Haven't checked it yet, but does this also mean that a list comprehension for a class variable can now reference are pre-defined class variables? Previously it would fail as class definitions are not an enclosing scope.
I usually use f"""{"somthing"}"""" if the double quote cannot be avoided. But usually f"{' something '}" would get the job done. Also use f"{{ {something} }}" when dealing with some command prompt requires {} as parameters.
6:30 Just a note that, as with any function, generation of the code happens only once, at compile time, but creation of the function _object_ (ref to code + environment binding) happens every time that statement is executed.
@@headlikeahole6548 " in python they are just for IDE hints." Even hints would be better than the instanity of gender fluid variables and non binary booleans !!! Python is a mess on every level.
@@jamesflames6987 But they've really internalised the lesson from C++ that you can say you're OO without having to actually enforce it, so you can still have functions floating around any old place and of course none of the flow-control is OO at all. But, you know - whitespace!
I think with `kwargs`, it's not that having the same type for each `kwarg` is a natural solution but it's more like it's consistent with the typing of `args` where it *does* make more sense to have the same type for each arg.
11:51 There's exactly one problem with `itertools.batched`, it's passed tense. All the other itertools are present tense, i.e. `groupby` rather than `groupedby`.
never thought of that before, but It's pretty much same as in build-in python functions sorted() and reversed(). I mean we have ex. not mapped() but map(); not enumerated() but enumerate() tho I have never see anyone complain about it, even tho it's kinda not consistent
@@simp- Sorted and reversed are different, they suggest that the collection is not modified in place, but a copy/iterator is returned, whereas you can't really map or enumerate in place. As for batched, I think they chose that name because batch can also be interpreted as a singular noun. Kotlin also uses the name chunked for a similar method, and their naming conventions are pretty consistent.
@@nigeltan7766 It would be "bad" to use it now since that version is no longer getting any updates including security fixes. Newer versions have the new language features and much better performance anyway.
@@sid6645 yes, I the real world it would definitely be premature to fast track deprecation. It's just something some python package maintaners dream about because it can greatly reduce their work.
10:26: C++ typing is a good thing! Typing has the two purposes of 1. helping the programmer catch errors, 2. creating an efficient compiler producing lightning fast machine code (if the programmer chooses to). If the answer on purpose 2. is "use another language" then Python isn't that language.
Ever since python 2 went out with python e (2.7.18), I've been thinking it would be aesthetically pleasing for the python 3 line to end with 3.14.15. Perhaps if GIL removal by default gets released as Python 4...
Hi James, if you are gathering points for a new "Nooby Python habits" video, I might have one. Not sure if I missed it in the list or it is just so stupid that you did not think about it. In most of my codes I initialized instance attributes as class attributes instead of initializing them in the __init__() method. Mainly to make the code shorter and more readable. I mean, sure... Those are shared by all instances of a class, but if they are immutable objects they work like instance attributes. I realized my very stupid nooby habit when I created mutable objects (list and dict in my case) which is then shared between all instances. I think for newcomer this behavior is not quite obvious and can lead to silent misbehavior. I thought it is worth mentioning. Thank you for your great videos. Even considering the statement above, I'd say they made me a quite good Python dev :)
As an NVIDIA Jetson which shipped Ubuntu 20.04 (i.e. python38) and lots of proprietary NVIDIA libraries which only have python binding for 3.8, I start sweating at the thought of 3.8 being dropped next year.
I know exactly what you mean. As much as I love the new features of updates to python I get really worried the closer 3.8 gets to being dropped haha. Too many things would stop working for me :(
You'd probably be fine for a couple more years until 2025 when ubuntu 20.04 LTS standard security maintenance also runs out. Probably Nvidia calculated it based on that. That's definitely within reason. Most consider a newly deprecated python version to have year or two more in them in enviroments like that.
@@volbla "Let's update them ". Let us .... ? That will be you and who else picking up the burden of the massive work load imposed by the lamers at Python constantly doing non compatible "upgrades" and breaking everything in sight?
