> Date: Tue, 11 Jun 2013 19:53:21 +0300
> From: Eli Zaretskii <e...@gnu.org>
> Cc: l...@gnu.org, guile-user@gnu.org
>
> What eventually did the trick was configuring --with-threads=no. Once
> I did that, the build ran successfully and almost 100% cleanly to
> completion. (I will report the details about "almost" later.)
Here's part 1 of those details.
First, running the test suite resulted in some failures.
test-system-cmds failed because it uses '..' quoting on the command
line, which the Windows shell doesn't support. Fixed thusly:
--- test-suite/standalone/test-system-cmds~0 2010-12-08 11:07:11.000000000
+0200
+++ test-suite/standalone/test-system-cmds 2013-06-12 13:52:14.333269200
+0300
@@ -10,7 +10,7 @@
"test-system-cmds: (system) did not return a boolean\n")
(exit 1)))
- (let ((rs (status:exit-val (system "guile -c '(exit 42)'"))))
+ (let ((rs (status:exit-val (system "guile -c \"(exit 42)\""))))
(if (not (= 42 rs))
(begin
(simple-format
@@ -39,4 +39,4 @@
;; Local Variables:
;; mode: scheme
-;; End:
\ No newline at end of file
+;; End:
Next, test-unwind failed because it assumed that either $TMPDIR or
'/tmp' will exist, none of which can be counted upon on Windows.
Here's the fix for that (the declaration of mkstemp avoids compiler
warning):
--- test-suite/standalone/test-unwind.c~0 2012-01-31 00:32:38.000000000
+0200
+++ test-suite/standalone/test-unwind.c 2013-06-12 14:11:47.967231800 +0300
@@ -200,9 +200,19 @@ check_ports ()
#define FILENAME_TEMPLATE "/check-ports.XXXXXX"
char *filename;
const char *tmpdir = getenv ("TMPDIR");
+#ifdef __MINGW32__
+ extern int mkstemp (char *);
if (tmpdir == NULL)
+ tmpdir = getenv ("TEMP");
+ if (tmpdir == NULL)
+ tmpdir = getenv ("TMP");
+ if (tmpdir == NULL)
+ tmpdir = "/";
+#else
+ if (tmpdir == NULL)
tmpdir = "/tmp";
+#endif
filename = alloca (strlen (tmpdir) + sizeof (FILENAME_TEMPLATE) + 1);
strcpy (filename, tmpdir);
Finally, the tests in check-guile fail because they unconditionally
use features that are not compiled into the MinGW build or not
supported by it, like 'lstat', AF_UNIX in sockets, etc. Moreover,
whenever a test fails, the run stops without running the rest of the
test. Is there a way to skip tests that are known to fail? I see
that renaming the corresponding test files to not have the .test
extension would probably do the trick, but is there a better way? For
that matter, is there a way of forcing the test run to continue even
if some test failed? (I would actually suggest to skip tests
automatically based on unsupported features.)
Thanks.
