I'm honestly still shocked any of this happened.

Prior to SHA1 we had MD5, a decade earlier. MD5 collision attacks had already been widely documented and known. It was the most obvious thing on Earth that this would happen to SHA1 too. Apparently, Linus never realized there was a need for cryptographic security and that the hash was purely internal.

Here's what I honestly think was a factor. I think C programmers fell in love with the implementation that you could throw around a fixed hash record on the stack. It's incredibly efficient. But it's an efficiency that doesn't really matter because as soon as you read from or write to a disk or a network or even memory, any cost saving is completely gone.

More than a decade ago, some people wrote a Java implementation of git (jgit?) and despite all their optimizations, it was (IIRC) only half as fast as C git. It is of course because Java at the time had no concept of stack values for non-primitive types so couldn't compete. Personally, I was impressed: only half the speed? That's pretty good.

For something that's only 20 years old, the Git SHA1 assumption is some of the worst technical debt we have in the modern era.

Here's another thought: when people make a lot of these programs, they often make the mistake of not separating the program version and the network protocol (or just the external API). So you end up with brittle client-server implementations where you have to upgrade both the client and the server at the same time because they lack a network abstraction.

The other end of the spectrum is video streaming where you have codex, container formats, transport protocols and so on.

What a mess.