is done. In error handling, the caller must make sure that printing is unbuffered. And it is: we use unbuffered io (fdStream) which needs no flushing and where flush() is a noop.

So, I'd just remove the explicit flush() here.

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

PR Review: https://git.openjdk.org/jdk/pull/16804#pullrequestreview-1747369587
PR Review Comment: https://git.openjdk.org/jdk/pull/16804#discussion_r1403980260