On Sun, Jan 15, 2006 at 10:14:38PM -0000, Andy Hassall wrote:
> > Here's another stake-in-the-ground so we know where we're at:
> >
> > http://www.data-plan.com/public/DBD-Oracle-1.17-RC4.tar.gz
>
> Test results for RC4 here:
>
> http://www.andyh.co.uk/temp/DBD-Oracle/1.17RC4/
>
> Just Linux and native Windows on there - all the combinations from last
> time, plus new runs against an XE database (which appears to act like a
> plain 10.2 database) - previously only tested the XE client, not the server.
> 100% pass on all of those.
Yeah!
> For the Cygwin failures, something to do with environment variables is
> prime suspect at the moment, specifically how it sets NLS_NCHAR at runtime
> for some of the tests.
>
> It seems to depend on the value of NLS_NCHAR before running the tests under
> Cygwin. If it's set to AL32UTF8, then 22nchar_al32utf8 passes. If it's blank
> or set to another value, the test fails. But on other platforms (even native
> Windows on the same machine), the value is overridden at runtime by the
> tests, so it doesn't matter what it's initially set to.
>
> The debug output that shows the current value of NLS_NCHAR shows it as
> AL32UTF8, but still it behaves as if it were the value in the parent process
> environment.
>
> There's a Perl bug "#36665: delete $ENV{FOO} leaves $FOO visible in
> subprocesses on Cygwin Perl"
> https://rt.perl.org/rt3/Ticket/Display.html?id=36665 which indicates there's
> some issues with env vars under Cygwin Perl, but it doesn't quite explain
> this behaviour - and it persisted even with trying the latest dev version of
> Perl which has the fix for that particular bug.
>
> Haven't got it quite worked out yet though, but seem to be getting closer.
It sounds like the behavior is what you'd expect if %ENV was cached
instead of live (ie manipulating %ENV didn't alter the processes own
'native' environment variables until a subprocess was spawned).
I've CC'd p5p in the hope someone can shed some light on it and save us
some time digging...
Tim.