What steps will reproduce the problem? Unknown, just compiling a huge .py with many classes.

What is the expected output? What do you see instead? I added a print lcp to display the [ ... ] in the log. if len(lcp) > 1: + print lcp in /usr/lib/python2.7/dist-packages/shedskin/typestr.py:~190

[class none, class int_, class str_] [class none, class str_, class int_] Traceback (most recent call last): File "/usr/bin/shedskin", line 3, in <module> shedskin.main() File "/usr/lib/python2.7/dist-packages/shedskin/init.py", line 138, in main start(gx, main_module_name) File "/usr/lib/python2.7/dist-packages/shedskin/init.py", line 128, in start generate_code(gx) File "/usr/lib/python2.7/dist-packages/shedskin/cpp.py", line 2715, in generate_code gv.header_file() File "/usr/lib/python2.7/dist-packages/shedskin/cpp.py", line 248, in header_file self.visit(self.module.ast, True) File "/usr/lib/python2.7/compiler/visitor.py", line 57, in dispatch return meth(node, *args) File "/usr/lib/python2.7/dist-packages/shedskin/cpp.py", line 570, in visitModule self.module_hpp(node) File "/usr/lib/python2.7/dist-packages/shedskin/cpp.py", line 348, in module_hpp self.class_hpp(child) File "/usr/lib/python2.7/dist-packages/shedskin/cpp.py", line 711, in class_hpp self.class_variables(cl) File "/usr/lib/python2.7/dist-packages/shedskin/cpp.py", line 807, in class_variables self.output(nodetypestr(self.gx, var, cl, mv=self.mv) + self.cpp_name(var) + ';') File "/usr/lib/python2.7/dist-packages/shedskin/typestr.py", line 119, in nodetypestr return typestr(gx, types, None, cplusplus, node, check_extmod, 0, check_ret, var, mv=mv) File "/usr/lib/python2.7/dist-packages/shedskin/typestr.py", line 124, in typestr ts = typestrnew(gx, types, cplusplus, node, check_extmod, depth, check_ret, var, tuple_check, mv=mv) File "/usr/lib/python2.7/dist-packages/shedskin/typestr.py", line 194, in typestrnew elif not node or infer.inode(gx, node).mv.module.builtin: AttributeError: 'NoneType' object has no attribute 'module'

So basically, what I understand is that processing a certain string/int is incorrect in my code, I'll try to dive more in printing node incriminated.

What version of the product are you using? On what operating system? shedskin 0.9.4, python 2.7.6

Cheers, Olivier

thanks for reporting. could you please send me the sourcecode, possibly in private, so it's much easier to have a look? :-)

Ok so the fix here is: /usr/lib/python2.7/dist-packages/shedskin/typestr.py:191 if set(lcp) == set([python.def_class(gx, 'int_'), python.def_class(gx, 'float_')]): return conv['float_'] elif not node or not infer.inode(gx, node).mv or infer.inode(gx, node).mv.module.builtin:

Afterwards I get a list of types that are marked as "expression has dynamic" I'll fix my code with displayed warnings and keep you updated.

I get that it's easier to have a look with the python sourcecode. My problem is that his code is developed under a NDA with a thirdparty company. So I can't provide it until you sign the NDA as well :) Legals will definitely lead us to a sad end !

Cheers, Olivier

the real problem is not here, but I guess not crashing here is a definite improvement ^^ will have a look at your patch and probably commit it later today.

btw how many lines of code is your program, and how long does it take to analyze?

The program is composed of 2 files, one of 700 lines, the other of 2500 lines. It takes approx 10-15 minutes to get to the warning list (Core2 L9400) The type analysis is very long. But well it definitely performs as I required and the execution time gain later is tremendeous (even compared to pypy). Basically the program I write performs optimizations in assembly code for an embedded platform before the final link pass. Cheers, Olivier

Well I just saw I used my wife's google acocunt to post that. I've already reported a bug sometimes ago.

sounds like an interesting tool. please let me know when you run into other problems or if I can help optimize the result. you've probably already seen the standard optimization tips in the wiki documentation.

I'm afraid I hadn't checked out how to optimize the conversion process as it's definitely not a show stopper !

Ok, I finally have a valid conversion, so basically, the unsupported thing is this:

map = {} def function(self,arg1,arg2): map["somestring"] = ["string1","string2",int(arg1),int(arg2),"string3"]

function(3,4) map["somestring"][2] += map["somestring"][3]

or something alike. The unsupported part is the array of mixed type. when replaced by a class, it works like a charm.

Cheers, Olivier

once you applied above fix, did shedskin give you useful warnings that pointed you to the problem..?

oh sorry you already said you got 'dynamic type' warnings.

No problem, Yes after the fix, warnings were relevant. So I was able to find a solution, the only tricky stuff was the array with mixed item type.

this line was already fixed in GIT, good.. :-)

https://gitorious.org/shedskin/mainline/commit/78b4ef969f3b321dcf7a0a43a4d86d114de005b2

did the analysis get much faster after fixing the dynamic types?