[pypy-dev] Bug hunting in rctime becomes wtf

[Lawrence Oluyede]

I had a nasty bug like this one in the compilation process of the rctime module:

[translation:ERROR] Error:
[translation:ERROR]  TyperError: non-constant tuple index
[translation:ERROR] .. block at 34 with 1 exits
[translation:ERROR] .. v0 = getitem(tup_0, ___i_0)

You can reproduce it decommenting asctime function from interp_time.py
and in the __init__.py file of the rctime module.

After some digging I've found the source of the problem in the code
Armin and I have checked-in at the sprint to add the support for
"*args" arguments (see visit_args_w in pypy.objspace.cpy.function). I
tried patching it but I'm not very into cpy objspace and the compiler
mechanism under the hood.

The "wtf" thing came out when I tried to use the file with this
"stupid" patch applied:

rhymes at groove:~/scc/pypy/pypy/objspace/cpy $ svn diff
Index: function.py
===================================================================
--- function.py (revision 29953)
+++ function.py (working copy)
@@ -61,7 +61,7 @@
         tramp.inputargs.append('*' + basename)
         tramp.wrappings.append('%s = []' % (argname,))
         tramp.wrappings.append('for ___i in range(len(%s)):' % (basename,))
-        tramp.wrappings.append('    %s.append(___W_Object(%s[___i]))' % (
+        tramp.wrappings.append('    %s.append(___W_Object(%s[0]))' % (
             argname, basename))
         tramp.passedargs.append(argname)

The thing surprised me is the compilation after the patch goes fine
and (!!) the module works correctly. asctime() does it's dirty job.

So, why does it work?

-- 
Lawrence
http://www.oluyede.org/blog