Fix pipelined response deadlock in tests

There's a race condition that can cause mod_perl's test suite to hang

in t/filter/in_str_declined.t.  The problem is that the response handler

starts generating response body, and so triggers header output, before

it reads the request body.  If LWP::Protocol::http, which is the client

for this test, receives a complete set of response headers, it will stop

sending the request body.  (However, if the request body is no more than

8192 octets then it will send the whole body before it starts looking

for a response.  The failure only shows up with an appreciably large

request body.)

RFC 2616 doesn't explicitly address this sort of pipelining, but the

start of section 6 does say "After receiving and interpreting a request

message, a server responds with an HTTP response message.", which can be

read as prohibiting sending any part of the response before the entire

request has been received.

The attached patch fixes this issue by making all the POST handlers in

the test suite read the body before doing anything that generates output

(specifically plan()).

-zefram

CPAN RT#82409

Debian bug #676754

--- a/t/filter/TestFilter/in_str_declined.pm	2011-02-08 02:00:11.000000000 +0000

+++ b/t/filter/TestFilter/in_str_declined.pm	2013-01-04 16:08:14.000000000 +0000

@@ -35,13 +35,17 @@

 sub response {

     my $r = shift;

+    my $data;

+    if ($r->method_number == Apache2::Const::M_POST) {

+        # consume the data so the input filter is invoked

+        $data = TestCommon::Utils::read_post($r);

+    }

+

     plan $r, tests => 2;

     $r->content_type('text/plain');

     if ($r->method_number == Apache2::Const::M_POST) {

-        # consume the data so the input filter is invoked

-        my $data = TestCommon::Utils::read_post($r);

         ok t_cmp(length $data, 20000, "the request body received ok");

     }

--- a/t/filter/TestFilter/in_str_declined_read.pm	2011-02-08 02:00:11.000000000 +0000

+++ b/t/filter/TestFilter/in_str_declined_read.pm	2013-01-04 16:06:28.000000000 +0000

@@ -31,14 +31,19 @@

 sub response {

     my $r = shift;

+    my $err;

+    if ($r->method_number == Apache2::Const::M_POST) {

+        # this should fail, because of the failing filter

+        eval { TestCommon::Utils::read_post($r) };

+        $err = $@;

+    }

+

     plan $r, tests => 1;

     $r->content_type('text/plain');

     if ($r->method_number == Apache2::Const::M_POST) {

-        # this should fail, because of the failing filter

-        eval { TestCommon::Utils::read_post($r) };

-        ok $@;

+        ok $err;

     }

     Apache2::Const::OK;

--- a/t/filter/TestFilter/in_str_msg.pm	2011-02-08 02:00:11.000000000 +0000

+++ b/t/filter/TestFilter/in_str_msg.pm	2013-01-04 16:08:27.000000000 +0000

@@ -76,10 +76,10 @@

 sub response {

     my $r = shift;

-    plan $r, tests => 1;

-

     my $received = TestCommon::Utils::read_post($r);

+    plan $r, tests => 1;

+

     ok t_cmp($received, $expected,

              "request filter must have upcased the data");

--- a/t/response/TestModperl/post_utf8.pm	2011-02-08 02:00:12.000000000 +0000

+++ b/t/response/TestModperl/post_utf8.pm	2013-01-04 16:04:39.000000000 +0000

@@ -29,14 +29,14 @@

 #    $r->content_type("text/plain; charset=utf-8");

 #    $r->print("expected: $expected_utf8\n");

+    my $received = TestCommon::Utils::read_post($r) || "";

+

     # utf encode/decode was added only in 5.8.0

     # XXX: currently binmode is only available with perlio (used on the

     # server side on the tied/perlio STDOUT)

     plan $r, tests => 2,

         need need_min_perl_version(5.008), need_perl('perlio');

-    my $received = TestCommon::Utils::read_post($r) || "";

-

     # workaround for perl-5.8.0, which doesn't decode correctly a

     # tainted variable

     require ModPerl::Util;