On types or dictionary question you asked about. In c# we have enums that you r able to cast the value or str. And values could change depending on what the enum is used for. Then you want to use kwargs because you don't know how the structure changed. Like in reflection scenarios. Thanks great update And you explanation about square and override is a nice explanation of polytechnic methods if they were wondering how polymorphism relate back to real life scenarios instead of foo
Would like to hear more about your other preferred utils. How is batched implemented over an unindexed object, does it cast to a list or uses a generator ?
The buffer protocol is nice. Many times I've come across functions in the standard library that say they work with anything that implements the buffer protocol, when it's C api only and only bytes and bytearray implement it. Not that useful but kinda nice.
8/10 video. im gonna start using itertools batched. f string stuff was annoying having to use different quotes, etc. kinda wish single-quotes in python weren't a thing.
How did you learn about all the python language features you currently know how to use and what motivated you to learn them? Many times I watch your videos and I see you use something that I wasn't aware of before like __slots__ and other items like that.
Is the debugger improvement applied to the integrated vscode debugger or do you have to use the sys.monitoring yourself ? Sorry I don't really know how debuggers work internally
Not that I have a use case for it, but I am curious. How do immortal objects actually work? Is there a way from within python to mark an object as immortal or is that fully up to the C implementation to handle - so only code that is partially C can benefit from it atm?
Good stuff. Generics have come a long way. Now we need currying. Nits: is that min_max generic at 0:13 supposed to return triples? And at 0:49 shouldn't it return self.seq[item] or something?
I don't get why you would type hint kwargs that way instead of just type hinting the named kwargs after the "/, *," in the function signature, e.g. "def thing(self, /, *, kwarg1: int, kwarg2: str)"
The generic typing is the best! I'm disappointed of the performance though. 3.11 vs 3.10 was a clear upgrade, while 3.12 barely changes, some programs run even slower.
Immortal objects would be a no-op in pypy. Refcounting is just inherently ugly if you care about performance at all. I wish that python3 had just dropped refcounting and used a proper GC in python.
2:03 I don't understand, what's wrong with that? if you want to use same quotes, you have to escape them - seems very reasonable why was this changed? it feels more confuaing now that before
You mentionned setuptools. But not that *distutils* is completely removed from 3.12! setuptool still monkey-patches in its special vendored version. But that means if you never imported setuptools, then imports of distutils will fail.
Do you have to change the reference count for None anyway? I wrote a C python module a few years ago. And for numbers, I was told to not change the reference count when returning an integer from 0-255, because those are special objects that are never deallocated. It may be the same for None.
@@Spencer-wc6ew If you use PyLong_FromLong(x), then you don't need to increment the reference count, because the object you've just received has had its count incremented. But you can definitely crash Python by calling a function that returns `Py_None` without incrementing it. Call sys.getrefcount(None), find None's refcount, and call your function that many times. I assume the same holds true for the small integers, True, False, etc if you save the Python object in a global and repeatedly return it, rather than calling a getter.
I always set reference counts properly, for None as for everything else. I once wrote a test program that did nothing but decrement the refcount for Py_None until it crashed. I got up to about 512 decrements or so, maybe slightly more.
You are aware that anything in Python is a generic? "def foo(bar)" means bar is generic (as is the return value). It's not untyped either, because of "object" being the base for all Python values? There's only limited amount of usefulness in something like "[X where Parent]" as documentation.
I think the C++ syntax was stupid. And it still doesn’t let C++ (or Java, for that matter) expression Python concepts like function factories and class factories.
@@FrederikSchumacher Yes. Python is too generic, which doesn't allow to express the intent of the programmer. The generic specification is just a hint for IDEs to check the sanity of the code. For example, a class of type T, has methods that adhere to the same type T.
3:04 Having recently converted an application away from WSGI to ASGI instead, I could care less about things that might improve WSGI behaviour. WSGI is Python’s answer to PHP. ASGI goes beyond that, to let Python do things (like handle WebSocket connections) that PHP cannot.
