Home Original page

nested scopes

Tim Peters tim.one at home.com
Sat Feb 3 14:57:50 EST 2001 
[Tim]
> ...
> The determination of which names are local is still done at
> compile-time in the presence of "import *" and "exec", but then
> the results can't be trusted, and the compiler generates
> different code that does all sorts of crap under the covers to
> give "the expected" results -- most of the time. ....

[Rainer Deyke]
> "Most of the time" is significant here.

I used to lose about 100 hairs a day.  Now I think it's closer to 101.  Much
more significant to me <wink>.

> There are at least two inconsistencies in functions with 'import *'
> as of 2.0.  Can you say "source of subtle bugs"?

In this case, no.  Yes, "import *" is a notorious source of subtle bugs, but
all of them reported in real life are due to its advertised behavior (like
"hey, 'open()' is broken!", "works for me", "no, it's broken", "what did you
change?", "nothing!", "did it used to work?", "well, yes, but then it broke
for no reason at all!", "and you changed nothing", "well, nothing
relevant -- it just broke!", "so you did change something -- what (exactly)
did you change?", "I added a global counter, so maybe the compiler generates
bad code now ... and, OK, sure, I added 'from os import *'", ... and that
one is so frequent the replybots automatically respond  "from os import *"
whenever somebody complains "hey, 'open()' is broken!" now).

I posted some cases after the 0.9.9 release that showed endcase differences
in "import *" and "exec" semantics due to the then-new local variable
optimization.  All of them were contrived.  Guido didn't care; neither did
I.  In the approximately seven years since then, to the best of my knowledge
no bug report on "import *" has ever been filed, and perhaps two were filed
on "exec" but in non-contrived cases that were clearly broken.

Nevertheless, if you fear the import* "bugs" *you* contrived, SourceForge is
open 24 hours a day for patches <wink>.

never-use-import-*-myself-except-in-an-interactive-shell-ly y'rs  - tim
More information about the Python-list mailing list