Hello,

I jump on the opportunity offered by the previous thread to tell a little story (serious onl
y people disgard). Guess this is a candidate for Bug of the Year.

I was filtering (parse) results to keep only nodes from a given set of types:

for child in node:
    if child.pattern in item_types:
        items.append(child)
    else:       # recursion
        ...

For a weird reason, _all_ child nodes were collected. So, I printed out node types at start 
of loop, but all was ok. After a while wondering, I printed patterns in the if branch, too -
- which seemed to myself completely irrational:

for child in node:
    print child.pattern
    if child.pattern in item_types:
        print "*** type OK\t%s" %child.pattern
        items.append(child)
    else:       # recursion
        ...

This gave me results such as:
digit::[0..9] 
*** type OK     section::header body footer

The second expression is the one of another pattern in play during this test. Well, had a ha
rd time finding out how the pattern, or its expression, could change between both prints. Ac
tually, the reasons were (hope you're confortable):
* "section" is a recursive pattern (some items in the body can be sections)
* I had overloaded "==" in the top Pattern type as a syntactic sugar to name patterns.

Both together mean that, when the "in" test is performed, any child is (1) renamed and (2)re
turned (for the renaming to work), like a false result to the "==" comparison, so that the a
ssertion always seems True.
[I have long considered this rule of python logic a flaw, but it's only me]
Actually, the logical relation is a bit obscured because, for me, patterns where compared fo
r identity, not equality.

The funny joke here is that this trick for naming patterns is not necessary in normal use, o
nly handy for testing. So that, probably, I'm the only one who could step on the bug.
For the little story, I did not find the solution, rather it fell into my head from who-know
s-where while walking in the street on the way to market ;-)

Denis
------
la vita e estrany


_______________________________________________
Tutor maillist  -  Tutor@[...].org
To unsubscribe or change subscription options:
http://mail.python.org/mailman/listinfo/tutor