Add to_usize

[matklad]

The main benefit of .to_usize versus .into() is that the type is
fixed for the purposes of type-inference. This is handly if you want
to do "hello, world"[..ts.to_usize()].

To clarify, the only reason why we might want this function is because "hello, world"[..ts] doesn't work, because of perma-unstable indexing trait in std :(

[CAD97]

I think the string[..size] case is probably better served by a string[TextRange::to(size)] (or maybe TextRange::from(..size)) than the ".ix()" method I had previously (roughly, "turn this into a standard indexing type").

Rowan doesn't use to_usize() at all. rust-analyzer does, so let's walk through the uses:

• ra_text_edit::TextEdit::apply
  Looks like it can (should?) be using TextUnit directly.
• ra_text_edit::AtomTextEdit::apply
  Uses String::replace_range, which requires impl RangeBounds<usize>. Unfortunately, as RangeBounds returns Bound<&T>, this will continue to be required unless we provide and tell people to use our own replace_range. Better served by TextRange -> ops::Range<usize> than TextSize -> usize, however.
• Iterator for ra_ide_db::line_index_utils::OffsetStepIter
  • if size > 1 (also a use of from_usize)
  • RangeFrom (possible address: Idea: RangeFrom #15)
• ra_ide::syntax_tree::syntax_tree_for_token
  Should just be using TextUnit directly.
• ra_syntax::parsing::lexer::lex_single_syntax_kind
  Should just be using TextUnit directly.
• ra_syntax::parsing::lexer::lex_single_valid_syntax_kind
  Should just be using TextUnit directly.
• ra_ide_db::line_index::LineIndex::new
if size > 1
• ra_ide_db::line_index::LineIndex::utf8_to_utf16_col
  Legit. Could be unambiguously .into() in this position, but maybe should be converted to usize sooner.
• ra_assists::helpers::check
&before[range.start().to_usize()..range.end().to_usize()], written (moved?) by @matklad 16 days ago, who should've known that they could've just written &before[range] 😛
• ra_syntax::tests::dump_tokens_and_errors
  Should just be using TextUnit directly.
• ra_hir_def::body::scope::tests::do_check
  RangeFrom again.
• test_utils::add_cursor
  RangeFrom again.
• rust_analyzer::main_loop::handlers::handle_on_type_formatting
  RangeFrom again.

Conclusion: we definitely want TextRange::from(size), TextSize::one(), and a solution for replace_range. Otherwise, a nudge that to_usize is not an idiomatic code path (by not having a dedicated function) seems reasonable enough.

[CAD97]

Notes from rust-lang/rust-analyzer#3570:

• String::with_capacity(size.to_usize()) is a definite use of .to_usize(), but is equally served by .into()
• The utf8_to_utf16_col version that uses usize more where it should almost certainly needs a type-hinted .to_usize() to be written the way that I did.

[matklad]

Hm, given the amount of false-positives in rust-analyzer, I think there's a clear benefit in not providing to_usize method :)

If only there was a clearly good name for TextRange::to(size)...