Jokes aside, the one thing that really impresses me about ChatGPt when it comes to coding are its RegEx capabilities. I once needed a Lua Pattern (less capable version of RegEx without assertions or repeating words) to extract quoted parts from a string with the limitation that escaped quotes need to be ignored. It came up with a three step process where it first replaces every instance of escaped quotes in the string, then extracts the actual quoted substrings and then restores the original formating by rereplacing the temporary placeholder with escaped quotes again. I thought surely there must be a simpler way. I spent 2 hours reading the docs and fiddling around with it and only came up with uglier, slower solutions. That was the first time I felt that I could've really saved time using ChatGPT without sacrificing code quality.

why my aws bill 1.4m

ZAMN unintelligible regex 😍

Insanely plausible

Was literally gonna comment this LOL! I truly think that's what happened too.

You could write a regex for that

My brain almost didn't parse this comment

@@stacklysm must be an LL(1) comment

A sequence of catastrophic consequences.

it's a recursive catastrophe

Is he allowed to say that? lol

he did react to kevin once

I mean if you actually watched this video you will see that he starts watching Kevin Fang's video around 11 minutes in.

kevin fangs deez nutz

@@andrewcathcart You're right he watches some of it. I wasn't going to rewatch something I know.

Shit straight from Los Alamos pencil and paper days

A decade after Ken Thompson unleashed this unholy language into Unix, the e-mail server was born and it speaks only it's own Turing-complete version of regex. Sendmail will drive anyone into madness on the first meet, beware!

They also warned us: "Many a young programmer with a problem thinks, 'I know! I'll use a RegEx!' Now he has two problems." -JWZ

True, and using `contains()` wouldn't solve the problem anyway; the regex was much more involved than just `/=/`. People really must be regex-blind if they can't see that...

Funny how regex is basically the standard way of doing lexer step in a parser.

There are 2 cases where I've used regex: in-place text substitution and a personal long shell script that parses itself. I don't dare run regex while I sleep. Regex always execute in front of me.

I think the main problem here is that the person didn't understand the consequences of using .* and trying to match a character after that. To achieve the desired result, a negated set would've been perfect for the job. The problem is the readability.

@@theodorealenas3171 I've seen a thread in production that has spent more than a day running a very simple regex on an unexpectedly large input. The regex has then been replaced with a string contains and some change in the code. If it's not running in front of me, at least I must be confident that it's always run on text limited in length

@@CottidaeSEA I think the main problem is really that the person used a regex matching algorithm written by colossally incompetent fools. The entire _reason_ to use regular expression grammars is they're parseable by DFA, that's literally the whole point. Why on _earth_ is it backtracking?

How is that better? Oh so like [^=]*= got it

even better, make it a possessive quantifier: [^=]*+=

really, all you need is a properly implemented and optimized regex engine. It's not exactly cutting edge, finite automata theory is about 70 years old. Of course when computers are "fast enough" and neither the one implementing the regex engine nor the one using it cares to fully optimize their shit, desaster entails.

It won't be added to the capture groups but it can still be part of the outer match

exactly this

Or [^=]*(?:=.*) if you don't have non-greedy/minimal/lazy quantifiers.

Yes, and the regex in question did more than just checking whether a string contains "=". Prime's reaction... this comment... I expected better. Sad!

The words “use contains equals” is not the same phrase as “this regex was only checking for an equal sign” You have wrongly concluded that Prime meant the latter when he said the former, but you have failed to realize there is a better explanation for why Prime said “use contains equals” This better explanation appears more than once in the video. First, it appears before he says “use contains equals” you can hear him say “use contains” more than once while highlighting different parts of the text that talk about different parts of what the refer does. That’s because you can write a “contains” line for each part of the reflex statement that is checking to see if the string contains many different things and get the same result as using the regex. Later, we see a diagram of what was being matched in the regex. Again, Prime highlights the individual elements that are listed and says you can “use contains” for each thing being checked for besides the equal sign. Does you understand what I’m talking about here?

you just have to semi-die inside and become a semi-ghost

With a lot of yoga or prayers

I force every regex to have a shitload of tests, and then reviewers when they try to see what it could possibly do add more test cases. those test cases end up being better documentation than any comment or whatever.

@@gagiotter4114 Yes, tests are assumed here. But tests don't make the intent clear. You can test something fine but not know what exactly the author intends to match.

