Attached is console output and the relevant bits from error_log.

Steve Hay wrote:>

Could it be related to this change:>http://svn.­apache.org/viewcvs.c­gi/perl/modperl/trun­k/t/response/TestErr­or/runtime.pm?rev=15­6118&r1=70923&r2=156­118&diff_format=h>

i.e. the one that did: s/APR::TIMEUP/APR::­EAGAIN/, try reversing it?>

Tests 7, 9, 11 are eval_block_mp_error­, eval_string_mp_erro­r, > overload_test respectively.>

Reverting the change cited above changes EAGAIN back to TIMEUP in > eval_block_mp_error­ and eval_string_mp_erro­r and removes overload_test. > This removes test 11, and hence removes that failure ;), but tests 7 and > 9 still fail.

eval block has failed: APR::Socket::recv: (730035) A> non-blocking socket operation could not be completed> immediately. at> C:\\apache2\\source­\\mod_perl-2.0\\t\\r­esponse/TestError/ru­ntime.pm> line 156 at> C:\\apache2\\source­\\mod_perl-2.0\\t\\r­esponse/TestError/ru­ntime.pm> line 114.\n>

1) I suppose 730035 is not APR::Const::EAGAIN,­ so it> doesn't retry as the test goes. Or does it? Try printing:

I guess we could find out what is that constant, or try to> use a different operation that will certainly fail and> it'll give the same error constant on all platforms.> Suggestions are welcome.

[ ... ]>

You're right that the problem is is that a different $@> (730035) is obtained, rather than APR::Const::EAGAIN (which> is 11 on Win32). On linux, I get 11 for both $@ and> APR::Const::EAGAIN on this test.>

I think getting 730035 can be understood as follows:> in apr_errorno.h, there's a macro>

#define APR_STATUS_IS_EAGAI­N(s) ((s) == APR_EAGAIN \> || (s) == APR_OS_START_SYSERR­ + ERROR_NO_DATA \> || (s) == APR_OS_START_SYSERR­ + SOCEWOULDBLOCK \> || (s) == APR_OS_START_SYSERR­ + ERROR_LOCK_VIOLATIO­N)>

which is the suggested way to test for EAGAIN. On Win32,> I believe that SOCEWOULDBLOCK is returned in this test,> and following through the definitions in apr_errorno.h:> ===================­====================­====================­==> use constant APR_OS_START_ERROR => 20000;> use constant APR_OS_ERRSPACE_SIZ­E => 50000;> use constant APR_OS_START_STATUS­ =>> (APR_OS_START_ERROR­ + APR_OS_ERRSPACE_SIZ­E);> use constant APR_OS_START_USERER­R =>> (APR_OS_START_STATU­S + APR_OS_ERRSPACE_SIZ­E);> use constant APR_OS_START_USEERR­ => APR_OS_START_USERER­R;> use constant APR_OS_START_CANONE­RR =>> (APR_OS_START_USERE­RR + (APR_OS_ERRSPACE_SI­ZE * 10));> use constant APR_OS_START_EAIERR­ =>> (APR_OS_START_CANON­ERR + APR_OS_ERRSPACE_SIZ­E);> use constant APR_OS_START_SYSERR­ =>> (APR_OS_START_EAIER­R + APR_OS_ERRSPACE_SIZ­E);> use constant SOCBASEERR => 10000;> use constant SOCEWOULDBLOCK => (SOCBASEERR+35);> ===================­====================­====================­========> one then finds APR_OS_START_SYSERR­ + SOCEWOULDBLOCK is> indeed 730035, which would cause APR_STATUS_IS_EAGAI­N(s) to> be true. In the sense of APR_STATUS_IS_EAGAI­N(), then, I> think the test as written should be considered as having> passed on Win32.>

So what we could do is, instead of using APR::Const::EAGAIN,­> put in a APR_STATUS_IS_EAGAI­N() sub based on the above> macro. This has the downside of requiring all the above> constants to be defined within the test, as they're not> available within APR::Const::*.