Based on some semi-formal performance tests, read guards turned out to be
slower than the old code. With this change, we're using simple function calls again to perform security checks. But the calling sequence is intended to be easier to comprehend than the old code. Now instead of DT_String.String subclasses having a validate() method attached to them, they subclass AccessControl.DTML.RestrictedDTML, which provides a guarded_getattr() method and a guarded_getitem() method. Note that the functionality of guarded_getattr() used to be implemented both in C and Python (in cDocumentTemplate and DT_Util), but now it's in one place, ZopeGuards.py. Thus it's not only reusable but easy to optimize. I ran all the tests and ran the new code through the profiler again. The change sped up restricted code a little more than expected, which is definitely a good thing, but that may indicate that nested scopes have a hidden speed penalty. Also, RestrictedPython is now restrictive about printing to targets and two forms of augmented assignment had to be forbidden.
Showing
Please register or sign in to comment