Python 3.7 is now deprecated, huh? Feels weird, since that's the version I started coding on. But ofc it needed to go eventually. Other than it having a special place in my heart, there's no point in holding onto it
can someone explain to me how generics work? I've re-watched the entire 2 mins about 3 times and I don't quite understand what's going on. it really triggers me a big impostor syndrome because I work as a python dev, yet I've never dealt with this
I see generics get used a lot in languages like Java/Kotlin, C++, Typescript, and probably others. The idea is that the "Type" of the parameter is determined when the object is actually implemented/created. So when you're creating the structure of the object, you can't define what the type is, but you want to tell the type checker that "whatever is used during implementation, that type is also used (wherever you want it in your logic)" - this most often gets used when you're dealing with custom types, not really the built-ins. This allows you to build tools that accept any arbitrary types, but still tell the type checker that the type matters and reference it elsewhere. A basic abstract example might be: you want to track what movies, tv shows, and books you have in your "Library" - one of your methods on the library allows you to easily link sequels and prequels. In this case, you don't really care what the type of the object is (be it a movie or book or any other entertainment concept), but you want to ensure that whatever you link as the sequel is the same type. So you define the generic (commonly specified as "T" for "Type") and then you can now reference that in the rest of your code by saying "The parameter is a generic T and the sequel is the same type T" More commonly I see it used in situations where you are building packages or tooling, so your function/class can be typed even if you never know what might be used for it, all you know is that the type must match elsewhere in the logic.
from my understanding, in short, when you use generics with a function, whatever type you use as input for it, it also outputs that type. i guess it's useful for functions that are meant to process objects that behave similarly and have implemented similar methods or dunders. in the example in the video at 0:37, using generics in that function makes it clear that whatever numeric objects (or any object for that matter. it may also be objects like datetime where you can compare them) are used as input, the function will output object of that type, marking that specific type with 'T' , with a note that that object implements the greater than, less than, and equal dunders to allow comparison.
To me the easiest way to understand generics is to see how useful they are when typing lists, and more generally "container" classes. If you have to write your own list implementation, and you want to type your code, you'll soon face an impossible question to answer: what type do you use for the elements of your list? The answer is, well, you can't know. You can have lists of ints, of booleans, strings, etc. Clearly, the type of the elements of the list depends on each use that will be made of it -- therefore you have to tell the type checker that you have data whose type will change depending on how it is used. This is what type parameters are for, they're placeholders for the type checker to fill whenever your class is used.
Lists in python are generic. For example we can have a list of integers, or a list of strings, unfortunately, we can have a list of ints and strings ([1, "1"] is valid), which is often not what we actually want. To solve this, languages introduce a generic type T, that can be anything, int, string, double, myType, etc, but importantly, it can only be of that type only, not multiple types. So a generic list of type T[ ] can admit something like: [1, 2] and ["1", "2"] but not [1, "2"] .
The comprehension change is a 10/10. Literally no change to your code required but comprehensions become twice as fast.
twice?
Looking forward to my coworkers comprehending twice as fast!
I guess this also allows comprehensions to be evaluated in pdb session? Dunno, must test
Haven't checked it yet, but does this also mean that a list comprehension for a class variable can now reference are pre-defined class variables? Previously it would fail as class definitions are not an enclosing scope.
the f string change is so helpful, ive run into that one quite a few times
I usually use f"""{"somthing"}"""" if the double quote cannot be avoided.
But usually f"{' something '}" would get the job done.
Also use f"{{ {something} }}" when dealing with some command prompt requires {} as parameters.
you've run into using quotes inside quotes and deciding not to escape them?
6:30 Just a note that, as with any function, generation of the code happens only once, at compile time, but creation of the function _object_ (ref to code + environment binding) happens every time that statement is executed.
Great to see Python taking some inspiration from C++ such as typing and incomprehensible syntax.
Types in C++ affect your programs, in python they are just for IDE hints.
@@headlikeahole6548All of the complexity, none of the functionality.
