feat(complete): Add per-candidate trailing space control by AndreasBackx · Pull Request #6266 · clap-rs/clap

AndreasBackx · 2026-02-13T03:02:16Z

This PR was generated with the assistance of an AI agent and reviewed by me.

Summary

Adds a nospace field to CompletionCandidate so the engine can tell shells "don't add a trailing space after this completion." Each shell adapter translates this into its native mechanism, replacing ad-hoc heuristics.

Fixes #5587

Changes

clap_complete/src/engine/candidate.rs: Add nospace: bool field with builder .nospace(bool) and getter .is_nospace_set().
clap_complete/src/engine/custom.rs: Set nospace(true) on directory candidates in complete_path().
clap_complete/src/engine/complete.rs: Set nospace(true) on --flag=value completions, -F=value completions, delimiter-prefixed completions, and require_equals options with = suffix.
clap_complete/src/env/shells.rs:
- Bash: Output _CLAP_COMPLETE_NOSPACE sentinel; registration script detects it and calls compopt -o nospace. Replaces the [=/:]$ regex heuristic.
- Zsh: Nospace candidates prefixed with \x1f; registration script splits into nospace/other groups using _describe 'values' nospace -S ''. Replaces directory-only split.
- Elvish: Nospace candidates prefixed with \x1f; registration script wraps them in edit:complex-candidate $value &code-suffix=''.
- Fish/PowerShell: No changes needed.
clap_complete/tests/testsuite/engine.rs: 5 new tests for nospace on directories, equals values, short equals, require_equals, and delimiters.
Snapshot files: Updated registration snapshots for Bash, Zsh, and Elvish.

Test Plan

cargo test -p clap_complete --features unstable-dynamic — 112 tests pass
cargo clippy -p clap_complete --features unstable-dynamic — no warnings
cargo fmt -p clap_complete --check — clean

…ompletions Add a `nospace: bool` field to `CompletionCandidate` that tells shell adapters to suppress the trailing space after a completion. This is set on: - Directory path completions (ending in `/`) - `--flag=value` inline completions (long and short) - Delimiter-prefixed completions (e.g., `comma,space`) - `require_equals` options (completed with trailing `=`) The field defaults to `false` and is exposed via: - Builder: `pub fn nospace(mut self, nospace: bool) -> Self` - Getter: `pub fn is_nospace_set(&self) -> bool` This is the first part of clap-rs#5587; per-shell handling follows.

Each shell adapter now translates the `nospace` field into its native mechanism: - **Bash**: Completions output a `_CLAP_COMPLETE_NOSPACE` sentinel as the last entry when any candidate has nospace set. The registration script detects this sentinel, strips it, and calls `compopt -o nospace`. This replaces the previous `[=/:]$` regex heuristic. - **Zsh**: Nospace candidates are prefixed with `\x1f` (unit separator) in the wire output. The registration script splits candidates into `nospace` and `other` groups, using `_describe 'values' nospace -S ''` for the former. This replaces the previous directory-only split. - **Elvish**: Nospace candidates are prefixed with `\x1f`. The registration script pipes output through `from-lines` and wraps nospace values in `edit:complex-candidate $value &code-suffix=''`. - **Fish**: No changes needed — Fish naturally suppresses trailing space for `/`-terminated completions and handles `=`-joined tokens through its tokenization. - **PowerShell**: No changes — `CompletionResult` does not support per-candidate nospace control. Closes clap-rs#5587

epage · 2026-02-13T16:01:26Z

+/// Like `complete` but includes `[nospace]` marker for candidates with nospace set
+fn complete_with_nospace(


We generally ask for tests to be added in a prior commit, with them passing in that commit

epage · 2026-02-13T16:03:10Z

+    /// This is useful for completions like directory paths (ending in `/`),
+    /// `--flag=` values, and delimiter-separated values where a trailing space
+    /// would be incorrect.


