So I am a big fan of your work. It is really admirable what you've done, this latest push recently. Having the stabler `regex` vs more experimental/open-ended/user-assembly-required `regex-automata` is fantastic! I wish many more libraries did this, and I will happily spread around your blog post in arguing that specific ones should do so.
I wouldn't describe the difference between our views here as optimism vs pessimism. I like Rust decently enough to use it a fair amount and contribute to it somewhat. But I'm hungry for something more. It's fine if you aren't.
> But the thing is, coupling is generally how you make things fast, and making something more general usually involves de-coupling. So they are actually and usually exclusionary goals. But it takes a lot of domain knowledge to actually see this.
The biggest difference is that I don't believe this is true. The interface may be extremely gnarly and hard to express, but I do believe the interface can always in principle be sliced. It may not be practical (the interface is far more complicated than the thing before cutting) but it is still in principle possible.
---------------------
Let me put it this way, imagine some alternative world Unicode which was pathological in all the same ways (case, variable width encodings, all the other things I am forgetting). Imagine if someone forked your work to handle that. Clearly there would be lots of small things different --- it would be hard to abstract over --- but the broad brushstrokes would be the same?
I am not interested in supporting things which are not like Unicode, agree stray to far towards large alphabets etc. and all the algorithms one would want would be different anyways. But supporting things like Unicode or strictly easier than Unicode should be doable in some language, I think.
> The biggest difference is that I don't believe this is true. The interface may be extremely gnarly and hard to express, but I do believe the interface can always in principle be sliced. It may not be practical (the interface is far more complicated than the thing before cutting) but it is still in principle possible.
Well "practicality" matters. And I used weasel words such as "generally."
It's one thing to make concrete suggestions about how to actually achieve the thing you're looking for. But when you're way out in right field picking daisies, it just comes across (to me) as unproductive.
> but the broad brushstrokes would be the same?
This is basically impossible to answer. I could just as well say that the "broad brushstrokes" are the same between the `regex` and `regex-lite` crates.
I wouldn't describe the difference between our views here as optimism vs pessimism. I like Rust decently enough to use it a fair amount and contribute to it somewhat. But I'm hungry for something more. It's fine if you aren't.
> But the thing is, coupling is generally how you make things fast, and making something more general usually involves de-coupling. So they are actually and usually exclusionary goals. But it takes a lot of domain knowledge to actually see this.
The biggest difference is that I don't believe this is true. The interface may be extremely gnarly and hard to express, but I do believe the interface can always in principle be sliced. It may not be practical (the interface is far more complicated than the thing before cutting) but it is still in principle possible.
---------------------
Let me put it this way, imagine some alternative world Unicode which was pathological in all the same ways (case, variable width encodings, all the other things I am forgetting). Imagine if someone forked your work to handle that. Clearly there would be lots of small things different --- it would be hard to abstract over --- but the broad brushstrokes would be the same?
I am not interested in supporting things which are not like Unicode, agree stray to far towards large alphabets etc. and all the algorithms one would want would be different anyways. But supporting things like Unicode or strictly easier than Unicode should be doable in some language, I think.