Fail a trial if it produces empty csv file (#1006)

# Pull Request

## Title

Fail a trial if it produces empty csv file.
A follow-up to #998 
______________________________________________________________________

## Description

* Fail a trial if it produces empty csv file.
* Add unit test to check for the condition.
* Small fix in `split_cmdline()` function to allow empty token strings
(occurs in the new unit test)

______________________________________________________________________

## Type of Change

- 🛠️ Bug fix
- 🧪 Tests

______________________________________________________________________

## Testing

Run unit tests as usual

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Brian Kroth <bpkroth@users.noreply.github.com>

Author: 3 people

mlos_bench/mlos_bench/environments/local/local_env.py

@@ -195,19 +195,23 @@ def run(self) -> tuple[Status, datetime, dict[str, TunableValue] | None]:
             _LOG.debug("Not reading the data at: %s", self)
             return (Status.SUCCEEDED, timestamp, stdout_data)
 
-        data = self._normalize_columns(
-            pandas.read_csv(
-                self._config_loader_service.resolve_path(
-                    self._read_results_file,
-                    extra_paths=[self._temp_dir],
-                ),
-                index_col=False,
+        try:
+            data = self._normalize_columns(
+                pandas.read_csv(
+                    self._config_loader_service.resolve_path(
+                        self._read_results_file,
+                        extra_paths=[self._temp_dir],
+                    ),
+                    index_col=False,
+                )
             )
-        )
+        except pandas.errors.EmptyDataError:
+            _LOG.warning("Empty metrics file - fail the run")
+            return (Status.FAILED, timestamp, None)
 
         _LOG.debug("Read data:\n%s", data)
         if len(data) == 0:
-            _LOG.warning("Empty metrics file - fail the run")
+            _LOG.warning("No data in the metrics file - fail the run")
             return (Status.FAILED, timestamp, None)
         elif list(data.columns) == ["metric", "value"]:
             _LOG.info(

mlos_bench/mlos_bench/services/local/local_exec.py

@@ -46,7 +46,7 @@ def split_cmdline(cmdline: str) -> Iterable[list[str]]:
     cmdline_tokens.whitespace_split = True
     subcmd = []
     for token in cmdline_tokens:
-        if token[0] not in cmdline_tokens.punctuation_chars:
+        if token[:1] not in cmdline_tokens.punctuation_chars:
             subcmd.append(token)
         else:
             # Separator encountered. Yield any non-empty previous subcmd we accumulated.

mlos_bench/mlos_bench/tests/environments/local/local_env_test.py

@@ -104,8 +104,31 @@ def test_local_env_wide(tunable_groups: TunableGroups) -> None:
     )
 
 
+def test_local_env_results_null_file(tunable_groups: TunableGroups) -> None:
+    """When the results file is of zero length, do not crash but mark the trial
+    FAILED.
+    """
+    local_env = create_local_env(
+        tunable_groups,
+        {
+            "run": [
+                "echo '' > output.csv",
+            ],
+            "read_results_file": "output.csv",
+        },
+    )
+
+    check_env_success(
+        local_env,
+        tunable_groups,
+        expected_status_run={Status.FAILED},
+        expected_results=None,
+        expected_telemetry=[],
+    )
+
+
 def test_local_env_results_empty_file(tunable_groups: TunableGroups) -> None:
-    """When the results file is empty, do not crash but mark the trial FAILED."""
+    """When the results file has no data, do not crash but mark the trial FAILED."""
     local_env = create_local_env(
         tunable_groups,
         {