@@headlikeahole6548 " in python they are just for IDE hints." Even hints would be better than the instanity of gender fluid variables and non binary booleans !!! Python is a mess on every level.
@@jamesflames6987 But they've really internalised the lesson from C++ that you can say you're OO without having to actually enforce it, so you can still have functions floating around any old place and of course none of the flow-control is OO at all. But, you know - whitespace!
@@jamesflames6987 LOL,🤣
I think with `kwargs`, it's not that having the same type for each `kwarg` is a natural solution but it's more like it's consistent with the typing of `args` where it *does* make more sense to have the same type for each arg.
1:52 and it used to be that you coudln't use backslashes in f-strings. Which bit a lot of people in the backside.
11:51 There's exactly one problem with `itertools.batched`, it's passed tense. All the other itertools are present tense, i.e. `groupby` rather than `groupedby`.
I read it as an adjective, rather than a verb. Like functools.partial.
never thought of that before, but It's pretty much same as in build-in python functions sorted() and reversed(). I mean we have ex. not mapped() but map(); not enumerated() but enumerate() tho I have never see anyone complain about it, even tho it's kinda not consistent
@@simp- Sorted and reversed are different, they suggest that the collection is not modified in place, but a copy/iterator is returned, whereas you can't really map or enumerate in place. As for batched, I think they chose that name because batch can also be interpreted as a singular noun. Kotlin also uses the name chunked for a similar method, and their naming conventions are pretty consistent.
@@piaIy oh, okey I get it, thanks a lot
As someone who maintains a lot of packages.. 3.7 disappearing is definitely the best. Can we fast track it to 3.10? :D
is python 3.7 bad? im no expert but am curious so pls elaborate
Losing 3.8 might be bad because of nvidia, as someone pointed out below.
@@nigeltan7766 It would be "bad" to use it now since that version is no longer getting any updates including security fixes. Newer versions have the new language features and much better performance anyway.
@@sid6645 yes, I the real world it would definitely be premature to fast track deprecation. It's just something some python package maintaners dream about because it can greatly reduce their work.
And I'm still using 3.6 for an old custom-build tensorflow version that was specifically compiled to run on an ancient GPU
10:26: C++ typing is a good thing! Typing has the two purposes of 1. helping the programmer catch errors, 2. creating an efficient compiler producing lightning fast machine code (if the programmer chooses to). If the answer on purpose 2. is "use another language" then Python isn't that language.
2:05 spooky, I literally just got bit by that this morning; I had no idea that's being/been addressed
im already excited for python pi (or just python 3.14) as it is now very very close
Ever since python 2 went out with python e (2.7.18), I've been thinking it would be aesthetically pleasing for the python 3 line to end with 3.14.15. Perhaps if GIL removal by default gets released as Python 4...
@@Liam3851cant wait for it’s end really..
pithon
Did python 1 stop with 1.414?
@@jan-lukas 1.6.1. Golden ratio!
Hi James,
if you are gathering points for a new "Nooby Python habits" video, I might have one. Not sure if I missed it in the list or it is just so stupid that you did not think about it. In most of my codes I initialized instance attributes as class attributes instead of initializing them in the __init__() method. Mainly to make the code shorter and more readable. I mean, sure... Those are shared by all instances of a class, but if they are immutable objects they work like instance attributes. I realized my very stupid nooby habit when I created mutable objects (list and dict in my case) which is then shared between all instances. I think for newcomer this behavior is not quite obvious and can lead to silent misbehavior. I thought it is worth mentioning.
Thank you for your great videos. Even considering the statement above, I'd say they made me a quite good Python dev :)
10:15 PUBLIC STATIC VOID MAIN STRING ARGS 🗣‼
As an NVIDIA Jetson which shipped Ubuntu 20.04 (i.e. python38) and lots of proprietary NVIDIA libraries which only have python binding for 3.8, I start sweating at the thought of 3.8 being dropped next year.
you're not alone
Let's update them 🙂
I know exactly what you mean. As much as I love the new features of updates to python I get really worried the closer 3.8 gets to being dropped haha. Too many things would stop working for me :(
You'd probably be fine for a couple more years until 2025 when ubuntu 20.04 LTS standard security maintenance also runs out. Probably Nvidia calculated it based on that. That's definitely within reason. Most consider a newly deprecated python version to have year or two more in them in enviroments like that.