This is useful when there are more completion candidates if this completion is selected, like ...

epage · 2026-02-13T16:04:28Z

                        complete_arg_value(value.to_str().ok_or(value), arg, current_dir)
                            .into_iter()
-                            .map(|comp| comp.add_prefix(format!("--{flag}="))),
+                            .map(|comp| comp.add_prefix(format!("--{flag}=")).nospace(true)),


Why is this unconditionally adding nospace?

epage · 2026-02-13T16:04:49Z

+                            let comp = comp.add_prefix(format!("-{leading_flags}{sep}"));
+                            if has_equal {
+                                comp.nospace(true)
+                            } else {
+                                comp
+                            }


epage · 2026-02-13T16:05:26Z

        values = values
            .into_iter()
-            .map(|comp| comp.add_prefix(prefix))
+            .map(|comp| comp.add_prefix(prefix).nospace(true))


epage · 2026-02-13T16:16:24Z

+                    if a.is_require_equals_set() {
+                        populate_arg_candidate(CompletionCandidate::new(format!("--{s}=")), a)
+                            .nospace(true)
+                    } else {
+                        populate_arg_candidate(CompletionCandidate::new(format!("--{s}")), a)


Doesn't this conflict with your requires_equals PR?

I also think I prefer the approach taken in that PR.

epage · 2026-02-13T16:19:56Z

            let candidate = CompletionCandidate::new(suggestion.as_os_str().to_owned())
-                .hide(is_hidden(&raw_file_name));
+                .hide(is_hidden(&raw_file_name))
+                .nospace(true);


Should we limit this to !is_wanted? I'm guessing not because something may look like a valid candidate but the user may instead want a directory inside of the current one.

Granted, besides sort order, we don't have a good way of communicating which directories match is_wanted and which don't. Maybe the way to deal with that is to provide a way to add a description for wanted paths.

epage · 2026-02-13T16:22:18Z

+        let has_nospace = completions.iter().any(|c| c.is_nospace_set());
+
        for (i, candidate) in completions.iter().enumerate() {
            if i != 0 {
                write!(buf, "{}", ifs.as_deref().unwrap_or("\n"))?;
            }
            write!(buf, "{}", candidate.get_value().to_string_lossy())?;
        }
+        if has_nospace && !completions.is_empty() {
+            write!(buf, "{}", ifs.as_deref().unwrap_or("\n"))?;
+            write!(buf, "_CLAP_COMPLETE_NOSPACE")?;
+        }


This makes it so if any candidate doesn't have a space, none of them do.

Instead, if nospace is supported, could we always append a space and have bash always add nospace?

epage · 2026-02-13T16:23:26Z

+        local -a nospace=()
        local -a other=()
        local completion
        for completion in $completions; do
-            local value="${completion%%:*}"
-            if [[ "$value" == */ ]]; then
-                local dir_no_slash="${value%/}"
-                if [[ "$completion" == *:* ]]; then
-                    local desc="${completion#*:}"
-                    dirs+=("$dir_no_slash:$desc")
-                else
-                    dirs+=("$dir_no_slash")
-                fi
+            if [[ "$completion" == $'\x1f'* ]]; then
+                completion="${completion:1}"
+                nospace+=("$completion")
            else
                other+=("$completion")
            fi
        done
-        [[ -n $dirs ]] && _describe 'values' dirs -S '/' -r '/'
+        [[ -n $nospace ]] && _describe 'values' nospace -S ''


Again, this is doing it all or nothing

AndreasBackx added 2 commits February 13, 2026 02:40

AndreasBackx marked this pull request as draft February 13, 2026 03:02

epage reviewed Feb 13, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

feat(complete): Add per-candidate trailing space control#6266

feat(complete): Add per-candidate trailing space control#6266
AndreasBackx wants to merge 2 commits intoclap-rs:masterfrom
AndreasBackx:feat/completion-space-control

AndreasBackx commented Feb 13, 2026

Uh oh!

epage Feb 13, 2026

Uh oh!

epage Feb 13, 2026

Uh oh!

epage Feb 13, 2026

Uh oh!

epage Feb 13, 2026

Uh oh!

epage Feb 13, 2026

Uh oh!

epage Feb 13, 2026

Uh oh!

epage Feb 13, 2026

Uh oh!

epage Feb 13, 2026

Uh oh!

epage Feb 13, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

		/// Like `complete` but includes `[nospace]` marker for candidates with nospace set
		fn complete_with_nospace(

Uh oh!

Conversation

AndreasBackx commented Feb 13, 2026

Summary

Changes

Test Plan

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants