helix-tui: Skip line-endings in Buffer::set_string

[the-mikedavis]

The update of unicode-width to 0.1.13 changes the width of line endings to 1 (from 0), so we need to detect line-endings when skipping graphemes in Buffer::set_string_truncated_at_end.

Connects #10950

[RoloEdits]

Tried this fix, but I still had the same issues. This branch is what I build my custom build with. I cherrypicked the commit from this pr right on top.

This can be built with a simple command via https://github.com/casey/just

just

Which just expands to this, if you'd like:

RUSTFLAGS='-C target-cpu=native' cargo build --profile opt --bin hx

These are my rust-analyzer settings. (Most of it is still hold overs and random testing 😆)

[language-server.rust-analyzer]
config.check.command = "clippy"
config.check.extraArgs = ["--", "-W", "clippy::pedantic", "-W", "clippy::nursery", "-W", "clippy::cargo"]
cargo.buildScripts.rebuildOnSave = true
completion.fullFunctionSignatures.enable = true
cargo.features = "all"
hover.actions.references.enable = true
hover.show.structFields = 7
hover.show.traitAssocItems = 7
hover.memoryLayout.enable = true
hover.memoryLayout.alignment = "both"
hover.memoryLayout.niches = true
hover.memoryLayout.offset = "both"
hover.memoryLayout.size = "both"
highlightRelated.breakPoints.enable = true
highlightRelated.closureCaptures.enable = true
highlightRelated.exitPoints.enable = true
highlightRelated.references.enable = true
highlightRelated.yieldPoints.enable = true
inlayHints.bindingModeHints.enable = true
inlayHints.chainingHints.enable = true
inlayHints.closingBrace.enable = true
inlayHints.closureCaptureHints.enable = true
inlayHints.closureReturnTypeHints.enable = "always"
inlayHints.discriminantHints.enable  = "always"
inlayHints.implicitDrops.enable = true
inlayHints.parameterHints.enable = true
inlayHints.lifetimeElisionHints.enable = "always"
inlayHints.lifetimeElisionHints.useParameterNames = true
inlayHints.rangeExclusiveHints.enable = true
inlayHints.expressionAdjustmentHints.enable = "always"
typing.autoClosingAngleBrackets.enable = true
semanticHighlighting.doc.comment.inject.enable = true
typing.continueCommentsOnNewline = true

[RoloEdits]

It could also be that #6417 has to make fixes for the new changes as well, this has just been the most apparent way to show the issue at hand. Hate to make you go on a wild goose chase like this, kind of unfortunate that this wasn't considered a breaking change by unicode-width. There could be many areas where this could unknowingly crop up.

[the-mikedavis]

For inline diagnostics I believe it fixes it to copy this same width == 0 || str_is_line_ending(s) to the other if width == 0 checks in Buffer. But I found other breakages as well, for example opening a PDF causes many more artifacts than on master. Let's revert the unicode-width update for now, the update doesn't seem to benefit us in any way as far as I can tell anyways

[RoloEdits]

Sounds good.

[the-mikedavis]

Actually I think unicode-width's intepretation of this as non-breaking is correct since it's a bug-fix. It's just unfortunate that tui-rs (the crate that historically became helix-tui and ratatui) assumes line-endings are zero-width. I'm not well read on the unicode width vs. terminal symbol-width issue though so maybe @pascalkuthe has a more informed opinion here

[pascalkuthe]

Newlines are treated as zero width by all terminal emulators. For us they are width one because we actually show them but that is very helix specific.

Our width definition must align with the terminal emulator to avoid bugs (or we would need to start adding all kinds of special casing to the rendering).

I have my own implementation of unicode width in a repo. That crate also allows customizing which unicode version to use (also differs between emulators... fun). I didn't think it was worth bringing it in. I wrote.it originally because I thought I could solve the issues for multicodepoint graphemes but there is not standard for that and emulators can't agree on anything so it doesn't solve that problem