@@volbla "Let's update them ". Let us .... ? That will be you and who else picking up the burden of the massive work load imposed by the lamers at Python constantly doing non compatible "upgrades" and breaking everything in sight?
Please let the 3.14 release be called “Pi-thon” 😂
uyese
The new generic type syntax is sooo welcome
On types or dictionary question you asked about. In c# we have enums that you r able to cast the value or str. And values could change depending on what the enum is used for. Then you want to use kwargs because you don't know how the structure changed. Like in reflection scenarios. Thanks great update
And you explanation about square and override is a nice explanation of polytechnic methods if they were wondering how polymorphism relate back to real life scenarios instead of foo
Finally!!! I've been waiting for this to start learning Python.
Been waiting for a 3.12 take from mCoding. Awesome!
Would like to hear more about your other preferred utils. How is batched implemented over an unindexed object, does it cast to a list or uses a generator ?
Generator. Unlike batched_even which needs the iterator length, batched does not
The buffer protocol is nice. Many times I've come across functions in the standard library that say they work with anything that implements the buffer protocol, when it's C api only and only bytes and bytearray implement it.
Not that useful but kinda nice.
10:14 That would need a “@final” decorator on the base class method. No doubt that will be coming ...
Actually, @final is already in the `typing` module since python 3.8 ;)
@@peco_17 Ha!
Only two more to go until we can start calling it πthon...
8/10 video. im gonna start using itertools batched. f string stuff was annoying having to use different quotes, etc. kinda wish single-quotes in python weren't a thing.
I just wish they did something different than double quotes, feels like wasted opportunity to make two characters do the same thing
@@michaelflynn6952Yeah. Almost every other scripting language has different handling for the two characters.
"and python 3.8 has a year to go so I can start pushing clients to upgrade to 3.9" :D
I also have had to write my own batched function. Since itertools has pairwise you might think it should also have an n-wise, but not so until now.
0:47 is (float, int) here mean that T will be a tuple, or will it be either a float or int?
Love this form of video with the ratings! Very engaging...
How did you learn about all the python language features you currently know how to use and what motivated you to learn them? Many times I watch your videos and I see you use something that I wasn't aware of before like __slots__ and other items like that.
Great vid as always 🎉
discord gang 🤙
Finally a youtuber who knows how to pronounce et cetera properly. You deserve a like just for that.
Can't wait for python 3.14, python π
Is the debugger improvement applied to the integrated vscode debugger or do you have to use the sys.monitoring yourself ?
Sorry I don't really know how debuggers work internally
Bump
Marvellous. Loved it.
woa! I love the static typing stuff
Static type gang here
👇
Can’t wait for Py 3.14
man you are literally the best python programmer i've ever seen
I'll be waiting for 3.14
wonderful update
Not that I have a use case for it, but I am curious. How do immortal objects actually work? Is there a way from within python to mark an object as immortal or is that fully up to the C implementation to handle - so only code that is partially C can benefit from it atm?
Looking forward to 3.14 version. I hope they gonna call it Pi-thon
Python 3.14 will be the Pi update.
Thank you!
Streamlit is not available in python 3.12
Good stuff. Generics have come a long way. Now we need currying.
Nits: is that min_max generic at 0:13 supposed to return triples? And at 0:49 shouldn't it return self.seq[item] or something?
Currying has long been in Python: functools.partial or functools.lru_cache
We already have function factories and class factories. What more do you need?
one of the best updates ever
`batched` function is GOAT
new way of creating types is a huge change.
Override is my favorite. I love it.
