@terdon For that purpose, you can also use a CRC. It depends on what the result of a collision is. Anyway for an n-bit checksum of any kind, collision probability between two random inputs is 2^(n/2). The birthday paradox gives probability of any collisions between m distinct inputs.

But unless the inputs are specially designed to cause a collision, you can treat MD5 as any other checksum with a 128-bit output, and SHA-256 as one with a 256-bit output. Note that for CRCs, you are guaranteed to detect any differences up to "burst size" that aren't a multiple of the CRC polynomial (which is good for detecting random errors).

I despise the birthday paradox

The birthday paradox is a paradox in itself

or at least the way that most people quote it

In cryptography, it's usually quoted correctly. :P

at my last job we used to store redundant copies of data on nodes called "shrimp" and so if two shrimp were down in a region they would say "there is a high likelyhood of them storing copies of the same data because of the birthday paradox"

but there is an even higher likelyhood they didn't share the same data

Well with a 128-bit hash, there's virtually no chance they'd have any duplicates.

in a group of 23 people there is a 50% chance that two of them share a birthday, but the chances that two of them chosen at random share a birthday is significantly lower

I guess if you combine the birthday paradox and murphy's law things could go sideways

Sure. That's the difference between a preimage and a collision.

If you have a person with a specific birthday and try to find someone else with the same birthday, it's much less likely than if you have multiple people with random birthdays.

But 365 << 2^128 for sure!

365.25

nevermind me, I'm all shimmed up

365.2425

According to Wikipedia. Just skimming.

But it's still << 2^128!

github.blog/…

This is one of the many reasons I don't have polkit installed on my system.

> CVE-2021-3560: local privilege escalation using polkit_system_bus_name_get_creds_sync()

9 year old user->root bug. :P

Jokes on them, all my users are privileged anyway

heh

@forest Unfortunately, it's not up to me. I mean, I am not implementing this, only using the existing compression tool which provides the md5sum checking itself. Sure, I could do it manually and just checksum the original file and store that and then compress, decompress and check manually but that would be considerably more complicated.

don't think those characters really add any value to the title

that's-a me!

@JeffSchaller I know exactly how I feel about it. grumble, grumble, millennials, grumble

:P

7 bits of ASCII were enough for me, they're enough for anybody!

heh, reminds me of;

img.newspapers.com/img/…

hah! fair :)

back in my day ascii was just a nibble

Ah, damn:

> While it’s certainly possible that someone, somewhere, at sometime uttered the phrase “If English was good enough for Jesus, it’s good enough for us” in earnest, this quote is most often used to paint politicians for being, perhaps, more ignorant of history and religion than they actually are.

snopes.com/fact-check/marjorie-greene-jesus-english

alternative facts

I'm really disappointed. I've always loved that story. :(

supermicro has a tool called supermicro update manager (sum) and hp has a tool called smart update manager (sum) (╯°□°)╯︵ ┻━┻

@jesse_b ... and IBM has suma (not the official link; a question that came up recently)

rabble rabble rabble

@JeffSchaller If the emoji or symbol does not add anything important to the actual title, delete that symbol. If it does, keep it. Example of good use: unix.stackexchange.com/q/648352/116858

@JeffSchaller "I wanted to parse HTML, so I thought I'd use regular expressions. Now I have two </p>roblems</body></html>"