One time I decided to learn Python. I bought books, registered for a class at a local university. After 2 classes I realized I already knew python. Because Python is just C++ for kids. And I have the maturity of a 7 year old.
No, they might grow resistant to it. Show them MY code Also, you can make them learn python, while simultaneously not dropping C++. Then, make them write a program (without compiling it and without copying from other sources) in python and in 10 minutes(after start) make them write the same one in C++(also without compiling it and without copying from other sources). Repeat until they're done. After that, compile both programs. Profit
Now the good part: in C++ you can improve the code so it runs much faster. In python you can do not much to improve unless you write a wrapper which calls C++.
Well actually you CAN improve interpreted languages perfomance a lot. In any algo there are optimizable bottlenecks. The difference is that you can push optimization farther if you are using language with memory control
1) IO operations take about, I dunno, like 1000 times as long as most other instructions, and are completely done by the OS. Which means that IF you managed to write the same code in both languages, you would get basically the same speed. 2) However you did not write the same code, the Python program (probably) reads larger chunks of the file at a time, and since reading 1 byte and reading 4kB takes approximately the same amount of time, it obviously takes less time 3) You are using completely different data structures. In Your Python program, you are using IMMUTABLE STRINGS, so they can safely avoid copying, and instead just reference the input data, while your C++ strings are MUTABLE, so they (likely) have to copy the data
To add to Point 1, IO operations are slow cause, if i remember correctly, they don’t include buffering so the program have to fetch data from the main memory instead of using a cache
@@pixelitedragon1720 is it? I think cache useage is transparent to all programs (OS included) (programs can't choose whether to use main memory or cache)
@@dysrhythmic percentage points is a thing but it doesn't apply in this context; it only applies when the values being compared are themselves percentages
C++ cin and cout streams are very slow by default because they have to wait and sync with the C standard streams. Add std::ios_base::sync_with_stdio(false); and run the test again
@@knowledgenews5343 there's a whole stack overflow page about the issues with it and I just think typing std:: before everything is a good habit anyway
you can't scare a c++ deeloper dude, in python you've read the whole chunk at once, but in case of C++ you're reading it line by line --- causing more input operations resulting in more time, in fact i'm impressed that c++ takes this less time for a million input operations more than python.
Yes, sys.stdin is buffered. The iterator iterates over the internal buffer line by line. The assignment does not have any effect. The bytecode-compiler will remove this line. If you want to test the speed of the languages, then you have to take IO away. Both tests require the same conditions, and it's not the case with the example code.
It has more flexibility with {} and ; Also you do not need the right amount of SPACES.. yes right SPACES! not TABS! It has to compile and tells you about lang mistakes before running program not in the middle of running. And I forgot. You define type of variable so you know what it is. In python you come bck to large old code and have no idea what type a variable is cD
The fact that a python programmer would think using I/O to test code performance is something reasonable, is exactly why this video makes the opposite point it's trying to.
@@computerscience3836 You sound like my college professor. He refuses to teach anything but Fortran 77 (in 2024). He doesn't understand that different languages have its places. Why the hell would I create a plot in C++ when I can scratch together a notebook and do it faster in matplotlib? Every language has its place. (except Java).
Python will often use a single memread to open and contain files, which allows for a speed-use of large files, but C++ would not optimize that unless specifically instructed! Fun video, thanks!