Love your music by the way
10:25 you forgot noexcept(noexcept(L)) lol
as someone who learned programming with java, the new generics make me so happy 😊
Getting ready for the next billionaire to be someone who made “don’t Java my Python” bumper stickers
My crazy mind though mathematicians just dropped a new Pi value 😭😂
I really hope 3.14 will be named Pithon
Missed a great opportunity to put the Twitter logo at 9:00 haha
I don't get why you would type hint kwargs that way instead of just type hinting the named kwargs after the "/, *," in the function signature, e.g. "def thing(self, /, *, kwarg1: int, kwarg2: str)"
I don't really understand the type params section. Can someone explain it?
The generic typing is the best!
I'm disappointed of the performance though. 3.11 vs 3.10 was a clear upgrade, while 3.12 barely changes, some programs run even slower.
great video. thank you
I can't decide if a smaller update like this with less new syntax is refreshing or boring
Best reviewer 🎉 :)
These changes all sound amazing.
Hey question how do u get this app like the same ui and stuff because i have to use windows terminal
looks like vscode to me
It would be awesome if you made a video about generic functions!
What's "type variable"?
They are basically copying TypeScript and changing the design a bit.
When can we get 3.14thon?
I was waiting for release really patiently, the new type syntax and f-string syntax are such amazing changes
Immortal objects may be useless to most projects, but for those, for which it is a problem, it may resolve a bit PITA.
Immortal objects would be a no-op in pypy. Refcounting is just inherently ugly if you care about performance at all. I wish that python3 had just dropped refcounting and used a proper GC in python.
2:03 I don't understand, what's wrong with that?
if you want to use same quotes, you have to escape them - seems very reasonable
why was this changed?
it feels more confuaing now that before
im strugle to find pyaudio for this version of python3.12
You mentionned setuptools. But not that *distutils* is completely removed from 3.12!
setuptool still monkey-patches in its special vendored version. But that means if you never imported setuptools, then imports of distutils will fail.
Does immortality mean I can get rid of all the Py_INCREF(Py_None) that are in my code because I'm returning None from a function?
Do you have to change the reference count for None anyway?
I wrote a C python module a few years ago. And for numbers, I was told to not change the reference count when returning an integer from 0-255, because those are special objects that are never deallocated. It may be the same for None.
@@Spencer-wc6ew If you use PyLong_FromLong(x), then you don't need to increment the reference count, because the object you've just received has had its count incremented. But you can definitely crash Python by calling a function that returns `Py_None` without incrementing it. Call sys.getrefcount(None), find None's refcount, and call your function that many times.
I assume the same holds true for the small integers, True, False, etc if you save the Python object in a global and repeatedly return it, rather than calling a getter.
I always set reference counts properly, for None as for everything else.
I once wrote a test program that did nothing but decrement the refcount for Py_None until it crashed. I got up to about 512 decrements or so, maybe slightly more.
I love python type hints, but not gonna lie, without rigorous type checking, I’m skeptical whether this kind of “advanced” typing is actually useful…
Do they rename Python to PIthon at Version 3.14?
3:14 you mention PEP 648 when it's PEP 684 :P
I really hope python will be called as Pie-thon when 3.14 is released
Can't wait for 3.14 - Python Pi 😊
Feeling like the error message update has been a long time coming, wonder if rust launching with readable traceback added to the push
Let's gooooo!!!
Thank you for the summary. It was super useful. The rating at the end of ever ysection feels kind of subjective and arbitrary.
kwargs typing is useful for framework-like code. Because it is not obvious which parameters are accepted by **kwargs.
3:13
its PEP six- eighty-four, not fourty-eight!
Batched finally here! I no longer have to produce weird magic tricks with iterator multiplication
Python 3.7? Was that the one that made me continue using 3.6 well into its "security updates only" stage?
For me the new Generic syntax and @override is 10/10. The generic syntax would have made it easier for me to read though!
You are aware that anything in Python is a generic? "def foo(bar)" means bar is generic (as is the return value). It's not untyped either, because of "object" being the base for all Python values? There's only limited amount of usefulness in something like "[X where Parent]" as documentation.
I think the C++ syntax was stupid. And it still doesn’t let C++ (or Java, for that matter) expression Python concepts like function factories and class factories.
