[Python-Dev] alpha problems -- need input

[Neal Norwitz]

These issues are on HEAD.  There might be some others I missed.

With cc there are at least 2 issues:
 * test_file causes interpreter exit due to sys.stdin.seek(-1)
 * test_pty fails apparently due to whitespace differences

http://www.python.org/dev/buildbot/all/alpha%20Tru64%205.1%20trunk/builds/18/step-test/0

Should we skip the test for an invalid seek on stdin on osf1?
I haven't investigated the test_pty failure further.

With gcc, there are also several issues:
 * test_float and test_struct fail due to NaN handling
 * test_long fails
 * test_ctypes

The question is how to fix these.  test_float and test_struct fail due
to a Floating Point Exception signal (SIGFPE).  test_long fails due to
float(shuge) raising a ValueError.  There is a comment:

            # XXX Perhaps float(shuge) can raise OverflowError on some box?
            # The comparison should not.

Note: it raises value error, not OverflowError.

Should I just wrap the float(shuge) in a try/except and skip the test
if an exception is raised?

The patches below fix the float and struct problems.  I'm not sure if
we should ignore SIGFPE in test_float.  I'm also not sure if the
change to floatobject.c should be made.  It's possible there is a
compiler flag that would help.  I tried -mieee and a few others, but
none seemed to make a difference for gcc.

test_ctypes fails because _findLib() doesn't seem to work properly on
OSF1.  Note:  libc.so  =>   /usr/shlib/libc.so

ERROR: test_find (ctypes.test.test_loading.LoaderTest)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/net/ringneck/scratch1/nnorwitz/python/trunk/Lib/ctypes/test/test_loading.py",
line 40, in test_find
    cdll.find(name)
  File "/net/ringneck/scratch1/nnorwitz/python/trunk/Lib/ctypes/_loader.py",
line 205, in find
    raise OSError("Library %r not found" % name)
OSError: Library 'c' not found

======================================================================
ERROR: test_load (ctypes.test.test_loading.LoaderTest)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/net/ringneck/scratch1/nnorwitz/python/trunk/Lib/ctypes/test/test_loading.py",
line 25, in test_load
    cdll.load(name)
  File "/net/ringneck/scratch1/nnorwitz/python/trunk/Lib/ctypes/_loader.py",
line 112, in load
    return self._load(libname, mode)
  File "/net/ringneck/scratch1/nnorwitz/python/trunk/Lib/ctypes/_loader.py",
line 124, in load_library
    return self._dlltype(libname, mode)
  File "/net/ringneck/scratch1/nnorwitz/python/trunk/Lib/ctypes/__init__.py",
line 292, in __init__
    self._handle = _dlopen(self._name, mode)
OSError: dlopen: Cannot map library libc.so.6

======================================================================

Below are the patches.

Index: Objects/floatobject.c
===================================================================
--- Objects/floatobject.c       (revision 43416)
+++ Objects/floatobject.c       (working copy)
@@ -1430,10 +1430,19 @@
                return -1;
        }
        else {
-               float y = (float)x;
+               float y;
                const char *s = (char*)&y;
                int i, incr = 1;

+               /* The conversion of x to a float causes a floating exception
+                  on some platforms (at least OSF1 v5.1 gcc-4.0.1)
+               */
+#ifndef MAXFLOAT
+#define MAXFLOAT ((float)3.40282346638528860e+38)
+#endif
+               if (x > MAXFLOAT || x < -MAXFLOAT)
+                       goto Overflow;
+               y = (float)x;
                if ((float_format == ieee_little_endian_format && !le)
                    || (float_format == ieee_big_endian_format && le)) {
                        p += 3;
Index: Lib/test/test_float.py
===================================================================
--- Lib/test/test_float.py      (revision 43416)
+++ Lib/test/test_float.py      (working copy)
@@ -1,4 +1,5 @@

+import signal
 import unittest, struct
 from test import test_support

@@ -83,6 +84,12 @@
 # is accident (today).

 class IEEEFormatTestCase(unittest.TestCase):
+    def setUp(self):
+        signal.signal(signal.SIGFPE, signal.SIG_IGN)
+
+    def tearDown(self):
+        signal.signal(signal.SIGFPE, signal.SIG_DFL)
+
     if float.__getformat__("double").startswith("IEEE"):
         def test_double_specials_do_unpack(self):
             for fmt, data in [('>d', BE_DOUBLE_INF),
Index: Lib/test/test_file.py
===================================================================
--- Lib/test/test_file.py       (revision 43416)
+++ Lib/test/test_file.py       (working copy)
@@ -100,12 +100,14 @@
     print "writelines accepted sequence of non-string objects"
 f.close()

-try:
-    sys.stdin.seek(-1)
-except IOError:
-    pass
-else:
-    print "should not be able to seek on sys.stdin"
+# This causes the interpreter to exit on OSF1 v5.1.
+if sys.platform != 'osf1V5':
+    try:
+        sys.stdin.seek(-1)
+    except IOError:
+        pass
+    else:
+        print "should not be able to seek on sys.stdin"

 try:
     sys.stdin.truncate()
Index: Lib/test/test_long.py
===================================================================
--- Lib/test/test_long.py       (revision 43416)
+++ Lib/test/test_long.py       (working copy)
@@ -372,10 +372,15 @@

             self.assertRaises(OverflowError, eval, test, namespace)

-            # XXX Perhaps float(shuge) can raise OverflowError on some box?
-            # The comparison should not.
-            self.assertNotEqual(float(shuge), int(shuge),
-                "float(shuge) should not equal int(shuge)")
+            # float(shuge) can raise OverflowError on an Alpha built with gcc.
+            try:
+                fshuge = float(shuge)
+            except ValueError:
+                pass
+            else:
+                # The comparison should not.
+                self.assertNotEqual(fshuge, int(shuge),
+                    "float(shuge) should not equal int(shuge)")

     def test_logs(self):
         import math
Index: Lib/test/test_struct.py
===================================================================
--- Lib/test/test_struct.py     (revision 43416)
+++ Lib/test/test_struct.py     (working copy)
@@ -429,11 +429,12 @@
     # The same, but tack on a 1 bit so it rounds up to infinity.
     big = (1 << 25) - 1
     big = math.ldexp(big, 127 - 24)
-    try:
-        packed = struct.pack(">f", big)
-    except OverflowError:
-        pass
-    else:
-        TestFailed("expected OverflowError")
+    for value in (big, -big):
+        try:
+            packed = struct.pack(">f", value)
+        except OverflowError:
+            pass
+        else:
+            TestFailed("expected OverflowError")

 test_705836()