[tbszz]

Description

--quiet does not suppress warnings/info messages when stdout is redirected (piped to a file or another process), contradicting the documented behavior of the flag:

Do not print to stdout or stderr, except for errors and warnings when provided once. Provide twice to suppress warnings as well.
— http --help

Fixes #1978

Reproduction (on unmodified master)

Any warning emitted while stdout is redirected leaks to stderr even with -qq. Using a local server that returns HTTP 500:

$ http --quiet --quiet --check-status GET http://127.0.0.1:18974/ > out.json   # note: stderr shown below
http: warning: HTTP 500 Internal Server Error

Same happens for e.g. an invalid config file (http: warning: invalid config file ...).

Root cause

Environment.log_error() gates the quiet-threshold check on self.stdout_isatty (httpie/context.py):

if self.stdout_isatty and self.quiet >= LOG_LEVEL_DISPLAY_THRESHOLDS[level]:
    stderr = self.stderr          # under --quiet this is devnull
else:
    stderr = self._orig_stderr    # always the real stderr

When stdout is not a TTY the condition short-circuits to _orig_stderr, so the message is printed regardless of the quiet level. The stdout_isatty gate was introduced together with the original --quiet implementation in #1175 / #1283; no test or discussion back then named piping semantics, and test_quiet_quiet_with_check_status_non_zero_pipe ended up pinning the buggy behavior.

Fix

Drop self.stdout_isatty from the condition so the quiet level decides alone:

• -qq + redirected stdout: warnings/info are suppressed (via the devnull'd self.stderr), same as on a TTY.
• Single -q: warnings still shown, on TTY and pipes alike.
• Errors: never hidden (LOG_LEVEL_DISPLAY_THRESHOLDS[ERROR] is infinite).
• No-quiet path: unchanged.

test_quiet_quiet_with_check_status_non_zero_pipe is flipped to assert the documented suppression (it previously asserted the bug).

Verification

• Regression evidence: the flipped test fails on unmodified master (assert not r.stderr vs actual 'http: warning: HTTP 500...') and passes with the fix. The TTY twin (test_quiet_quiet_with_check_status_non_zero) passes before and after (control).
• Live before/after with the same local HTTP 500 server and > out.json redirection: unmodified master prints http: warning: HTTP 500 Internal Server Error on stderr with -qq; with the fix stderr is empty; with a single -q the warning is still printed.
• Full suite (Windows, Python 3.13): with fix 989 passed / 5 failed / 11 errors; unmodified master in the same environment yields the identical failure set (Windows-specific DNS error text, CLI usage-text formatting, update-warnings fixture errors) — no regressions introduced.

[tbszz]

Noticed #1992 addresses the same issue (#1978) with the same one-line root-cause fix (dropping stdout_isatty from the log_error condition) — happy to defer to whichever the maintainers prefer.

One difference worth flagging: this PR also updates test_quiet_quiet_with_check_status_non_zero_pipe, which currently pins the buggy behavior (assert 'http: warning: HTTP 500' in r.stderr). With the fix applied, that assertion fails, so the test has to be flipped along with the change for the suite to stay consistent. Also included a CHANGELOG entry.

For context on the red CI: the test workflow is failing on master as well (e.g. runs from 2026-08-29 on the latest master commit are all failure/cancelled), so those failures don't come from this change.