No way to capture stderr #11

Problem

_run_stream captures stderr via subprocess.PIPE but never exposes it to the caller or subclasses. The captured process.stderr and process.returncode are silently discarded.

This makes it impossible to:

• Log diagnostic output from TF_LOG during streaming operations (plan, apply)
• Detect and report errors that terraform/tofu writes to stderr
• Build subclasses that add logging or monitoring without fully reimplementing _run_stream

In contrast, _run already returns a CommandResults dataclass that includes both stderr and returncode, so the pattern is established.

Proposed Solution

Store the process result on the instance after _run_stream executes, reusing the existing CommandResults dataclass. This is non-breaking — the generator API stays the same.

def _run_stream(self, args):
    args = [self.binary_path] + [str(x) for x in args]

    process = subprocess.run(
        args,
        cwd=self.cwd,
        stdout=subprocess.PIPE,
        capture_output=False,
        stderr=subprocess.PIPE,
        universal_newlines=True,
        encoding="utf-8",
        bufsize=1,
        env={
            **os.environ,
            "TF_IN_AUTOMATION": "1",
            "TF_LOG": self.log_level,
            **self.env,
        },
        timeout=None,
    )

    # NEW: store result for callers/subclasses to inspect after iteration
    self._last_stream_result = CommandResults(
        command=" ".join(args),
        stdout=process.stdout,
        stderr=process.stderr,
        returncode=process.returncode,
    )

    event_line = ""
    for buffer in process.stdout:
        if buffer == "\n":
            yield json.loads(event_line)
            event_line = ""
        else:
            event_line += buffer

Context

We use tofupy in a Celery worker that runs terraform operations. Without stderr access, we have no visibility into terraform diagnostic output, which makes debugging production issues difficult.

P.S: TF_LOG_PATH env var: tofupy already supports passing env={"TF_LOG_PATH": "/tmp/terraform.log"} to redirect terraform logs to a file. This works but requires managing temp files and reading them after each operation, which is awkward in long-running workers that want to stream logs to Python's logging in real time. And it's for debug, it anyway doesn't capture the stderr...