I like this bit at the end:
As a side note, the program is amazingly performant. For small numbers the results are instantaneous and for the large number close to the 2^32 limit the result is still returned in around 10 seconds.
Really makes you question your sanity when optimizing jumps in code without benchmarks.
For a long time I’ve been of the opinion that you should only ever optimize for the next
suckercolleague who might need to read and edit your code. If you ever optimize for speed, it needs to be done with massive benchmarking / profiling support to ensure that the changes you make are worth it. This is especially true with modern compilers / interpreters that try to use clever techniques to optimize your code either on the fly, or before making the executable.The first rule of optimization: Don’t do it
The second rule of optimization: Don’t do it yet (experts only)I’m absolutely on-board …in application code.
I do feel like it’s good, though, when libraries optimize. Ideally, they don’t have much else to do than one thing really well anyways.
And with how many libraries modern applications pull in, you do eventually notice whether you’re in the Python ecosystem, where most libraries don’t care, or in the Rust ecosystem, where many libraries definitely overdo it. Because well, they also kind of don’t overdo it, since as a user of the library, you don’t see any of it, except the culmulative performance benefits.
Libraries are also written and maintained by humans.
It’s fine to optimize if you can truly justify it, but that’s going to be even harder in libraries that are going to be used on multiple different architectures, etc.
I’m still mad he didn’t use the size of the number to tell the system which block to read first. I feel like that would be a great use of division or maybe modulus?
I just like how he used “% 2” in the Python code he used to generate the C++ code.
Now we just need to someone to package it and upload it to NPM.
What’s another 40 gb of node_modules anyway
Could be easily made 50% space saving by only iffin all odds and return even on else. Maybe one if before to handle overflow to avoid wrong even if over the last if.
Well yeah, if you allow cheating!
Yeah but then ALL even numbers would be slow to compute because you would have to chain through every odd before you know that
2is even.Depends on the expected distribution of input values
Heuristic: keep it until 512, afterwards powers of 2, and numbers like 1000, 2000,…, 10000, 20000,… (regex: [0-9]000+)
Let’s be real though, everything is IF statements all the way down
There’s not a single thing in this universe that cannot be accomplished with enough IF statements… as long as you’ve got infinite time to wait
…you mean IF you’ve got infinite time to wait?
…you mean if you’ve got IFinite time to wait?
What if there was an unintentional infinite loop in your code. You could be waiting for infinite time only to learn the code had a bug. D:
The problem with if is the answer comes from user. There’s no mathematical reason or scientific explanation, only programmer who thinks the answer should include the subject.
True…
But even on a more metaphorical level, every single thing that has or will happen in this universe, down to even the smallest quantum fluctuations could be encapsulated into IF statements as long as you had enough of them.
This is poetry.
My favourite part is that he uses the modulo operator in his Python script to generate the C code.
@programming_horror , anyone? We have an in production version of this used in Wikipedia
Instance friendly link: !programming_horror@programming.dev
Thank you!
I’m pretty sure that my link also works regardless of instance. It works when I visit it on your instance.
Andreas is a maniac
I’m not a good reader - I skim most articles and often miss most of the meaning. I read, and enjoyed, every word of that!
Thanks, I totally would’ve skipped it without this comment.
Well i hate this:
PS > .\program.exe 0
THIS:
.\
This is peak YandereDev.
I’ve lost some of my sanity reading this !
It works don’t it
Those are rookie numbers. Professionals came up with the nested logic monstrosity that is the JSON-LD specification:
https://www.w3.org/TR/json-ld11-api/#context-processing-algorithm
Looks through the algorithm bits in the various sections. How would you implement that? The answer is invariably by copying the highly nested statements of the spec directly into your language. Maybe there’s a better way, but you’d have to understand all that nested logic first, and you’d be exhausted at that point and just want to move on.
This is why every programmer needs to understand the modulo operator.
Or bitwise AND.
This is what I prefer too! I also some times prefer to use bitshift when it comes to division or multiplication of power of 2.
I would divide by two (floating point) and check the fractional part.
turns out that 2^53 + 1 is an even number
The article only covers unsigned 32-bit numbers, so floating point division would be fine.
I first saw this joke back in the days of 8-bit home microcomputers. Of course then it only needed 256 lines of code, and took up about 8k of your precious, precious RAM.
