[sungbin1015]

Problem

dotenv run rejects a .env path that is a FIFO, although the library (load_dotenv, dotenv_values, find_dotenv, since #586) and the list/get subcommands read it fine:

$ dotenv -f <(echo A=1) list
A=1
$ dotenv -f <(echo A=1) get A
1
$ dotenv -f <(echo A=1) run printenv A
Usage: dotenv run [OPTIONS] [COMMANDLINE]...
Try 'dotenv run --help' for help.

Error: Invalid value: Invalid value for '-f' "/dev/fd/63" does not exist.

The same happens with a named pipe created by mkfifo and picked up as the default ./.env, which is the 1Password-mounted .env case #586 was written for: load_dotenv() works there, dotenv run -- cmd does not.

Cause

run guards the path with os.path.isfile(file) before handing it to DotEnv. DotEnv._get_stream and find_dotenv use _is_file_or_fifo instead; the guard in cli.py was not updated when FIFO support was added.

Fix

Use _is_file_or_fifo in the run guard. Missing paths and directories are still rejected with the same message. No change on Windows, where _is_file_or_fifo is equivalent to os.path.isfile.

Tests

• tests/test_fifo_dotenv.py::test_cli_run_from_fifo (Unix only, like the existing FIFO test): fails before the change with the error above, passes after.
• pytest: 285 passed, 4 skipped. ruff check src tests, ruff format --check src tests and mypy for 3.10-3.14 are clean.

Related work

• Support reading .env from FIFOs (Unix) #586 (merged) added FIFO support to the library functions; this PR extends the same check to the CLI's run path, which Support reading .env from FIFOs (Unix) #586 did not cover.
• feat: Add waiting logic for FIFO files that may not be immediately readable #599 (open, for Bug: python-dotenv doesn't load 1Password “mounted” .env files (looks like a FIFO / fd-backed file) #598) adds waiting logic in src/dotenv/main.py for FIFOs that are not immediately readable. It does not touch cli.py, so dotenv run would still reject a FIFO with it applied.
• Resolve the default CLI env path on each invocation #718 (open) also edits cli.py, but for a different problem (when the default .env path is resolved).

[Riccardo-Vecchi]

Independent Linux validation of 92216393627596e62e7e7cf6152a43d33c64db63 (OpenAI Codex-assisted).

A private subprocess matrix uses real os.mkfifo pipes and the actual python -m dotenv run entry point, with synthetic values and a five-second timeout per command:

• default and explicit FIFO paths, a symlink to a FIFO, and a producer delayed by 200 ms;
• --no-override preserving a parent value and --override replacing it;
• child exit status 7, a multiline UTF-8 value, and Bash <(...) process substitution;
• regular-file success, missing-path rejection and directory rejection as controls.

On unchanged base 0b2880591780a426b0551436f47a17e7a76d954f, all nine FIFO cases return the existing exit-2 error; the three controls pass. On this PR head, 12/12 pass. Writers use bounded nonblocking opens and are joined, so a rejection cannot leave a blocked producer behind.

CPython 3.12.3: the repository suite passes 289 tests on the PR, versus 288 on the base. Both runs emit the same 35 warnings (Click's isolated-filesystem deprecation and IPython's temporary-directory fallback). Ruff check/format and mypy targets 3.10–3.14 pass; those mypy targets are static checks, not runs on five interpreters. The checkout is unchanged.

This validates the POSIX CLI/FIFO path and environment/exit-code behavior. It does not establish compatibility with a real 1Password mount, macOS or Windows.