Andrew Gallant

If we were to add something like this, it should probably start with something like, "**If** you'd like to define an MSRV, then ..."

The existing API guidelines don't really seem to have any particular goal of listing desirable permissive licenses. A few extremely popular ones are mentioned, but otherwise, I think the moniker...

> but it would be good if it listed more than just BSD-style licenses It does. The Apache-2.0 license is also mentioned and it is permissive. The guidelines say "permissive"...

I'm not sure I agree with promoting this to an API guideline at this point in time. Certainly, if we had _started_ with the current definition of the `std::error::Error` trait,...

I had to [remove the minimal version check from `regex`'s CI tests](https://github.com/rust-lang/regex/commit/8f9ca96a844b07e6b378573ec6fbac6cf9fe5430) because of `rand`. The `rand` crate continues to, for example, advertise support for `libc 0.2.0` even though `rand`...

@vks Sorry, but I'm not going act as rand's unofficial CI by submitting PRs whenever downstream detects that rand's minimal versions aren't correct. I've either just been removing the minimal...

Empirically, I believe you're right. And if so, it leads to a pretty huge simplification: #93. I haven't had time to think through why this is okay and why I...

See https://docs.rs/aho-corasick/0.7.18/aho_corasick/struct.AhoCorasick.html#resource-usage And: https://docs.rs/aho-corasick/0.7.18/aho_corasick/struct.AhoCorasick.html#method.heap_bytes Here is where the main construction algorithm happens: https://github.com/BurntSushi/aho-corasick/blob/4499d7fdb41c6279ea1367bccf3daca9cb06c36c/src/nfa.rs#L671 Patches that decrease memory usage are welcome, although it would be good to discuss them first. I...

OK, so my analysis above looks correct then? If there's a bug, it's that `heap_bytes` isn't accurate here. Indeed, it looks like it doesn't count the sparse state memory usage...

Using heaptrack, about 75MB comes from adding sparse states, and about 42MB comes from the sparse transitions themselves. That also seems in line with my analysis above.