SDS VS Git

Let me reframe this. What we're saying to do is stop using C string manipulation such as strcat, strcpy, etc. Particularly, I'm saying simply don't use C-style null terminated strings until you actually go to call a C ABI interface where it is necessary.

The argument against this is that you might call something that already does this. Yes, sure, that IS true, but what this betrays is the fact that you have to deal with that regardless of whether or not you add additional error-prone C string manipulation code on top of having to worry about memory ownership, mutation, etc. when passing blobs of memory to "untrusted" APIs.

It's not about passing the buck. Passing a blob of memory to an API that might do horrible things not defined by an API contract is not safe if you do strcat to construct the string or you clone it out of an std::string or you marshal it from Go or Rust. It's about not creating a bigger mess than you already have.

Okay fine, but what if someone hates C++ and Rust and Go and Zig? No problem. There are a slew of options for C that can all handle safer, less error-prone string manipulation, including interoperability with null-terminated C strings. Like this one used in Redis:

https://github.com/antirez/sds

And on top of everything else, it's quite ergonomic, so it seems silly to not consider it.

This entire line of thinking deeply reminds me of Technology Connection's video The LED Traffic Light and the Danger of "But Sometimes!".

https://youtube.com/watch?v=GiYO1TObNz8

I think hypothetically you can construct some scenarios where not using C strings for string manipulation requires more care, but justifying error prone C string manipulation with "well, I might call something that might do something unreasonable" as if that isn't still your problem regardless of how you get there makes zero sense to me.

And besides, these hypothetical incorrect APIs would crash horrifically on the DS9K anyways.

Even better, use a string handling library. Personally I am a big fan of (sds)[https://github.com/antirez/sds] from the Redis creator. It's not even a dependancy you can just copy the .c and .h to your project.

One nice application is length-prefixed string literals to complement dynamic string libraries:

The better answer would be to add data types like SDS[0] to the standard library, and use them as ADTs (Abstract Data Types) [1].

Unfortunely WG14 has proven in 30 years of existence, that it isn't something that they care to fix, and while 3rd party solutions exist, without vocabulary types on the stardard library adoption will never take off.

[0] - https://github.com/antirez/sds

[1] - https://en.wikipedia.org/wiki/Abstract_data_type

With the woes of string.h being known, why not just use an alternative like https://github.com/antirez/sds ?

I’ve also been having a blast with C because writing C feels like being a god! But the biggest thing that I like about C is that the world is sort of written on it!

Just yesterday I needed to parse a JSON… found a bunch of libraries that do that and just picked one that I liked the API.

Please note that the discussion started with requirement for no dynamic allocation in critical code what virtually eliminates std::string. I agree that std::string code tends to be simpler but the main reason is that the standard C library sucks on strings. There are better alternatives like sds but they are ... not standard.

Another day, another post about a writing a bespoke string lib instead of using SDS

d) everything being a macro seems overkill for me (and possibly dangerous, see b)). Maybe implement more as static inline functions, see the sds header: https://github.com/antirez/sds/blob/master/sds.h (which does a similar thing with the header struct).

For example, you can use the C language with sds strings (see https://github.com/antirez/sds) if you want to have an easier time with string formatting and don't want to worry about using the famously unsafe string.h functions correctly. You'll still program in ISO C, but just not in the standard library. The same applies to pretty much all parts of the standard library, the only part unsurpassed is pretty much just printf and the math headers (math.h, fenv.h, tgmath.h, complex.h) imo, and the occasional call to exit. A good place to look for libraries if you want to go that route is the awesome-c collection: https://github.com/oz123/awesome-c

One way around this problem is to declare the container as a pointer to the element type and then store the container’s metadata, alongside its elements, in the heap block to which the pointer points. This approach is already used for dynamic arrays in several container libraries, most notably stb_ds and sds. They place the metadata before the elements and provide the user with a pointer to the elements themselves (this has the nice effect that users can use the [] operator to access elements).

Look at any PR/patch series that got merged into the Git project. https://github.com/git/git/

Any random one. Because those that did not meet the minimum criteria for a well-crafted history would not have passed review.

Boy, I can't find this either (but also, the kernel mailing list is _really_ difficult to search). I really remember Linus saying something like "it's not a real SCM, but maybe someone could build one on top of it someday" or something like that, but I cannot figure out how to find that.

You _can_ see, though, that in his first README, he refers to what he's building as not a "real SCM":

https://github.com/git/git/commit/e83c5163316f89bfbde7d9ab23...

Here is the direct link, as HN somehow removes the query string: https://github.com/git/git/commits?author=peff&since=2023-10...

I understand all that.

I'm saying, if you write a survey and one of the possible answers is "diff", but you don't clearly define what you mean by "diff", then don't be surprised if respondents use any reasonable definition that makes sense to them. Ask an ambiguous question, get a mishmash of answers.

The thing that Git uses for packfiles is called a "delta" by Git, but it's also reasonable to call it a "diff". After all, Git's delta algorithm is "greatly inspired by parts of LibXDiff from Davide Libenzi"[1]. Not LibXDelta but LibXDiff.

Yes, how Git stores blobs (using deltas) is orthogonal to how Git uses blobs. But while that orthogonality is useful for reasoning about Git, it's not wrong to think of a commit as the totality of what Git does, including that optimization. (Some people, when learning Git, stumble over the way it's described as storing full copies, think it's wasteful. For them to wrap their heads around Git, they have to understand that the optimization exists. Which makes sense because Git probably wouldn't be practical if it lacked that optimization.)

The reason I'm bringing all this up is, if you're trying to explain Git, which is what the original article is about, then it's very important to keep in mind that someone who is learning Git needs to know what you mean when you say "diff". Most people who already know Git would tend to gravitate toward the definition of "diff" that you're assuming (the thing that Git computes on the fly and never stores), but people who already know Git aren't the target audience when you're teaching Git.

---

[1] https://github.com/git/git/blob/master/diff-delta.c

Didn't Git have a new default merge strategy, `ort` https://github.com/git/git/blob/master/Documentation/RelNote... ?