Switch to regex-cursor

[pascalkuthe]

Closes #185

Replaces our use of regex with regex-cursor, so we don't have to convert ropes to string. That implementation passes the regex testsuite so it should be stable but definitely needs testing.

See discussion at https://github.com/rust-lang/regex/issues/425.

I used the chance to replace split_at_newline with something that uses ropey so that unicode feature flags are correctly respected (should also be slightly faster).

[kirawi]

Crashed with (not sure what I did)

thread 'main' panicked at /home/sumire/.cargo/registry/src/index.crates.io-6f17d22bba15001f/regex-cursor-0.1.2/src/input.rs:204:13:
internal error: entered unreachable code: cursor does not support backtracking 506
stack backtrace:
   0: rust_begin_unwind
   1: core::panicking::panic_fmt
   2: regex_cursor::engines::dfa::search::find_fwd
   3: regex_cursor::engines::dfa::try_search
   4: regex_cursor::engines::meta::strategy::StrategyI::search
   5: regex_cursor::engines::meta::regex::Regex::search
   6: helix_term::commands::search_impl
   7: helix_term::commands::search_next_or_prev_impl
   8: helix_term::ui::editor::EditorView::handle_keymap_event::{{closure}}
   9: helix_term::ui::editor::EditorView::handle_keymap_event
  10: <helix_term::ui::editor::EditorView as helix_term::compositor::Component>::handle_event
  11: helix_term::compositor::Compositor::handle_event
  12: hx::main_impl::{{closure}}
  13: tokio::runtime::park::CachedParkThread::block_on
  14: tokio::runtime::context::runtime::enter_runtime
  15: tokio::runtime::runtime::Runtime::block_on
  16: hx::main

[pascalkuthe]

Crashed with (not sure what I did)

thread 'main' panicked at /home/sumire/.cargo/registry/src/index.crates.io-6f17d22bba15001f/regex-cursor-0.1.2/src/input.rs:204:13:
internal error: entered unreachable code: cursor does not support backtracking 506
stack backtrace:
   0: rust_begin_unwind
   1: core::panicking::panic_fmt
   2: regex_cursor::engines::dfa::search::find_fwd
   3: regex_cursor::engines::dfa::try_search
   4: regex_cursor::engines::meta::strategy::StrategyI::search
   5: regex_cursor::engines::meta::regex::Regex::search
   6: helix_term::commands::search_impl
   7: helix_term::commands::search_next_or_prev_impl
   8: helix_term::ui::editor::EditorView::handle_keymap_event::{{closure}}
   9: helix_term::ui::editor::EditorView::handle_keymap_event
  10: <helix_term::ui::editor::EditorView as helix_term::compositor::Component>::handle_event
  11: helix_term::compositor::Compositor::handle_event
  12: hx::main_impl::{{closure}}
  13: tokio::runtime::park::CachedParkThread::block_on
  14: tokio::runtime::context::runtime::enter_runtime
  15: tokio::runtime::runtime::Runtime::block_on
  16: hx::main

Thats a stranfe one that shouldn't happen but it's very hard to track down without reproduction case (the backtrack is too stripped because it's a release build)

[kirawi]

Ah, it's hx then /git nix. This is on a personal fork with multiple PRs pulled in (I forgot to mention) but I don't think any of them touch regex.

Edit: I can reproduce with this branch alone

[pascalkuthe]

oh yeah I can reproduce that is very odd. Is suspect it has do do with the special case of an empty file

[kirawi]

I don't think it's a special case with an empty file. If you delete the entire buffer and then do /git n it won't crash. If you add a new line and then 1G, it won't crash. If you add a new line without moving the cursor, it crashes. Maybe it has something to do with the cursor being on the line before the EOF newline?

[pascalkuthe]

It was an edgecase when you search for stuff at a one-past-the-end position which is what I suspected. Adapting the ropey cursor semantics to the one regex-cusor uses isn't trivial. I published a new version where that is fixed

[gabydd]

I also haven't hit any bugs with this anymore