polars-df 0.9.0-x86_64-linux → 0.11.0-x86_64-linux

data/lib/polars/{io.rb → io/csv.rb}

@@ -80,6 +80,8 @@ module Polars
     #   allocation needed.
     # @param eol_char [String]
     #   Single byte end of line character.
+    # @param truncate_ragged_lines [Boolean]
+    #   Truncate lines that are longer than the schema.
     #
     # @return [DataFrame]
     #
@@ -113,7 +115,8 @@ module Polars
       row_count_name: nil,
       row_count_offset: 0,
       sample_size: 1024,
-      eol_char: "\n"
+      eol_char: "\n",
+      truncate_ragged_lines: false
     )
       Utils._check_arg_is_1byte("sep", sep, false)
       Utils._check_arg_is_1byte("comment_char", comment_char, false)
@@ -138,7 +141,7 @@ module Polars
 
       df = nil
       _prepare_file_arg(source) do |data|
-        df = DataFrame._read_csv(
+        df = _read_csv_impl(
           data,
           has_header: has_header,
           columns: columns || projection,
@@ -161,7 +164,8 @@ module Polars
           row_count_name: row_count_name,
           row_count_offset: row_count_offset,
           sample_size: sample_size,
-          eol_char: eol_char
+          eol_char: eol_char,
+          truncate_ragged_lines: truncate_ragged_lines
         )
       end
 
