Python is easy, just use the contextmanager decorator to turn the generator into a context manager that is also at the same time a decorator factory! For real though i love this❤
This is cool, this decorator-context manager equivalence reflects that a decorator what basically is doing is "managing" the "context" of the wrapped function and that a context manager is basically is "decorating" an annonymous IIFE. One really beautiful abstraction.
Cool, didn't know you could do this. This can only be done when the decorator returns the unmodified output of the given function being called with unmodified input... correct? Clearly these kinds of decorators exist... like ones that register stuff, log stuff, etc... but a lot of decorators are used to modify either input, output, coerce stuff, etc. If this can't help in those cases the beginning of this video is actually a great reference for how to properly type regular decorators.
Excuse me, what the hell. This is super useful! Plus, much easier to test! Holy crap, I will try to remember to use this. Was this a planned feature? Or a coincidence by how Python is built? Or added afterwards...?
I don't even write python that much, randomly saw this vid in recommended. It looks amazing! Even without the fact that you don't need to type anything, it's so much nicer than defining 3 nested functions. Tbh python decorators are the only feature i miss in other languages, but they're so nice.
That's such a nice and smart simplification, completely decoupling the function typing rather than going generic with it. Can the context manager decorator do something like logging the first parameter while also timing it, or are the parameters/return values completely out of reach?
Wait how you made nano like this ?? I would like to look at your dotfiles or something like this as I can't believe that nano can be like this. It's like vim or neovim but it's nicer.
this sort of intentionally prevents you from doing nasty signature mutation. the concatenate example in the other video was to demonstrate the feature and not really something I would ever do in reality
How long has it been like that? Holy - soooo much easier. Is there any downside to this? Why would you ever use the "old" way? Imagine not using functools either...
Damn wish I knew this sooner. However the drawback of not accessing arguments and return values kinda limit the scope of this approach (e.g. custom caching decorator)
2:40 is there a specific reason why you use time.monotonic instead of time.perf_counter, because as far as i can remember you have always used time.monotonic in your videos?
Generator has a yield type (the type in the yield), send type (usually only required for coroutines) and return type (the final returned value after the generator is exhausted)
This is awesome, and makes for a much nicer typing experience. Quick question on the @timing_ctx example: is "try: finally:" really needed in this case, or just a "nice to have" for the sake of completeness? i.e. would the section after the "yield" be skipped on an exception inside the decorated function if we ditch the try/finally? I'm asking this because imho, the decorator "setup/teardown" sections would look even better/simpler/more readable without the extra indentation that the finally: block unfortunately imposes on the "teardown" section of it. Thanks in advance, asottile, you are a legend!
![เปิดบ้าน นาราเครปกระเทย 1 ไร่ ขายขาดทุนหลัก10ล้าน เอามาใช้หนี้!!! l [Nickynachat]](http://i.ytimg.com/vi/0jz84VFWVP4/mqdefault.jpg)
Python is easy, just use the contextmanager decorator to turn the generator into a context manager that is also at the same time a decorator factory!
For real though i love this❤
This is cool, this decorator-context manager equivalence reflects that a decorator what basically is doing is "managing" the "context" of the wrapped function and that a context manager is basically is "decorating" an annonymous IIFE. One really beautiful abstraction.
wow that's a hack that I can use a lot more of
I had no idea that creating a context manager like that also created a decorator. Blowing my mind, here.
good god man this is fantastic - thanks!
Cool, didn't know you could do this. This can only be done when the decorator returns the unmodified output of the given function being called with unmodified input... correct?
Clearly these kinds of decorators exist... like ones that register stuff, log stuff, etc... but a lot of decorators are used to modify either input, output, coerce stuff, etc.
If this can't help in those cases the beginning of this video is actually a great reference for how to properly type regular decorators.
Excuse me, what the hell. This is super useful! Plus, much easier to test! Holy crap, I will try to remember to use this.
Was this a planned feature? Or a coincidence by how Python is built? Or added afterwards...?
yeah it got added after contextmanager was introduced
@@anthonywritescode so you basically dismissed decorators and then you used decorators
???
