4 ms·

> Most compilers make clever code silly. You can write a 15 line hard to read pipe or method chain, or a 25 line double for loop with the same runtime character

> Most compilers make clever code silly. You can write a 15 line hard to read pipe or method chain, or a 25 line double for loop with the same runtime characteristics. The latter is always better to maintain, while the former is always more clever. Weirdly, in rust there's lots of cases where the compiler will generate better assembly if you give it map/filter/fold than it will if you give it a for loop. The reasons are weird - like, its easier for the compiler to know it doesn't need bounds checks for list.iter(), and to work around some integer overflow errors while iterating through ranges and things like that. The difference usually doesn't matter in practice, but I still find myself thinking about it in performance critical sections.

cookieperson 3y ago
Ignoring rust or any language specific thoughts... Sure, but there are many cases where it's still better to have slightly slower code at the expense of readability/extendability/maintenance. Even in HPC applications. I know I know some very smart people are about to throw their sharpened axe at me. But in my experience very rarely does someone truly need to sacrifice a bounds check to deliver their product or save meaningful amounts of $ in prod. Not saying all loops must be bound checked, that's a dumb hill to die on. But I've seen devs hyperfocus on hypothetical minimal gains that get blown away two days after the code lands, or a minor requirement changes. The challenge with DO NOT TOUCH this code comments that intense performance tuning leads too(figuratively and literally) is it makes people who don't understand the code build bizarre(and often slow) monuments all around it that don't usually need to exist and deter from the minor gains won over a cute optimization. I've seen this happen ALOT. Of course there are times where you should opinionate your code to better serve the machine. But my big thing is, it's rarer then the overly complex design decisions favoring it are and compilers are getting better every day.

cookieperson 3y ago

Ignoring rust or any language specific thoughts... Sure, but there are many cases where it's still better to have slightly slower code at the expense of readability/extendability/maintenance. Even in HPC applications. I know I know some very smart people are about to throw their sharpened axe at me. But in my experience very rarely does someone truly need to sacrifice a bounds check to deliver their product or save meaningful amounts of $ in prod. Not saying all loops must be bound checked, that's a dumb hill to die on. But I've seen devs hyperfocus on hypothetical minimal gains that get blown away two days after the code lands, or a minor requirement changes. The challenge with DO NOT TOUCH this code comments that intense performance tuning leads too(figuratively and literally) is it makes people who don't understand the code build bizarre(and often slow) monuments all around it that don't usually need to exist and deter from the minor gains won over a cute optimization. I've seen this happen ALOT. Of course there are times where you should opinionate your code to better serve the machine. But my big thing is, it's rarer then the overly complex design decisions favoring it are and compilers are getting better every day.