metapensiero.asyncio.tasklocal
- author:
Alberto Berti
- contact:
- license:
GNU General Public License version 3 or later
An asyncio’s Task-local variable container
Usage
A drop-in replacement for threading.local object that stores per-task variables (instead of per-thread). You can use it like this:
import asyncio
from metapensiero.asyncio.tasklocal import Local
mylocal = Local()
@asyncio.coroutine
def task_a():
mylocal.a_var = 'foo'
yield from asyncio.sleep(0.001) # switches to task_b
assert mylocal.a_var == 'foo'
print(mylocal.a_var)
@asyncio.coroutine
def task_b():
mylocal.a_var = 'bar'
yield from asyncio.sleep(0.001) # switches back to task_a
assert mylocal.a_var == 'bar'
print(mylocal.a_var)
loop = asyncio.get_event_loop()
a = asyncio.Task(task_a())
b = asyncio.Task(task_b())
loop.run_until_complete(asyncio.wait((a, b)))
# this prints out:
# foo
# bar
Really the kind of object to be tracked as “current” is configurable by passing-in a different Discriminator instance to the Local class.
By default it will use the TaskDiscriminator supplied with the package. It will use standard asyncio’s loop detection techniques (by delegating it to asyncio.Task.current_task()) but if you want to track tasks of a particular event loop, you can supply your own TaskDiscriminator instance to the Local instance initialization, like this:
from metapensiero.asyncio.tasklocal import Local, TaskDiscriminator mylocal = Local(discriminator=TaskDiscriminator(loop=my_loop))
You can also track another set of object by subclassing the BaseDiscriminator class:
from metapensiero.asyncio.tasklocal import BaseDiscriminator
class ASampleSessionDiscriminator(BaseDiscriminator):
def current(self):
return current_session()
mylocal = Local(discriminator=ASampleDiscriminator())
You have to implement at least the current method.
Like threading.local class, this class supports subclassing, in which case it will re-execute the __init__ method for each one of the tracked objects:
call_counter = 0
class CustomLocal(Local):
a_value = None
def __init__(self, a_value):
self.a_value = a_value
nonlocal call_counter
call_counter += 1
mylocal = CustomLocal('foo')
Here mylocal.a_value will be initialized to foo for every tracked object (asyncio’s tasks by default). Here call_counter will count the number of every tracked object in which the mylocal object has been accessed.
Testing
To run the tests you should run the following at the package root:
python setup.py test
Build status
Changes
0.1 (2015-12-14)
First release complete with tests
0.2 (2015-12-26)
Fix CHANGES and packaging
Metadata
Release files for metapensiero.asyncio.tasklocal 0.2
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| metapensiero.asyncio.tasklocal-0.2.tar.gz | 4.7 kB | Without provenance | Details for metapensiero.asyncio.tasklocal-0.2.tar.gz |
Release files / metapensiero.asyncio.tasklocal-0.2.tar.gz
| Download URL | metapensiero.asyncio.tasklocal-0.2.tar.gz |
|---|---|
| Size | 4.7 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
a77ca539ea68636d5b682032c327d038d315b6dd0228646ca4f8998662643c0d
|
|
BLAKE2b-256 checksum How to use checksums |
9df518985d83c21bd58d0d6ab680938a1125fb395bd389f08fdd80f8c167bcce
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
Security for metapensiero.asyncio.tasklocal 0.2
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.