Respect TERM=dumb for cursor position queries

[catlover-bot]

Summary

Respect TERM=dumb in Reedline's cursor-position queries and prompt painting.

Painter caches whether TERM=dumb at construction and refreshes the flag on resize. Cursor-position queries, right-prompt layout, and buffer cursor save/restore use this cached flag rather than reading the environment on each paint.

When cursor-position queries are deliberately skipped, anchor_prompt keeps a stale bottom-row anchor without printing an extra newline. Actual query errors retain #1202's newline recovery and stale bottom-row fallback.

Right prompts are suppressed for TERM=dumb. Both small- and large-buffer painting skip SavePosition, and cursor placement skips restoring or repositioning the cursor, leaving it where the last Print put it.

TERM=dumb also disables ANSI coloring without changing the configured with_ansi_colors value. ANSI coloring continues to use the engine's existing TERM-aware policy.

Testing

• Regression tests for skipped cursor-position queries and anchoring without an extra blank line.
• Output-capture regression tests for no ESC 7 / ESC 8 in small- and large-buffer painting.
• Painter tests explicitly set the cached terminal flag instead of depending on the process environment.
• Full all-features suite under both TERM=xterm-256color and TERM=dumb, using --test-threads=1: 1794 passed, 1 ignored, and 30 doctests passed in each run.
• cargo clippy --locked --all-targets --all-features -- -D warnings
• cargo fmt --all -- --check
• git diff --check

Fixes #1009

[fdncred]

I wonder if it's worth centralizing all interactions with the env in calls like std::env::var_os("TERM");? It may be nice to have one function to rule them all since dealing with the env isn't threadsafe. Thoughts?

[catlover-bot]

That makes sense to me.

I checked the current codebase and found only three direct process environment-variable reads: TERM, USERPROFILE, and HOME, so the scope is small enough to centralize here.

I pushed a follow-up commit that routes those reads through a single utils::environment::var_os helper. I left APIs such as current_dir() out since those are process-state operations rather than environment-variable reads.

One nuance is that the wrapper itself does not make concurrent external environment mutation thread-safe. The benefit is that Reedline now has one place where a future synchronization or snapshot policy can be implemented instead of having environment-variable reads scattered around the crate.

The full library tests and clippy are still green after the refactor.

[fdncred]

The benefit is that Reedline now has one place where a future synchronization or snapshot policy can be implemented instead of having environment-variable reads scattered around the crate.

yup, that's exactly what I was asking.

Thanks for the update. We'll wait to see what @kronberger-droid thinks. I'm not sure it's a regression so I doubt this will be for the upcoming patch.

[kronberger-droid]

Definitely not a regression, thats a longstanding issue.
I would not rush it in before the patch.

I will take a closer look tomorrow.

[catlover-bot]

Thanks!

[kronberger-droid]

Well I think it would make sense to thread in the ansi-coloring removal here to.
It should be easy enough since we already handle it consistently through the config.

[catlover-bot]

Thanks, I threaded TERM=dumb through the existing ANSI-color configuration as well.

The configured with_ansi_colors value is preserved, while the effective value is disabled when the current TERM is dumb. I compute that effective value once per paint operation, so the same render uses a consistent setting while later runtime changes to TERM are still respected.

I also rebased onto current main and kept the newer W::cursor_position() abstraction, so non-dumb terminals continue through the existing testable terminal-I/O path.

I added coverage for TERM=dumb, a regular terminal, and an unset TERM; the full test suite and clippy are green.

[kronberger-droid]

Tested this in Emacs shell-mode, the hard error is gone, nice!

One problem still remains with the right prompt:

Reedline parks the cursor with ESC 7 / ESC 8 to paint it, which is terminfo sc/rc, and dumb doesn't have either.

They land in the buffer literally and CurrentDateTime gets inlined into the prompt line:

~/Projects/rust/reedline> ^[708/26/2026 03:35:09 PM^[8^[7^[8hello world

I would also just skip the right prompt since the terminal can't position it anyways.

[catlover-bot]

Thanks for testing this in Emacs shell-mode. I pushed a follow-up that suppresses the right prompt when TERM=dumb, so the right-prompt paint path no longer emits its cursor save/restore sequence. The TERM read remains dynamic. I added regression coverage for the right-prompt layout; the full suite passes with 1606 tests passed, 1 ignored, plus 30 doctests, and clippy is clean.

[catlover-bot]

Thanks for testing this in Emacs shell-mode. I now skip the right prompt entirely for TERM=dumb, so Reedline will not emit the save/restore cursor sequences needed to paint it there.

[kronberger-droid]

Just those two comments and a rebase and then I think we are ready.

[catlover-bot]

Thanks! I addressed both comments and rebased the branch onto the current main. Local tests, clippy, rustfmt, diff checks, and CI are all green.

[kronberger-droid]

Hey there, sorry that I come back with other requests.
But the changes should be small and contained:

Blank line before every prompt.
The query answers Ok(None) on every read_line, so anchor_prompt takes the #1202 fallback each time and prints a \r\n first.
Split the match: Err(_) keeps the print_crlf, Ok(None) only sets the bottom row Stale anchor.

ESC 7 / ESC 8 still lands in the buffer.
The right prompt pair is gone, the buffer paint one is not: SavePosition after before_cursor, RestorePosition in queue_cursor_placement.
Skip both under the same dumb predicate and leave the cursor where the last Print put it.

That is four TERM reads per paint now.
Read it once into a bool on the Painter, at construction and on resize, and take it from there in compute_layout and the print paths.
Then the right prompt tests can set the flag instead of the environment, six of them fail when the suite runs under TERM=dumb.

[catlover-bot]

Thanks for catching these. I've pushed a follow-up addressing all three points.

anchor_prompt now separates Ok(None) from Err(_): skipping the query keeps the stale bottom-row anchor without emitting a newline, while actual query errors retain #1202's newline recovery.

Both buffer paint paths now skip SavePosition for dumb terminals, and queue_cursor_placement returns without restoring or repositioning the cursor.

Painter caches the dumb-terminal flag at construction and refreshes it on resize. Its cursor queries, right-prompt layout, and buffer print paths use that flag. The affected tests set it explicitly instead of depending on the process environment.

I added output-capture regressions for the extra blank line and the remaining save/restore escapes. The full all-features suite passes under both TERM=xterm-256color and TERM=dumb with --test-threads=1: 1794 passed, 1 ignored, and 30 doctests passed in each run. Clippy, rustfmt, and diff checks also pass.

I also updated the PR description to reflect the cached Painter policy.

[kronberger-droid]

Nice thanks a lot.