@@andrewwhiteman8892 Sure! In C++ the code is "memtemplate" or "memcpy", which are implemented through a memset operation in C-lib. All code in the C-lib environment is pretty outdated, but you can create a similar function in this way: global ptr* load-file func { # pseudo-code for loading a file from disk into memory, using chunks aligned with page and disk-block size findChunkSize(useCurrentDisk is true) store in localVar ChunkSize; # this needs to find the largest, common size that fits in both one page size, which is how many bytes are loaded by the operating system for a single-use, and one block size, which is how many bytes are read by the hard-disk in a single read operation readOneChunk(useSingleChunkIO is true) store in localVar SingleChunk; # would need the block-place of the file in question, as well as localVar ChunkSize, etc, which are often omitted in pseudo-code, as it is a guide only, but I thought to mention as I would write it in! ConfirmChunkIsRead() returns true or false, test; # sometimes the disk is too busy to read and will fail, etc. By pseudo-code rules, no exception handling is needed, but here would be one UseSingleChunkFollowOn(pass-in SingleChunk, useDigitalSideIO is true) store in localVarArray AllChunksOfFile; # the SingleChunk is already measured, and includes where it was found for easy read-on-- DigitalSideIO is a boolean true or false for whether to follow the chunks as an abstract "file" (which We want) or literally as blocks on disk (for simpler I/O) # all chunks are read, and now are stored for access outside of the same-func. UseGlobalAlloc(sizeof AllChunksOfFile, cache-memread pass-in AllChunksOfFile) as return } Python uses a smaller form of this same function to cache exactly 1 MB or so. The later versions of Python will attack the file in different ways to further improve system-side caching without having to test if there is a large memread cache on disk, in the operating system, or other ways. All of this code is in Python's "memread" and "system IO call-func" parts, in case you're further interested. Take care!
Haha good one. This is why python has been traditionally used for embedded, realtime, OS, games, heavy desktop apps, and writing other major tools and runtimes...oh wait
python is slow tortoise ; when it becomes larger your program becomes slow as tortoise ; maybe this dude has 256 gb ram and that's why python is fast ; go and test it in the slow pc with 2 or 4 gb ram and you will see
@@Phantom-lr6cs fun fact: you said to test it in a slow pc with 2 or 4gb ram, so you just pushed the blame on depending on how fast the pc is, also dude doesnt have 256gb no one has 256gb ram plugged to one pc
hot take: if you really wanna scare c++ programmes, you should ask them to use std::find on an array of some structs and then try to read the compile error, lmao.
Testing LANGUAGES speed by calling functions handled by OPERSTING SYSTEM is like comparing milk quality by comparing their price - may look connected, but sometimes can fool you
As someone who is a c++ programmer who also does a bit of python occasionally, sure, as others have pointed out, there's ways of making the c++ version much faster. But unless I really needed the optimization, for something simple like this, I'd probably use python.
How do you start as a programmer and what should be learnt to pass as a programmer myself? I only know std::cout, struggling with std::cin, and thought namespace is useless.
The reason here is that Python doesn't read and write the line into a string, it's only reading the new line characters from the stack and counting. C++ is reading and writing every line into the heap, this is vastly different and therefore the reason it runs slower.
@@aaarkidI just tried this code and the perfomance is worse than the one in the video, the best method I found was to read the file with fstream, it became insanely faster, and when we read the whole file with Python its performance keeps the same
i have done alot of coding in both cpp and python, and honestly, cpp feels alot more comfortable. one of my friends said this : "when i code in cpp, it feels like i m coding, but in python, it feels like writing english."
@@davidwuhrer6704 One can use Pycharm and pygame. But it'll be much harder since most game engines use C#. So, you've to do it without Advanced features which increase the difficulty level tremendously.
@@Sumirevins Akshually scripting in python does not sacrifice any features, quite the opposite, it makes them more accessible and adds new ones. Nor is scripting in python more difficult that writing C#. One of the major selling points of python is how easy it is to write something readable in it. And most game engines are written in C++. Only Microsofties use C#.
@@Raspredval1337 you think so? Do you mind explaining why is that the case (I think C++ is way harder, the amount of syntax and features are just insane)?
@@Raspredval1337 yeah especially when you should be doing c in another few more lifetimes, but that lifetime lifespan you're having is 50+ years < my lifetime lifespan of 2 lifetimes Xd
If anyones confused on why python is faster in this situation is due to the fact that python *is* actually faster than c++ when it comes to numbers and counting, any other situation c++ is faster
-O3 enables some aggressive optimisations and TRIES to vectorize it, but, since the user didn't specify the vectorization options available to the compiler, it didn't vectorize it. Just pass -msse or -mavx to the compiler and you'd see the difference. Plus std::iostreams are kinda slow, should've used the cstdio primitives.
