Message123312
It seems to me that an explicit raise inside an except block should *not* chain exceptions by default. The rationale for chaining exceptions is to detect bugs in the exception handler:
try:
something
except SomeError:
y = 1/x # oops, what happens when x == 0?
but an explicit raise isn't a bug. If you want it to chain for some reason, it's easy to do:
try:
something
except SomeError as exp:
raise MyException from exp
or otherwise set __content__, but there's no simple way to avoid chaining except via boilerplate code.
I frequently use the EAFP idiom to detect error conditions, and chained exceptions exposes details to the caller that are irrelevant implementation details. Here's a common example:
def process(iterable):
try:
x = next(iterable)
except StopIteration:
raise ValueError("can't process empty iterable")
continue_processing()
The fact that ValueError was raised in response to a StopIteration is an irrelevant implementation detail that shouldn't be, and wasn't in 2.x, exposed.
Another practical example is avoiding isinstance checks:
def func(x):
try:
x + 0
except (ValueError, TypeError):
raise MyException('x is not a number')
do_something_with(x)
I note that PEP 3134 explicitly listed the issue of *not* chaining exceptions as an open issue. Now that chained exceptions are being seen in the wild, it seems that the inability to easily suppress chaining *is* a real issue for some users:
http://mail.python.org/pipermail/python-list/2010-October/1258583.html
http://mail.python.org/pipermail/python-list/2010-December/1261738.html |
|
Date |
User |
Action |
Args |
2010-12-04 01:45:45 | steven.daprano | set | recipients:
+ steven.daprano, ncoghlan, draghuram, poke |
2010-12-04 01:45:45 | steven.daprano | set | messageid: <1291427145.5.0.342288561466.issue6210@psf.upfronthosting.co.za> |
2010-12-04 01:45:41 | steven.daprano | link | issue6210 messages |
2010-12-04 01:45:41 | steven.daprano | create | |
|