minor improvements everywhere

[CAD97]

Because none of this was quite specific enough to break into its own PR

• Update translation docs
• Officially compiler_error!-forbid cfg(target_pointer_width = "16")
• TextRange::before (alt options: TextRange::to, TextRange::up_to, TextRangeTo, TextSize::prefix)
• TextRange::after (alt options: TextSize::suffix)
• TextRange::offset (alt options: TextRange::add, Add::add, TextRange::reanchor)
• TextSize::one (alt options: )
• TextSize::ONE (alt options: TextSize::ASCII
• #[inline] capability for a bunch of things

This PR is mainly just to get opinions on all of these little things, then I'll re-force-push with a commit just with those.

[CAD97]

I consider TextRange::after good enough to close #15.

[matklad]

so we currently have panic at runtime

Oh indeed, I’ve missed that....

…

On Sunday, 15 March 2020, Christopher Durham ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In src/lib.rs <#18 (comment)> : > @@ -13,3 +13,6 @@ mod traits; mod serde_impls; pub use crate::{range::TextRange, size::TextSize, traits::TextSized}; + +#[cfg(target_pointer_width = "16")] +compile_error!("text-size assumes usize >= u32 and does not work on 16-bit targets"); When the goal is to compile on multiple cfgs, I definitely agree that specifying the behavior without cfg is better, if possible. But for cases like this where some cfg isn't going to work at all, I don't agree that the array shenanigans are better. To begin with, what we have on master is "cfn_assert! <https://crates.io/crates/const_fn_assert>" and not "const_assert! <https://crates.io/crates/static_assertions>", so we currently have panic at runtime (or error-by-default lint if it gets MIR const folded) behavior, rather than guaranteed compile error. More importantly, though, is that #[cfg] compile_error! *exists* so that cases like this can give a useful compile error message rather than just an opaque "const will always panic at runtime" error. And it's not like we have to worry about other cfgs with usize < u32; Rust nominally only supports down to a 16 bit usize (and a system with a pointer size of 256 values is constrained enough that good luck doing anything useful with anything other than asm). We can do both (the cfg and assert), if you prefer, but the cfg pays for itself by providing a useful compile error and documentation of intent in this case. — You are receiving this because your review was requested. Reply to this email directly, view it on GitHub <#18 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AANB3M7VB6VEN4GL2ABHZE3RHTLQRANCNFSM4LIBIDBQ> .

[matklad]

it would have made sense to send #[inline] and API changes in separate PRs :D It's important to separate bikeshedable and non-bikeshedable changes.

[CAD97]

separate PRs

Yeah I know I just sort of ended up lumping everything in this PR by accident.

[CAD97]

I think I've pruned this back to the approved changes now.

[matklad]

bors r+

[bors]

Build succeeded

• continuous-integration/travis-ci/push