It might be purely because iostrams are slow. I looked at the disassembly of some C code i wrote that is of the same function but uses read() directly, it did actually vectorize something - it doubled the instruction count and there is vector register manipulation everywhere, but this had almost no effect on performance compared to an unoptimized version; however both were as fast as python, as a speed about twice that of the optimized C++ version. i would assume the same happens to the C++ code if you remove iostreams, although i did not examine that for some reason.
C++ streams are slow because they are forced to wait to sync with C's stdio by default. There's a parameter in std::ios_base that you can change to disable sync with C stdio which makes C++'s streams much faster. Obviously they are still slower than raw stdio because streams internally make a bunch of checks to convert the data that you're inputting/outputting to and from the streams.
@@AlFredo-sx2yy just paste std::cin.tie(0); std::cout.tie(0); std::io_base::sync_with_stdio(false); before doing any IO and cin and cout should be same speed as scanf and printf (iirc cin and cout are even faster than scanf and prinf with this)
Tries to show language X is faster than language Y on an IO bound operation without any understanding of how the IO is being handled in either language. I'm sure everyone is quaking in their boots over it.
It's because there is a mutex to have thread safety with cin and cout in c++. You can disable it. And Python is monothreaded so there is no mutex needed.
it's probably because sys module is implemented with cpython, so you're essentially comparing c against c++ otherwise looping in python is extremely slow
C++ IO is slow by default because C++ streams are programmed to sync with C's IO. There's a parameter in std::ios_base that you can use to disable that sync and that would make any IO operation in C++ much faster.
By the time python reads the input buffered in memory, C++ has done 700 000 stdin read calls of arbitrary length. It's like reaching for a shelf in the same room and walking a few blocks to get a book.
@@thelemoid not even that, the fact alone that a std::string is being used is enough for me to be annoyed at this. std::string classes are mutable, while the strings in python are immutable (reminder that python is backed by c). trying to argue that python is faster than c++ while not even using the same data types. after rewriting the C++ code to read into a char buffer, i get pretty much the exact same time as python (which is exactly to be expected since python is also just doing c stuff under the hood) if i rewrite to use buffered input, i could definitely improve the speed by a lot more but im lazy right now lmao
The thing that scares me about python is not so much the performance (not everything needs to be lightning fast), but rather the lack of curly braces and static typing which can easily get messy and confusing if you're not careful
I've done this by accident when working on college project, my nerd friend lost his shit when he saw my absolute disaster of a code I've created. He literally screamed "NO WHAT ARE YOU DOING?!?"
Python isn't that much different from C++ in some ways. I have learned some Swift, C,C++, Rust, Python, Nim and probably more from the C/C++ family of language as well as several more from other families of languages. All the languages from the "C/Algol family" are much the same. They are all imperative languages with a similar syntax. Languages like Haskell and Prolog are totally different, they don't feel like your programming in the same way and you need to convert the programs from an imperative train of thought to a declarative logic. The biggest difference between Python and C++ is that Python is interpreted and doesn't have the duality of compile time vs runtime. I cannot comprehend how people can use dynamic languages for big projects, I get nervous just thinking about it. Python is great for engineers and data scientist and using jupyter notebook is an incredible tool where you can do things in a way that maybe wasn't feasible before. But actually creating application software in it sounds almost irresponsible unless your writing millions of unit tests(and some do).
depends on the project... python developers are cheaper, also it IS quicker to build something in python... then, it is also possible to design things wrong - and for c++ that could be costlier... no programming language prevents errors from happening, in C++ they can be harder to diagnose and fix than in python
The problem are IO operations which are not meant to be used on production by any means. In any real C++ application (Most probably a multimedia system) console outs and console ins are a big no no. In audio programming for example, there is a constant loop through really small slices which together makes frequencies and therefore sound. In the loop you should never use a cout or cin for debugging because it will damage completely the audio output.
