what is going on here
Timeline
Post
Remote status
Context
7
@sickburnbro LLMs are really good at turning working code into other working code, especially with sufficient testing which with working code can include feeding it the same inputs to see if it gets the same outputs. This is actually significantly easier than creating code from scratch.
hardest hit: proprietary blobs (the objdump'd binary is also "working code"), legacy sunk-cost systems, and companies that fired all the devs 10 ten years ago and have been just milking the cow until it dies off, since.
hardest hit: proprietary blobs (the objdump'd binary is also "working code"), legacy sunk-cost systems, and companies that fired all the devs 10 ten years ago and have been just milking the cow until it dies off, since.
@apropos but why is this even something people care about? how much of that code still exists?
@sickburnbro COBOL's attention-getting but most of software is like this. There's are also huge Java codebases and legacy Ruby-on-Rails systems. There are companies running 20-year-old Perl programs with 20-year-old backdoors. There's incredible amounts of investment in Python which is all much slower and much harder to maintain because it's in Python. There's JS and C++ where even the people advancing the language can barely contain their contempt when talking about it. Banks have paleolithic software that can't be touched because any upgrade would need to replicate all of the bugs exactly.
Everyone's working with some software he wish he could rewrite.
Everyone's working with some software he wish he could rewrite.
@apropos it feels like the people that want to rewrite the shit are just talking up new languages which then have their own bugs? It seems pretty stupid all around
@sickburnbro there's a lot of that, yeah.
>new guy joins team
>new guy: woah, this is in Java? Java sucks!
>new guy: let me rewrite it in Haskell. I'm a super genius so it'll go great.
>boss: really? OK.
fast forward
>new guy#2
>new guy#3
>new guy#4
fast forward, it gets harder to argue the economic case of letting part of the team piss around for two years without doing anything productive.
>new guy joins team
>new guy: woah, this is in Java? Java sucks!
>new guy: let me rewrite it in Haskell. I'm a super genius so it'll go great.
>boss: really? OK.
fast forward
>new guy#2
>new guy#3
>new guy#4
fast forward, it gets harder to argue the economic case of letting part of the team piss around for two years without doing anything productive.
@apropos like there was the cloudflare failure where they had something they rewrote in .. rust I think? and it was supposed to be "oh this programming language is perfect" or some shit
@sickburnbro yeah it was Rust. Rust's error-handling is such a pain in the ass that the language makes it more convenient to lie in the function signature and say that the code is written to be infallible. Without any of the self-aware architecture that Erlang would have.
Replies
0
No replies yet.