Regression-test driver hides errors: stderr is not captured, failures appear "without a message"

[luwang00]

Summary

When a regression test that goes through the Python driver layer fails, the actual error (Fortran fatal error, Python traceback, or shared-library load failure) is not shown anywhere the user normally looks. The test simply reports ***Failed with no diagnostic, and the per-case .log file ends mid-run with no error text. This makes these failures very hard to triage.

I hit this with py_wavetank_test1, but the defect is in the shared driver helper and affects every case routed through it (the run*DriverCase family and the C-binding Python tests).

Root cause

_runCase in reg_tests/lib/openfastDrivers.py redirects only stdout to the log file; stderr is never redirected:

def _runCase(executable, inputFile, logFile, stdout, restart=False, ExtraFlags=""):
    if logFile is None:
        command = f"{executable} {inputFile} {ExtraFlags}"
    elif restart:
        command = f"{executable} -restart {os.path.splitext(inputFile)[0]} > {logFile}"
    else:
        command = f"{executable} {inputFile} {ExtraFlags} > {logFile}"
    print(command)
    return subprocess.call(command, stdout=stdout, shell=True)

Three effects combine to make the failure look silent:

1. stdout → log file. The > {logFile} redirect sends normal progress output into <case>.log, so nothing appears on the console.
2. stderr is dropped. There is no 2> / 2>&1, so Python tracebacks, the ctypes "cannot find module / one of its dependencies" DLL-load error, and Fortran fatal errors raised back through the C-binding (e.g. RuntimeError: ... Fatal Error: ...) go to stderr and are written neither to the log file nor merged with stdout.
3. ctest hides output by default. Without --output-on-failure (or -V), ctest captures the streams and only prints ***Failed, so the stderr text that would otherwise reach the console is suppressed too.

Net result: the real error exists but is split — benign stdout into the .log file and the actual error on stderr, which is discarded. The .log for py_wavetank_test1, for example, ended at Running AeroDyn. with no indication of what went wrong.

Steps to reproduce

1. Cause any driver-based case to fail for a reason that reports on stderr, e.g. a C-binding fatal error, a missing Python dependency, or a missing/incompatible shared library.
2. Run it via ctest: ctest -R py_wavetank_test1.
3. Observe ***Failed with no error message; inspect <build>/.../<case>.log and note it ends mid-run with no error text.
4. Run the driver directly capturing all streams (python <driver>.py *> out.txt, or 2>&1) to finally see the real error.

Expected behavior

The failing command's stderr should be captured into the case .log file (and/or shown by ctest on failure) so the actual error is discoverable without manually re-running the driver.

Proposed fix

Merge stderr into the redirected log in _runCase, e.g.:

    elif restart:
        command = f"{executable} -restart {os.path.splitext(inputFile)[0]} > {logFile} 2>&1"
    else:
        command = f"{executable} {inputFile} {ExtraFlags} > {logFile} 2>&1"

(Equivalently, pass stderr=subprocess.STDOUT and write to the log via the file object instead of a shell redirect.) This keeps the console clean while ensuring the .log always contains the error.

Environment

• File: reg_tests/lib/openfastDrivers.py (_runCase)