Now try it with 1 billion lines. Python will take ~128000ms. And C++ will still take like 500ms. But why? Python is an INTERPRETED language, which means it runs code line-by-line. So by multiplying the amount of work, the speed will vary a lot, A LOT. C++ is a COMPILED language, which means it converts the code into machine code, which runs VERY fast. So by multiplying the amount of work, the speed won't change really.
my benchmarks: py3: ~0.490 c++: ~0.800 I/O operations are very long, that does not depend on the language, and the times should thus be about the same. However python has I/O code more optimized than this C program written in 5 seconds, using less I/O operations thus the faster time. Optimizinf the C++ code by ditching C++ because i can’t write it, and rewriting it in C reading chunks of 2048, i can get it down to 0.465, which beats python ever so slightly, probably due to load times or something.
That's the shitty part about C++. Reading file line by line is very common idea. To do it efficiently you need to do lots of steps. In fact in C++ I avoid them entirely and read whole file. And only ~5 years ago c++ got normal api for telling file sizes. No, rewiding file to the end, using tellg, rewinding to beginning is not normal.
@@ItsCOMMANDer_ That could be faster, but memory allocations are quite fast on modern systems, and by doing that you lose the ability to re-use the code for reading from files on unknown length. It, as almost always, depends on context
@@chri-k you can do something like a function, wich takes an FILE object, uses fseek to go to the end of the file, say at what character it is (e.G. 25 charcater is theast so the contwnt of the file is 25 bytes. (not including \0 ) long and then just allocatge memory for that amount and use fread to read the entire file.
my C++ brain: "aren't you just using unoptimized code?" my Python brain: "wtf, and here I thought C++ was faster" *My entire brain:* this is a warcrime and must be investigated
Haha nice one. Well on a serious note this is just simple print operation and a counter. When writing more complex things Python provides libraries which are generally slower than C++. If you implement the same thing in Python yourself you can get equal speed etc. But Python is preferred because of its easy syntax and support of a LOT of libraries which allow you to do anything rather than implementing them. (C++ have as well, but Python has more)
I love both languages. Python bc it's easy to make a project, and C++ bc I feel like a pro programmer when programming in it. But Im seeing that the purpose of this video actually worked, there are so many C++ devs scared in comments. Including me
To solve this riddle: Ptyhon has many library functions that are natively coded in C and quite optimized (using low level system calls and stuff) As long as you use those packages, you will probably beat most languages (because C is very fast, if optimized)
Delusional. Practically all abstraction except polymorphism are resolved at compile time. C has a very clean and understandable look but C++ is more powerful, it's just the truth. Try benching any code.
@@torshepherd9334 templates are such a nice convenience when it comes to writing code, but a major pain in the rear when it comes to reading it as an outsider. I'd rather not touch those things beyond aliasing composite types, and maybe one or two inliners.
For me, the weird thing is that it's a bit complicated and hard for me to know python but for some reason I understand c++.. I don't know why but that just happened Although I actually managed to make turtle python art only circles and lines but in C++ basically just made my custom header file so instead using cout, I use print.. (I like to see the world burn slowly and painfully)
Fun fact: Rust code will be slower by default in IO because IO in Rust are not buffered (you van do it but it's different behavior than other popular languages std lib).
Python uses hyper-optimized C libraries and C++ is an extension of C, so basically you compared an extension of C to a wrapper of C and added your personal flavour to both. A C++-programmer understands that and isn't scared... which is why you could upload that.
Comparing c++ and python in what python is good at, is a bad comparison unless you try it the other way around. Compare Python and c++ at what c++ is good at, and you will find that c++ is faster by a factor of about 150 times. Besides python is super messy and a breading ground for very bad code. Not scared.
C++ programmers seeing this be like: "I'm understand this joke but still here is my explanation for why both code aren't the same thing and how this C++ code can be 1532% faster"
