functools.lru_cache does not cache exceptions

[Zero]

BPO	21537
Nosy	@rhettinger, @webknjaz

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = None
closed_at = <Date 2014-05-19.19:44:30.300>
created_at = <Date 2014-05-19.18:58:59.519>
labels = ['invalid', 'type-feature', 'library']
title = 'functools.lru_cache does not cache exceptions'
updated_at = <Date 2019-10-31.14:22:37.351>
user = 'https://bugs.python.org/Zero'

bugs.python.org fields:

activity = <Date 2019-10-31.14:22:37.351>
actor = 'Zero'
assignee = 'none'
closed = True
closed_date = <Date 2014-05-19.19:44:30.300>
closer = 'rhettinger'
components = ['Library (Lib)']
creation = <Date 2014-05-19.18:58:59.519>
creator = 'Zero'
dependencies = []
files = []
hgrepos = []
issue_num = 21537
keywords = []
message_count = 3.0
messages = ['218808', '218810', '302158']
nosy_count = 2.0
nosy_names = ['rhettinger', 'webknjaz']
pr_nums = []
priority = 'normal'
resolution = 'not a bug'
stage = None
status = 'closed'
superseder = None
type = 'enhancement'
url = 'https://bugs.python.org/issue21537'
versions = ['Python 3.4']

[Zero]

While examining the implementation for lru_cache, it came to my attention that the wrappers ignore the possibility of exceptions. Is this on purpose? If the cache is designed to reduce the overhead of running certain functions, it seems like lru_cache should handle returned values along with raised exceptions.

[rhettinger]

Is this on purpose?

The short answer is yes. It is a can of worms and there isn't much of a payoff.

[webknjaz]

Hi,

I've got similar need (caching exceptions). My use case is syscall, which is not going to result in something different regardless of the number of times it's called. But I'd like to be able to have exception already raised for me by lru_cache, so that I won't poke kernel with no need.

I'd suggest adding cache_exceptions=False key to the decorator to let user control whether it's worth enabling. This might be useful in stdlib.