[ZODB-Dev] BTree.setdefault(), BTree.pop()

[Tim Peters]

[Dmitry Vasiliev]
> I want to add .setdefault(key, default) (with mandatory 'default'
> argument) and .pop(key, default) methods to BTrees so BTree interface
> will be closer to Python 2.3 dictionaries interface.
>
> Any objections?

I'd like to hear people say they're going to *use* these in the context of
BTrees first.  Just piling on new methods for its own sake creates ongoing
maintenance burdens without payback.  Offhand, neither of these methods make
particularly good sense to me for the ways BTrees are used.  For example,
the fundamental reason for dict.pop() to have a default in Python is to
avoid the need for a critical section in multithreaded Python apps; but
BTrees are very rarely used in apps where two threads may be mucking with
the same in-memory BTree at the same time (note that two threads in ZODB
mucking with the same *database* BTree at the same time is common, but
because each thread has its own connection to the database these threads are
nevertheless mucking with distinct in-memory BTrees).

BTW, part of .setdefault() *functionality* has been in BTrees from long
before Python added it to dictionaries:

>>> from BTrees.OOBTree import *
>>> b = OOBTree([(1, 2)])
>>> b.insert(1, 8)   # does not replace b[1], and returns false
0
>>> list(b.items())
[(1, 2)]
>>> b.insert(2, 4)   # adds b[2], and returns true
1
>>> list(b.items())
[(1, 2), (2, 4)]

The difference is that, "win or lose", .setdefault() returns the value
associated with the key, but doesn't tell you whether you won or lost.
insert() tells you whether you won or lost, but doesn't tell you the current
value associated with the key.  The existence of insert() was driven by a
specific use case, explained in the BTrees Interfaces.py.  Use cases should
drive this.