[Swig-devel] [ swig-Bugs-2528941 ] Generated code breaks strict aliasing rules

[SourceForge.net <no...@so...>]

Bugs item #2528941, was opened at 2009-01-22 18:03
Message generated for change (Tracker Item Submitted) made by Item Submitter
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=101645&aid=2528941&group_id=1645

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: code generation (general)
Group: None
Status: Open
Resolution: None
Priority: 5
Private: No
Submitted By: Gran Uddeborg (goeran)
Assigned to: Nobody/Anonymous (nobody)
Summary: Generated code breaks strict aliasing rules

Initial Comment:
The code generated by SWIG 1.3.36 breaks the rules for strict aliasing.  Recent versions of GCC triggers this problem.  To reproduce, use the attached test case with these commands:

swig -c++ -java break_strict_aliasing.swg

c++ -shared -fPIC -O2 -o libbreak_strict_aliasing.so -I$JAVA_HOME/include -I$JAVA_HOME/include/linux break_strict_aliasing_wrap.cxx

$JAVA_HOME/javac break_strict_aliasing.java

CLASSPATH=. LD_LIBRARY_PATH=. $JAVA_HOME/java break_strict_aliasing

This should print the number "1", but it doesn't.  Using GCC 4.1.2 and Java 1.6.0 on a x86_64 RHEL 5 machine gives the result 1082477248.  In other combinations I've got "0", and yet others have crashed.

Adding the -Wall flag to the compilation will give warnings about the code triggering the problem.

Lowering optimisation with -O1, -O0, no -O flag, or -fno-strict-aliasing are workarounds for the problem.  with the obvious drawback of redusing the optimisers efficiency.

----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=101645&aid=2528941&group_id=1645