So obvious and straightforward! Love it
I don't even write python that much, randomly saw this vid in recommended. It looks amazing! Even without the fact that you don't need to type anything, it's so much nicer than defining 3 nested functions.
Tbh python decorators are the only feature i miss in other languages, but they're so nice.
That's such a nice and smart simplification, completely decoupling the function typing rather than going generic with it. Can the context manager decorator do something like logging the first parameter while also timing it, or are the parameters/return values completely out of reach?
You can get access to anything with the inspect module, but things will start getting slow.
nope. though arguably those are need special handling anyway and I would not recommend personally
There’s also the contextdecorator class that’s worth looking into
Wait how you made nano like this ?? I would like to look at your dotfiles or something like this as I can't believe that nano can be like this. It's like vim or neovim but it's nicer.
This isn't nano. This is babi. The gigachad here wrote it himself
framing this lol. this is my editor I made github.com/asottile/babi
@@anthonywritescode Wow what a crazy one. I will look at it :)
Very interesting! Is this possible for concatenate introduced in PEP 612? like the one example in your previous video adding a logger
this sort of intentionally prevents you from doing nasty signature mutation. the concatenate example in the other video was to demonstrate the feature and not really something I would ever do in reality
How long has it been like that? Holy - soooo much easier. Is there any downside to this? Why would you ever use the "old" way? Imagine not using functools either...
the only downside is no argument hackery signature modification but imo that's a good thing
This is cool but what if you want override/pass the args/kwargs of the func or you want to get the name of the func?
you can't, but I see that as a good thing
Amazing! So all these times I've been creating context managers, I could've also used them as a decorator? 🤯
This is why I was sad to see verbose typing in Python. Reminds of the same thing that happened in Java. Of course, it's still optional, but still.
This is incredible!
Damn wish I knew this sooner. However the drawback of not accessing arguments and return values kinda limit the scope of this approach (e.g. custom caching decorator)
2:40 is there a specific reason why you use time.monotonic instead of time.perf_counter, because as far as i can remember you have always used time.monotonic in your videos?
it's an alias on the platforms I care about and good enough on windows
@@anthonywritescode ah i see, thanks
This feels like black magic
Does it make any sense if you are not into typing?
yes, it is still a simpler way to make decorators
Is there no down sides to this?
Thanks!
That's super smart
What is the [None, None, None] typing for on the generator?
Generator has a yield type (the type in the yield), send type (usually only required for coroutines) and return type (the final returned value after the generator is exhausted)
@@philipreinhold1180 thank you!
th-cam.com/video/DTegfCNAXoM/w-d-xo.html for my video on the subject
I just wondered if you use an IDE at work
why wouldn't I use my own text editor?
Anyway to generalize this to create a decorator that decorates another decorator?
I mean, it should work
That's kind if new.... Contex manager working as decorator 🧐.... I need try....
This is awesome, and makes for a much nicer typing experience. Quick question on the @timing_ctx example:
is "try: finally:" really needed in this case, or just a "nice to have" for the sake of completeness?
i.e. would the section after the "yield" be skipped on an exception inside the decorated function if we ditch the try/finally?
I'm asking this because imho, the decorator "setup/teardown" sections would look even better/simpler/more readable without the extra indentation that the finally: block unfortunately imposes on the "teardown" section of it.
Thanks in advance, asottile, you are a legend!
yeah finally is necessary to work in the context of exceptions -- this is also true in the decorator code too (if I skipped it there it's a mistake!)
Blasphemy
first!
Am I noticing correctly that the context manager method has 1000x overhead? 2ms vs the 0.002ms of the decorator!?
It's 1.7us for the decorator and 1.9us for the context manager. The 200ms one is after he added the with block containing a sleep.
I intentionally added a sleep so the timing output was interesting
Ahh, I was just looking at the time beyond the sleep (like 0.207 at 6:19). Maybe it’s a sleep resolution thing.
ah probably just python show plus vm plus recording
At this you should probably go writing rust code. Easier than all this
toxic