id,summary,reporter,owner,description,type,status,priority,milestone,component,version,resolution,keywords,cc 113,Problems with dict item access/assignment using attribute syntax,darkness-keyword-genshi.e8bcba@…,cmlenz,"There are a few problems in `genshi.template.eval` related to the transformations that allow `d.k` to work like `d[""k""]`. Here's a short-ish executable example: {{{ #!python from genshi.template import MarkupTemplate def run(content): return MarkupTemplate(content).generate().render() run("""""" Tests

${d.k} is still 1 (should be 2)

${d.k} should now be 3

"""""") }}} Executing this with Genshi 0.4 or SVN (as of r569) will cause an exception because of the `d.k = 3`; comment that out and you'll see the wrong values output in the template. There are other things that fail with the current Genshi too, mostly related to AST nodes that weren't included in `genshi.template.eval`; see the tests included in the patch for more examples. The attached patch tries to solve these problems by essentially converting `d.k` to `AccessWrapper(d).k` and `d[""k""]` to `AccessWrapper(d)[""k""]`. `AccessWrapper` then tries both attribute and item access, in an order depending on which was used. To me this patch looks a little ugly. (In my defense I think the original use of `_lookup_attr`/`_lookup_item` wasn't very pretty either. :) ) I assumed that converting all those `_lookup_*` calls to an object instantiation followed by another function call for the operator overload would consume more time and memory. However, in my testing with both Genshi from SVN versus my patched Genshi, I was unable to see a significant difference in either time taken or memory used. My speculation is that other modules used from `genshi.template.eval` are so much more costly (i.e. the `compiler` package) that any performance degradation due to my patch becomes immaterial.",defect,closed,major,0.4.1,Expression evaluation,devel,fixed,genshi eval attr item dict assignment,