[image] [ANN] First Release of ppx_pyformat 0.1.1 Community Using ^ for concatenation for each elements looks inefficient. Generating a call to String.concat "" should be more efficient and quite comparable in term of complexity for the code generator. I recently rece…

Let’s examine this code: let align_left w c s = let len = String.length s in if len >= w then s else s ^ String.make (w - len) c (* 1: 2 allocations + 2 copies *) The line marked 1 has 2 allocations: one for the new string, and one for the concatenation, and 2 copies: again, one for …

[image] bn-d: When the complier sees a chain of ^ operation, it will be automatically compiled to String.concat or something else similar. This would be a really bad thing to do, I think. There are many equations between different expressions, and starting to implement this sort of stuff is …

[image] Chet_Murthy: OCaml (and ML generally) has a history of being a language where you can look at what you wrote and know what it’s going to do. That means that whey you write a ^ b ^ c ^ d you’re expecting it to be three binary-concats, and left-to-right. You defeated your point as soon …

[image] bluddy: It’s also worth considering the Buffer module, which was created for exactly these kinds of use-cases. Buffer will force an additional allocation+copy on you at the end, because it doesn’t give you access to its internal bytes. So if you’re counting allocations, just stay away…

[image] c-cube: Buffer will force an additional allocation+copy on you at the end, because it doesn’t give you access to its internal bytes . So if you’re counting allocations, just stay away from it. Buffer is still useful when you want to build a string gradually but don’t want to compute h…

That’s true, that’s the use case. If you’re going to allocate a string eventually it’s fine to use Buffer.

Thanks! That is an excellent explanation! I totally overlooked the extra allocation and copy caused by Bytes.to_string in the second implementation. It is also good to know all the unsafe functions are actually available through the Stdlib.Bytes. I compared the performance of the function with bo…

Best practice for performant string operation?

Learning

mseri February 17, 2022, 9:46am 5

Tangentially related, have a look also at this thread for a discussion and benchmarks on the subject

Topic		Replies	Views
What's the fastest way to concatenate strings? Learning	14	10751	October 30, 2018
Could we move string and bytes to sliced types? Ecosystem language	27	1038	December 31, 2024
Should we have a `String.rev` in `Stdlib`? Ecosystem stdlib	36	4054	January 25, 2022
Optimizing small vector operations Learning	35	2013	April 4, 2023
Learnings from benchmarking OxCaml-flavoured SWAR memchr Learning c , performance , benchmark , rust , oxcaml	9	496	August 4, 2026

Best practice for performant string operation?

Related topics