lisa: Remove uses of polars.StringCache()

FIX

StringCache has been a no-op for quite some time now (since polars
1.32). It has now been marked as deprecated in polars 1.41, so we can
remove it from LISA code.

Author: douglas-raillard-arm

lisa/analysis/base.py

@@ -1199,28 +1199,23 @@ def df_method(cls, f, index=None):
         # can be changed using the AnalysisProxy(params=dict(...)) when the
         # AnalysisProxy is instanciated in lisa.trace
         def wrapper(self, *args, df_fmt=None, **kwargs):
-            # Ease working with LazyFrames coming from various sources. When
-            # they are collect()'ed in f(), they will be created using a common
-            # StringCache so Categorical columns can be concatenated and such.
-            with pl.StringCache():
-
-                # We might get different types based on whether the content
-                # comes from the function directly (could be a pandas object)
-                # or from the cache (polars LazyFrame).
-                df = cached_f(self, *args, **kwargs)
-                assert isinstance(df, (pd.DataFrame, pl.DataFrame, pl.LazyFrame))
-
-                df_fmt = df_fmt or 'pandas'
-                df = _df_to(
-                    df,
-                    fmt=df_fmt,
-                    index=(
-                        ('Time' if 'Time' in df.collect_schema().names() else None)
-                        if index is None and isinstance(df, (pl.LazyFrame, pl.DataFrame)) else
-                        index
-                    ),
-                )
-                return df
+            # We might get different types based on whether the content
+            # comes from the function directly (could be a pandas object)
+            # or from the cache (polars LazyFrame).
+            df = cached_f(self, *args, **kwargs)
+            assert isinstance(df, (pd.DataFrame, pl.DataFrame, pl.LazyFrame))
+
+            df_fmt = df_fmt or 'pandas'
+            df = _df_to(
+                df,
+                fmt=df_fmt,
+                index=(
+                    ('Time' if 'Time' in df.collect_schema().names() else None)
+                    if index is None and isinstance(df, (pl.LazyFrame, pl.DataFrame)) else
+                    index
+                ),
+            )
+            return df
 
         return wrapper
 

lisa/analysis/tasks.py

@@ -317,9 +317,7 @@ def load(event, *args, **kwargs):
             maintain_order=True,
         )
         df = df.select('name', 'pid')
-
-        with pl.StringCache():
-            df = _polars_fast_collect(df)
+        df = _polars_fast_collect(df)
 
         def finalize(df, key_col):
             assert len(df.columns) == 2

lisa/trace.py

@@ -6367,20 +6367,19 @@ def _data_to_parquet(cls, data, path, compression=_DEFAULT_PARQUET_COMPRESSION,
         elif isinstance(data, pl.DataFrame):
             data.write_parquet(path, **kwargs)
         elif isinstance(data, pl.LazyFrame):
-            with pl.StringCache():
-                try:
-                    # TOOD: revisit when polars streaming engine is complete
-                    # and it does not raise a DeprecationWarning anymore.
-                    with warnings.catch_warnings():
-                        warnings.simplefilter("ignore", category=DeprecationWarning)
-                        data.sink_parquet(path, **kwargs)
-                # The streaming engine may have issues with some LazyFrames, so
-                # fall back on collecting.
-                except Exception as e:
-                    cls.get_logger().debug(f'Could not use sink_parquet(): {e}')
-                    path.unlink(missing_ok=True)
-                    data = _polars_fast_collect(data)
-                    data.write_parquet(path, **kwargs)
+            try:
+                # TOOD: revisit when polars streaming engine is complete
+                # and it does not raise a DeprecationWarning anymore.
+                with warnings.catch_warnings():
+                    warnings.simplefilter("ignore", category=DeprecationWarning)
+                    data.sink_parquet(path, **kwargs)
+            # The streaming engine may have issues with some LazyFrames, so
+            # fall back on collecting.
+            except Exception as e:
+                cls.get_logger().debug(f'Could not use sink_parquet(): {e}')
+                path.unlink(missing_ok=True)
+                data = _polars_fast_collect(data)
+                data.write_parquet(path, **kwargs)
         else:
             data.to_parquet(path, **kwargs)
 
@@ -7199,12 +7198,6 @@ def __init__(self,
             at_exit=True,
         )
 
-        # Make sure that we always operate with an active StringCache when
-        # manipulating a trace object. This prevents issues with LazyFrame
-        # built out of a DataFrame containing Categorical data, in places where
-        # the user does not control the creation of the DataFrame.
-        cm_stack.enter_context(pl.StringCache())
-
         trace_path = str(trace_path) if trace_path else None
         self.trace_path = trace_path
 
@@ -7849,7 +7842,7 @@ def _get_parser(
 
         @contextlib.contextmanager
         def cm():
-            with pl.StringCache(), self._cache._parser_temp_path() as temp_dir:
+            with self._cache._parser_temp_path() as temp_dir:
                 self._activity_log.log(
                     'spinup-parser',
                     {

setup.py

@@ -143,6 +143,7 @@ def _find_packages(toplevel):
             # Require:
             # polars >= 1.15.0: https://github.com/pola-rs/polars/issues/19994
             # polars >= 1.16.0: https://github.com/pola-rs/polars/issues/20000
+            # polars >= 1.32.0: https://github.com/pola-rs/polars/pull/23016
             # polars >= 1.34.0: https://github.com/pola-rs/polars/pull/24531
             # polars != 1.35.1: https://github.com/pola-rs/polars/issues/25261
             # polars != 1.35.2: https://github.com/pola-rs/polars/issues/25261