@@ -172,88 +176,24 @@ module Polars
       end
     end
 
-    # Lazily read from a CSV file or multiple files via glob patterns.
-    #
-    # This allows the query optimizer to push down predicates and
-    # projections to the scan level, thereby potentially reducing
-    # memory overhead.
-    #
-    # @param source [Object]
-    #   Path to a file.
-    # @param has_header [Boolean]
-    #   Indicate if the first row of dataset is a header or not.
-    #   If set to false, column names will be autogenerated in the
-    #   following format: `column_x`, with `x` being an
-    #   enumeration over every column in the dataset starting at 1.
-    # @param sep [String]
-    #   Single byte character to use as delimiter in the file.
-    # @param comment_char [String]
-    #   Single byte character that indicates the start of a comment line,
-    #   for instance `#`.
-    # @param quote_char [String]
-    #   Single byte character used for csv quoting.
-    #   Set to None to turn off special handling and escaping of quotes.
-    # @param skip_rows [Integer]
-    #   Start reading after `skip_rows` lines. The header will be parsed at this
-    #   offset.
-    # @param dtypes [Object]
-    #   Overwrite dtypes during inference.
-    # @param null_values [Object]
-    #   Values to interpret as null values. You can provide a:
-    #
-    #   - `String`: All values equal to this string will be null.
-    #   - `Array`: All values equal to any string in this array will be null.
-    #   - `Hash`: A hash that maps column name to a null value string.
-    # @param ignore_errors [Boolean]
-    #   Try to keep reading lines if some lines yield errors.
-    #   First try `infer_schema_length: 0` to read all columns as
-    #   `:str` to check which values might cause an issue.
-    # @param cache [Boolean]
-    #   Cache the result after reading.
-    # @param with_column_names [Object]
-    #   Apply a function over the column names.
-    #   This can be used to update a schema just in time, thus before
-    #   scanning.
-    # @param infer_schema_length [Integer]
-    #   Maximum number of lines to read to infer schema.
-    #   If set to 0, all columns will be read as `:str`.
-    #   If set to `nil`, a full table scan will be done (slow).
-    # @param n_rows [Integer]
-    #   Stop reading from CSV file after reading `n_rows`.
-    # @param encoding ["utf8", "utf8-lossy"]
-    #   Lossy means that invalid utf8 values are replaced with `�`
-    #   characters.
-    # @param low_memory [Boolean]
-    #   Reduce memory usage in expense of performance.
-    # @param rechunk [Boolean]
-    #   Reallocate to contiguous memory when all chunks/ files are parsed.
-    # @param skip_rows_after_header [Integer]
-    #   Skip this number of rows when the header is parsed.
-    # @param row_count_name [String]
-    #   If not nil, this will insert a row count column with the given name into
-    #   the DataFrame.
-    # @param row_count_offset [Integer]
-    #   Offset to start the row_count column (only used if the name is set).
-    # @param parse_dates [Boolean]
-    #   Try to automatically parse dates. If this does not succeed,
-    #   the column remains of data type `:str`.
-    # @param eol_char [String]
-    #   Single byte end of line character.
-    #
-    # @return [LazyFrame]
-    def scan_csv(
-      source,
+    # @private
+    def _read_csv_impl(
+      file,
       has_header: true,
+      columns: nil,
       sep: ",",
       comment_char: nil,
       quote_char: '"',
       skip_rows: 0,
       dtypes: nil,
+      schema: nil,
       null_values: nil,
+      missing_utf8_is_empty_string: false,
       ignore_errors: false,
-      cache: true,
-      with_column_names: nil,
+      parse_dates: false,
+      n_threads: nil,
       infer_schema_length: 100,
+      batch_size: 8192,
       n_rows: nil,
       encoding: "utf8",
       low_memory: false,
@@ -261,417 +201,119 @@ module Polars
       skip_rows_after_header: 0,
       row_count_name: nil,
       row_count_offset: 0,
-      parse_dates: false,
-      eol_char: "\n"
-    )
-      Utils._check_arg_is_1byte("sep", sep, false)
-      Utils._check_arg_is_1byte("comment_char", comment_char, false)
-      Utils._check_arg_is_1byte("quote_char", quote_char, true)
-
-      if Utils.pathlike?(source)
-        source = Utils.normalise_filepath(source)
-      end
-
-      LazyFrame._scan_csv(
-        source,
-        has_header: has_header,
-        sep: sep,
-        comment_char: comment_char,
-        quote_char: quote_char,
-        skip_rows: skip_rows,
-        dtypes: dtypes,
-        null_values: null_values,
-        ignore_errors: ignore_errors,
-        cache: cache,
-        with_column_names: with_column_names,
-        infer_schema_length: infer_schema_length,
-        n_rows: n_rows,
-        low_memory: low_memory,
-        rechunk: rechunk,
-        skip_rows_after_header: skip_rows_after_header,
-        encoding: encoding,
-        row_count_name: row_count_name,
-        row_count_offset: row_count_offset,
-        parse_dates: parse_dates,
-        eol_char: eol_char,
-      )
-    end
-
-    # Lazily read from an Arrow IPC (Feather v2) file or multiple files via glob patterns.
-    #
-    # This allows the query optimizer to push down predicates and projections to the scan
-    # level, thereby potentially reducing memory overhead.
-    #
-    # @param source [String]
-    #   Path to a IPC file.
-    # @param n_rows [Integer]
-    #   Stop reading from IPC file after reading `n_rows`.
-    # @param cache [Boolean]
-    #   Cache the result after reading.
-    # @param rechunk [Boolean]
-    #   Reallocate to contiguous memory when all chunks/ files are parsed.
-    # @param row_count_name [String]
-    #   If not nil, this will insert a row count column with give name into the
-    #   DataFrame.
-    # @param row_count_offset [Integer]
-    #   Offset to start the row_count column (only use if the name is set).
-    # @param storage_options [Hash]
-    #   Extra options that make sense for a particular storage connection.
-    # @param memory_map [Boolean]
-    #   Try to memory map the file. This can greatly improve performance on repeated
-    #   queries as the OS may cache pages.
-    #   Only uncompressed IPC files can be memory mapped.
-    #
-    # @return [LazyFrame]
-    def scan_ipc(
-      source,
-      n_rows: nil,
-      cache: true,
-      rechunk: true,
-      row_count_name: nil,
-      row_count_offset: 0,
-      storage_options: nil,
-      memory_map: true
-    )
-      LazyFrame._scan_ipc(
-        source,
-        n_rows: n_rows,
-        cache: cache,
-        rechunk: rechunk,
-        row_count_name: row_count_name,
-        row_count_offset: row_count_offset,
-        storage_options: storage_options,
-        memory_map: memory_map
-      )
-    end
-
-    # Lazily read from a parquet file or multiple files via glob patterns.
-    #
-    # This allows the query optimizer to push down predicates and projections to the scan
-    # level, thereby potentially reducing memory overhead.
-    #
-    # @param source [String]
-    #   Path to a file.
-    # @param n_rows [Integer]
-    #   Stop reading from parquet file after reading `n_rows`.
-    # @param cache [Boolean]
-    #   Cache the result after reading.
-    # @param parallel ["auto", "columns", "row_groups", "none"]
-    #   This determines the direction of parallelism. 'auto' will try to determine the
-    #   optimal direction.
-    # @param rechunk [Boolean]
-    #   In case of reading multiple files via a glob pattern rechunk the final DataFrame
-    #   into contiguous memory chunks.
-    # @param row_count_name [String]
-    #   If not nil, this will insert a row count column with give name into the
-    #   DataFrame.
-    # @param row_count_offset [Integer]
-    #   Offset to start the row_count column (only use if the name is set).
-    # @param storage_options [Hash]
-    #   Extra options that make sense for a particular storage connection.
-    # @param low_memory [Boolean]
-    #   Reduce memory pressure at the expense of performance.
-    #
-    # @return [LazyFrame]
-    def scan_parquet(
-      source,
-      n_rows: nil,
-      cache: true,
-      parallel: "auto",
-      rechunk: true,
-      row_count_name: nil,
-      row_count_offset: 0,
-      storage_options: nil,
-      low_memory: false
+      sample_size: 1024,
+      eol_char: "\n",
+      raise_if_empty: true,
+      truncate_ragged_lines: false,
+      decimal_comma: false,
+      glob: true
     )
-      if Utils.pathlike?(source)
-        source = Utils.normalise_filepath(source)
+      if Utils.pathlike?(file)
+        path = Utils.normalize_filepath(file)
+      else
+        path = nil
+        # if defined?(StringIO) && file.is_a?(StringIO)
+        #   file = file.string
+        # end
       end
 
