[prev in list] [next in list] [prev in thread] [next in thread] 

List:       pykde
Subject:    [PyQt] sipdistutils.py - broken exceptions flags in VC++
From:       Matteo Italia <matteo () mitalia ! net>
Date:       2014-10-27 22:27:47
Message-ID: 544EC6E3.4000200 () mitalia ! net
[Download RAW message or body]

Hello,

when building a Python 2.7 extension on Windows (with VC++ 2008) through
sipdistutils.py, I spotted several warning about wrong exception flags.

The issue seems to be that distutils only bothers to provide a compiler
with build options adequate for C extensions
(http://bugs.python.org/issue1222585), while we need C++ settings. For
this reason, no /EH flag is provided, and we end up building extensions
with broken exception semantic (in particular, RAII seems to be
completely broken when no /EH is provided, and catch(...) catches also
SEH exceptions, which is _extremely_ bad).

To fix this issue, I patched sipdistutils.py to add the adequate
exception specification option (/EHsc) if not present (and complain if
some /EH which is not /EHsc is already present); attached you can find
the diff against the sipdistutils.py in the current repository (since it
already goes on checking command line arguments, there's also a check to
remove -Wstrict-prototypes on Unix, which just generates noise in the
build).

Could someone check if it looks ok and if it can be eventually merged
back in SIP? I already had several bizarre crashes that could be traced
back to this problem, so I think that this may be useful to others.

Regards,
Matteo

["sipdistutils-winexceptions.patch" (text/x-patch)]

diff -r 2a310fa9719a sipdistutils.py
--- a/sipdistutils.py	Mon May 26 11:36:04 2014 +0100
+++ b/sipdistutils.py	Mon Oct 27 23:26:05 2014 +0100
@@ -108,6 +108,35 @@
                     if old_sig != new_sig:
                         self.force = True
 
+        # The compiler by default starts with the arguments used when compiling \
Python itself +        # This means that we may get a configuration thought for C, \
that we have to patch +        # (actually, the fact that we get a C++ compiler and \
not a C one is accidental; see +        # http://bugs.python.org/issue1222585 )
+        try:
+            c=self.compiler
+            if c.compiler_type=='unix':
+                # On gcc, we kill -Wstrict-prototypes, that does not apply to C++
+                # and generates spurious warnings at each processed file
+                try:
+                    c.compiler_so.remove('-Wstrict-prototypes')
+                except:
+                    pass
+            elif c.compiler_type=='msvc':
+                if not c.initialized:
+                    c.initialize()
+                # On VC++, if we do not specify /EHsc the C++ exception semantics is \
broken +                # (destruction on stack unwinding is not performed correctly, \
and catch(...) +                # eats also Win32 SEH exceptions)
+                ehOptions=[opt for opt in c.compile_options if \
opt.startswith('/EH')] +                if '/EHsc' not in c.compile_options:
+                    if len(ehOptions):
+                        # We already have some unexpected /EH... option around
+                        print 'WARNING: strange exception semantics specified in \
compile_options:', ehOptions +                    else:
+                        c.compile_options.append('/EHsc')
+        except Exception as ex:
+            print "Couldn't patch compiler arguments, got exception: ", repr(ex)
+
         build_ext_base.build_extension(self, ext)
 
         self.force = oldforce


[Attachment #4 (text/plain)]

_______________________________________________
PyQt mailing list    PyQt@riverbankcomputing.com
http://www.riverbankcomputing.com/mailman/listinfo/pyqt

[prev in list] [next in list] [prev in thread] [next in thread]