@@FrederikSchumacher Yes. Python is too generic, which doesn't allow to express the intent of the programmer. The generic specification is just a hint for IDEs to check the sanity of the code. For example, a class of type T, has methods that adhere to the same type T.
Hmm... A video on Python 3.12 with a slightly over 12-minute runtime :D
Can't wait for 3.14 - then we'll have Piithon
Will it be renamed to pithon in 3.14?
Wow, python has changed a lot. I learned 3.1 and haven't had a reason to keep in the last few years
3:04 Having recently converted an application away from WSGI to ASGI instead, I could care less about things that might improve WSGI behaviour.
WSGI is Python’s answer to PHP. ASGI goes beyond that, to let Python do things (like handle WebSocket connections) that PHP cannot.
Yo the next version is 3.14 thats dope
The new generic type syntax looks like GO. and i love it actually
What about Linux perf support? I'd love to see a good example of it
Python 3.7 is now deprecated, huh? Feels weird, since that's the version I started coding on. But ofc it needed to go eventually. Other than it having a special place in my heart, there's no point in holding onto it
Can't wait for Python v 3.14159 running on Raspberry pi.
Just two versions left
can someone explain to me how generics work? I've re-watched the entire 2 mins about 3 times and I don't quite understand what's going on. it really triggers me a big impostor syndrome because I work as a python dev, yet I've never dealt with this
I see generics get used a lot in languages like Java/Kotlin, C++, Typescript, and probably others. The idea is that the "Type" of the parameter is determined when the object is actually implemented/created. So when you're creating the structure of the object, you can't define what the type is, but you want to tell the type checker that "whatever is used during implementation, that type is also used (wherever you want it in your logic)" - this most often gets used when you're dealing with custom types, not really the built-ins. This allows you to build tools that accept any arbitrary types, but still tell the type checker that the type matters and reference it elsewhere.
A basic abstract example might be: you want to track what movies, tv shows, and books you have in your "Library" - one of your methods on the library allows you to easily link sequels and prequels. In this case, you don't really care what the type of the object is (be it a movie or book or any other entertainment concept), but you want to ensure that whatever you link as the sequel is the same type. So you define the generic (commonly specified as "T" for "Type") and then you can now reference that in the rest of your code by saying "The parameter is a generic T and the sequel is the same type T"
More commonly I see it used in situations where you are building packages or tooling, so your function/class can be typed even if you never know what might be used for it, all you know is that the type must match elsewhere in the logic.
from my understanding, in short, when you use generics with a function, whatever type you use as input for it, it also outputs that type. i guess it's useful for functions that are meant to process objects that behave similarly and have implemented similar methods or dunders.
in the example in the video at 0:37, using generics in that function makes it clear that whatever numeric objects (or any object for that matter. it may also be objects like datetime where you can compare them) are used as input, the function will output object of that type, marking that specific type with 'T' , with a note that that object implements the greater than, less than, and equal dunders to allow comparison.
To me the easiest way to understand generics is to see how useful they are when typing lists, and more generally "container" classes.
If you have to write your own list implementation, and you want to type your code, you'll soon face an impossible question to answer: what type do you use for the elements of your list?
The answer is, well, you can't know. You can have lists of ints, of booleans, strings, etc. Clearly, the type of the elements of the list depends on each use that will be made of it -- therefore you have to tell the type checker that you have data whose type will change depending on how it is used. This is what type parameters are for, they're placeholders for the type checker to fill whenever your class is used.
Lists in python are generic. For example we can have a list of integers, or a list of strings, unfortunately, we can have a list of ints and strings ([1, "1"] is valid), which is often not what we actually want. To solve this, languages introduce a generic type T, that can be anything, int, string, double, myType, etc, but importantly, it can only be of that type only, not multiple types. So a generic list of type T[ ] can admit something like: [1, 2] and ["1", "2"] but not [1, "2"] .
Back in the day, at the dawn of object orientation, we'd call this parametric polymorphism@@dave_jones