-      LazyFrame._scan_parquet(
-        source,
-        n_rows:n_rows,
-        cache: cache,
-        parallel: parallel,
-        rechunk: rechunk,
-        row_count_name: row_count_name,
-        row_count_offset: row_count_offset,
-        storage_options: storage_options,
-        low_memory: low_memory
-      )
-    end
-
-    # Lazily read from a newline delimited JSON file.
-    #
-    # This allows the query optimizer to push down predicates and projections to the scan
-    # level, thereby potentially reducing memory overhead.
-    #
-    # @param source [String]
-    #   Path to a file.
-    # @param infer_schema_length [Integer]
-    #   Infer the schema length from the first `infer_schema_length` rows.
-    # @param batch_size [Integer]
-    #   Number of rows to read in each batch.
-    # @param n_rows [Integer]
-    #   Stop reading from JSON file after reading `n_rows`.
-    # @param low_memory [Boolean]
-    #   Reduce memory pressure at the expense of performance.
-    # @param rechunk [Boolean]
-    #   Reallocate to contiguous memory when all chunks/ files are parsed.
-    # @param row_count_name [String]
-    #   If not nil, this will insert a row count column with give name into the
-    #   DataFrame.
-    # @param row_count_offset [Integer]
-    #   Offset to start the row_count column (only use if the name is set).
-    #
-    # @return [LazyFrame]
-    def scan_ndjson(
-      source,
-      infer_schema_length: 100,
-      batch_size: 1024,
-      n_rows: nil,
-      low_memory: false,
-      rechunk: true,
-      row_count_name: nil,
-      row_count_offset: 0
-    )
-      if Utils.pathlike?(source)
-        source = Utils.normalise_filepath(source)
+      dtype_list = nil
+      dtype_slice = nil
+      if !dtypes.nil?
+        if dtypes.is_a?(Hash)
+          dtype_list = []
+          dtypes.each do|k, v|
+            dtype_list << [k, Utils.rb_type_to_dtype(v)]
+          end
+        elsif dtypes.is_a?(::Array)
+          dtype_slice = dtypes
+        else
+          raise ArgumentError, "dtype arg should be list or dict"
+        end
       end
 
-      LazyFrame._scan_ndjson(
-        source,
-        infer_schema_length: infer_schema_length,
-        batch_size: batch_size,
-        n_rows: n_rows,
-        low_memory: low_memory,
-        rechunk: rechunk,
-        row_count_name: row_count_name,
-        row_count_offset: row_count_offset,
-      )
-    end
+      processed_null_values = Utils._process_null_values(null_values)
 
