From 61c42889b4bd4b91fe8660a3f17d3ad0dedf247b Mon Sep 17 00:00:00 2001
From: 魏曹先生 <1992414357@qq.com>
Date: Mon, 3 Aug 2026 17:22:44 +0800
Subject: feat(comps): improve bash completion script word tracking

Rework bash completion script to fix word-index tracking when the
cursor is in the middle of a word, compute the current word from
COMP_LINE truncated at COMP_POINT, and use space-separated option
arguments. Also trim colon prefixes from completion items when
COMP_WORDBREAKS includes colons.
---
 CHANGELOG.md | 2 ++
 1 file changed, 2 insertions(+)

(limited to 'CHANGELOG.md')

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 69bfe43..590d566 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -59,6 +59,8 @@ None
 
 1. **[`comps:zsh`]** Fixed zsh completion script to properly escape colons in completion descriptions. The zsh completion script generated for the `zsh` output format now escapes colon characters in completion items (`${item//:/\\:}`) and description parts (`${match[1]//:/\\:}`) so that descriptions containing colons don't break the `_describe` command's parsing. Additionally, fixed the simple-completions branch to iterate over the original `completions` array (with the colon-escaped format matching) rather than the already-parsed `parsed_completions` array, correctly extracting the completion item when no description is present.
 
+2. **[`comps:bash`]** Reworked the bash completion script template to fix word-index tracking when the cursor is in the middle of a word and to add colon-ltrimming for completion descriptions. The script now computes the current word (`cur`) using `COMP_LINE` truncated to `COMP_POINT` (taking the last whitespace-delimited token) rather than relying on `COMP_WORDS[COMP_CWORD]`, which fails when the cursor is in the middle of a word. The word index is derived from the count of words preceding the current cursor position (`before_words`), and `prev` is the last word in that preceding set. The option flags to the underlying completion engine are now passed as `-f value`/`-C value`/etc. (space-separated) instead of `-f=value`/`-C=value`/etc. (equals-separated), since the engine expects space-delimited argument pairs. Additionally, when the current word contains a colon and `COMP_WORDBREAKS` includes `:`, the completion items are trimmed of the colon prefix before being inserted into `COMPREPLY` — this prevents completions like `foo:bar` from being double-prefixed with the literal `foo:bar` when bash would otherwise append the full word.
+
 #### Optimizations:
 
 1. **[`pathf`]** Added `is_module` field to `AnalyzeItem` and a new constructor `AnalyzeItem::local_module(module, item_name)` which sets `is_module: true`. The `type_mapping_builder` now tracks whether an item is a module: when generating `type_using.rs`, module items produce `use path::to::module::*;` (glob import) instead of the standard `use path::to::TypeName;` direct import. Non-module items continue to use direct imports as before. The internal data structure changed from `Vec<(String, String)>` to `Vec<(String, String, bool)>` to carry the `is_module` flag through the pipeline.
-- 
cgit