a couple of patches to deal with w32 semantics

[Ralf Wildenhues]

Hello,

here are two more patches for w32.  When applied, the testsuite passes
for me on MinGW and Cygwin.  I'd appreciate if someone firmer than me
with the specialities on these system could confirm that my rationales
are not nonsense.

OK to apply?

Thanks,
Ralf

    Fix bogus testsuite failure on Cygwin due to insufficient permissions.
    
    * tests/torture.at (AC_CONFIG_FILES, HEADERS, LINKS and COMMANDS):
    config.status may succeed to create an output file on Cygwin
    even in a directory without write permission, because it creates
    the file in a temporary directory and moves it to `.' later,
    which apparently succeeds on Cygwin.  Test for this setup and
    avoid false failure in this case.

diff --git a/tests/torture.at b/tests/torture.at
index ec86e22..191ab3f 100644
--- a/tests/torture.at
+++ b/tests/torture.at
@@ -211,8 +211,11 @@ AT_CHECK([grep OK $1], [], [OK
 AT_CHECK_CONFIGURE([what_to_test=var-$1 --no-create])
 # config.status might be stupidly expecting data on stdin, if it's
 # really broken...
+# If making a directory and moving files from there to . succeeds, we
+# are probably on cygwin.
 chmod a-w .
-AT_CHECK([./config.status var-$1 </dev/null], [1], [ignore], [ignore])
+AT_CHECK([./config.status var-$1 </dev/null], [1], [ignore], [ignore],
+        [AT_CHECK([{ test -d t || mkdir t; } && : >t/foo && mv -f t/foo 
./foo])])
 chmod u+w .
 ])# AT_CHECK_CONFIG_CREATION_NOWRITE
 

    On MinGW, substitution of CR and 0xFF fails.
    
    * tests/torture.at (Substitute and define special characters):
    MinGW awk cannot handle 0xFF, and on MinGW, the test does the
    wrong thing for CR.

diff --git a/tests/torture.at b/tests/torture.at
index 191ab3f..17383b2 100644
--- a/tests/torture.at
+++ b/tests/torture.at
@@ -769,7 +769,9 @@ baz=bla
     for j in 0 1 2 3 4 5 6 7; do
       for k in 0 1 2 3 4 5 6 7; do
        case $i$j$k in #(
-       000) ;; #(
+       000 | 015 | 377) ;; # MinGW awk dislikes 0xFF, and the test does
+                           # the wrong thing for CR on MinGW.
+                           #(
        *) printf \\$i$j$k's' ;; # The 's' works around a Solaris 8 /bin/bash 
bug.
        esac
       done