On 9/28/07, Guido van Rossum <[EMAIL PROTECTED]> wrote: > Well, maybe this is a good enough argument to give up.
Not quite yet... I still see two potential solutions, depending on whether or not the exclusion is sticky. Details below. ========= If the exclusion is sticky, then add (implicit) flags saying "seen a string" and "seen a byte". Similar logic is already there, in that "seen a non-string" replaces the lookdict function. The most common case (exact unicode in an exact unicode-only dict) would stay the same as today, but the other cases would have some extra type-checking. ========= If the exclusion is based on current contents, then we can add a count; my concern is that keeping this efficient may be too hacky. It looks like there is room for exactly one more pointer (-sized count variable) before small dicts bleed to a third cacheline. Because of this guard, bytes and strings can never appear in the same dict, so at least one count is zero. Because dict entries are 3 pointers long, there can never be more than (Py_ssize_t / 2) entries, so the sign bit can be repurposed to indicate whether the count refers to strings or bytes. (count==0 means no bytes or strings; count==5 means 5 string keys; count==-32 means 32 bytes keys.) -jJ _______________________________________________ Python-3000 mailing list Python-3000@python.org http://mail.python.org/mailman/listinfo/python-3000 Unsubscribe: http://mail.python.org/mailman/options/python-3000/archive%40mail-archive.com