​@@trapexittests definitely clarify what is intended to be matched and what isn't

Test only captures what you know. Test cannot assume what you don't know. It's more an assurance of the past implementation upon the future update. The worse part of regex is that it usually contains a lot of unknown unknowns that many pretend they know. Then the CPU starts to burn. My 2 cents: people who said test is enough never understand the purpose of test nor TDD. Documentation, review to double check if you truly need a regex (95% of the times you don't) is always the way to go.

from my recent bias experience, people who used regex extensively were usually juniors or college kids who obsessed to the fancy one line solution. or just direct copy paste answer from stack overflow or GPT. For their simplest case it works, but always fails in some edge cases nobody thinks of. The solution is the easiest copy-paste without requiring them to read through huge chunk of codes like other solutions (those are actually the one they should copy&paste). Often lead to issue later.

Only if you use a DFA-based regex implementation, which CloudFlare before the outage didn't. Also, contains, especially when looking for a single character, is much faster than any DFA, as it can use SIMD.

If this had been compiled to a DFA there would have been no issue - it would have been deterministic, so not needed any backtracking.

I'd rather do it slower but reliable.

@@chepossofare In this case, its both slower and less reliable - which I can't imagine anyone favouring.

@@vytah Do you think that their regex was only looking for a single character?

Yeah, I was just about to say that. I mean, what's the alternative? Starting the lowest priority at 1 (or 0) and then your highest priority would be, what, INT_MAX? xD I think every priority system I can think of works this way: DEFCON, protection rings, etc.; even just saying "it's my top priority," like, where do you put that in a list? At the beginning, number 1 spot of a list, not all the way at the bottom, right? "My top five movies! My favorite, at number 5..." Nobody would do that

if you put your priorities in an array, the 0th index would be the first priority.

*Julia would like to speak with you for a moment*

I mean you could see priority negative infinity as the highest priority considering you should have done that before all of the negative number priorities and positive number priorities so it isn't entirely unambiguous.

@@christopheriman4921 ok propose that to your boss xd

NFA specifically I think. If this is the cloud flare issue I'm thinking of.

As someone who learnt Finite Automata just weeks ago, this is the 1st thing that came to mind.

@@0xCAFEF00D Yep, NFA. Although I’m pretty sure it can be turned into DFA.

@@0xCAFEF00D nfas would be more or less fine in this situation (they have linear complexity), the regexp engine they used did *not* use an automaton, but a simple recursive backtracking algorithm (i.e. for Kleene stars consume as much characters as possible, then match one less on each failure, for alternatives just try them left to right until they succeed; this has exponential complexity for obvious reasons)

@@atijohn8135 NFAs use backtracking and have exponential complexity. If you first convert it to DFAs though then it's fine.

That's JWZ's quote that is at least 15 years old:.

@@another212shadow didn’t say we invented it, just that we had it. Btw, have no idea who JWZ is, sorry

Jamie Zawinski It was his .sig file and he says he took it off of an older sed joke. en.wikipedia.org/wiki/Jamie_Zawinski

I love regex, but yeah if you can split the string on a known delimeter, i find that is usually easier to work with.

What would you use otherwise?

@@beatboy6690 at this level of complexity, you should probably write your own parser by hand.

​@@sack-shawwhich will be 1000X more characters with possible errors.

I'd rather write a bunch of "contains", regexps are a viable option imho only if your output is insanely random.

@@sack-shaw +1 for the parser, which is testable and reliable.

Finally someone that knows how to use ?. I was searching the comments for this. I use it all the time.

This guy coming out with a generic ass ChatGPT comment 🤖

Where is your Regex license?

so, what was the bet he lost?

@@JonnyRobbie I don't even know

Bro got the #00FF00D7 hair tint

blue hair

​@@JonnyRobbie he basically did a sub goal

Yup, and also use ? to indicate not to be greedy.

Julia has a ReadableRegex library that lets you write `maybe(["-", "+"]) * maybe(zero_or_more(DIGIT) * ".") * one_or_more(DIGIT)` to match numbers, including signs and decimal points. It then auto converts that into a "normal" regex you can use anywhere.

yes, there are regex toolchains

Ya, the second problem is figuring out what to do with all of the free time that I got from solving the first problem with such great speed.

And only foreplay

Same. I don't know what's wrong with these people. I don't know how they can be so proud of their ignorance.