-    # Read into a DataFrame from Apache Avro format.
-    #
-    # @param source [Object]
-    #   Path to a file or a file-like object.
-    # @param columns [Object]
-    #   Columns to select. Accepts a list of column indices (starting at zero) or a list
-    #   of column names.
-    # @param n_rows [Integer]
-    #   Stop reading from Apache Avro file after reading ``n_rows``.
-    #
-    # @return [DataFrame]
-    def read_avro(source, columns: nil, n_rows: nil)
-      if Utils.pathlike?(source)
-        source = Utils.normalise_filepath(source)
+      if columns.is_a?(::String)
+        columns = [columns]
       end
-
-      DataFrame._read_avro(source, n_rows: n_rows, columns: columns)
-    end
-
-    # Read into a DataFrame from Arrow IPC (Feather v2) file.
-    #
-    # @param source [Object]
-    #   Path to a file or a file-like object.
-    # @param columns [Object]
-    #   Columns to select. Accepts a list of column indices (starting at zero) or a list
-    #   of column names.
-    # @param n_rows [Integer]
-    #   Stop reading from IPC file after reading `n_rows`.
-    # @param memory_map [Boolean]
-    #   Try to memory map the file. This can greatly improve performance on repeated
-    #   queries as the OS may cache pages.
-    #   Only uncompressed IPC files can be memory mapped.
-    # @param storage_options [Hash]
-    #   Extra options that make sense for a particular storage connection.
-    # @param row_count_name [String]
-    #   If not nil, this will insert a row count column with give name into the
-    #   DataFrame.
-    # @param row_count_offset [Integer]
-    #   Offset to start the row_count column (only use if the name is set).
-    # @param rechunk [Boolean]
-    #   Make sure that all data is contiguous.
-    #
-    # @return [DataFrame]
-    def read_ipc(
-      source,
-      columns: nil,
-      n_rows: nil,
-      memory_map: true,
-      storage_options: nil,
-      row_count_name: nil,
-      row_count_offset: 0,
-      rechunk: true
-    )
-      storage_options ||= {}
-      _prepare_file_arg(source, **storage_options) do |data|
-        DataFrame._read_ipc(
-          data,
-          columns: columns,
+      if file.is_a?(::String) && file.include?("*")
+        dtypes_dict = nil
+        if !dtype_list.nil?
+          dtypes_dict = dtype_list.to_h
+        end
+        if !dtype_slice.nil?
+          raise ArgumentError, "cannot use glob patterns and unnamed dtypes as `dtypes` argument; Use dtypes: Mapping[str, Type[DataType]"
+        end
+        scan = scan_csv(
+          file,
+          has_header: has_header,
+          sep: sep,
+          comment_char: comment_char,
+          quote_char: quote_char,
+          skip_rows: skip_rows,
+          dtypes: dtypes_dict,
+          null_values: null_values,
+          missing_utf8_is_empty_string: missing_utf8_is_empty_string,
+          ignore_errors: ignore_errors,
+          infer_schema_length: infer_schema_length,
           n_rows: n_rows,
-          row_count_name: row_count_name,
-          row_count_offset: row_count_offset,
+          low_memory: low_memory,
           rechunk: rechunk,
-          memory_map: memory_map
-        )
-      end
-    end
-
-    # Read into a DataFrame from a parquet file.
-    #
-    # @param source [Object]
-    #   Path to a file or a file-like object.
-    # @param columns [Object]
-    #   Columns to select. Accepts a list of column indices (starting at zero) or a list
-    #   of column names.
-    # @param n_rows [Integer]
-    #   Stop reading from parquet file after reading `n_rows`.
-    # @param storage_options [Hash]
-    #   Extra options that make sense for a particular storage connection.
-    # @param parallel ["auto", "columns", "row_groups", "none"]
-    #   This determines the direction of parallelism. 'auto' will try to determine the
-    #   optimal direction.
-    # @param row_count_name [String]
-    #   If not nil, this will insert a row count column with give name into the
-    #   DataFrame.
-    # @param row_count_offset [Integer]
-    #   Offset to start the row_count column (only use if the name is set).
-    # @param low_memory [Boolean]
-    #   Reduce memory pressure at the expense of performance.
-    # @param use_statistics [Boolean]
-    #   Use statistics in the parquet to determine if pages
-    #   can be skipped from reading.
-    # @param rechunk [Boolean]
-    #   Make sure that all columns are contiguous in memory by
-    #   aggregating the chunks into a single array.
-    #
-    # @return [DataFrame]
-    #
-    # @note
-    #   This operation defaults to a `rechunk` operation at the end, meaning that
-    #   all data will be stored continuously in memory.
-    #   Set `rechunk: false` if you are benchmarking the parquet-reader. A `rechunk` is
-    #   an expensive operation.
-    def read_parquet(
-      source,
-      columns: nil,
-      n_rows: nil,
-      storage_options: nil,
-      parallel: "auto",
-      row_count_name: nil,
-      row_count_offset: 0,
-      low_memory: false,
-      use_statistics: true,
-      rechunk: true
-    )
-      _prepare_file_arg(source) do |data|
-        DataFrame._read_parquet(
-          data,
-          columns: columns,
-          n_rows: n_rows,
-          parallel: parallel,
+          skip_rows_after_header: skip_rows_after_header,
           row_count_name: row_count_name,
           row_count_offset: row_count_offset,
-          low_memory: low_memory,
-          use_statistics: use_statistics,
-          rechunk: rechunk
+          eol_char: eol_char,
+          truncate_ragged_lines: truncate_ragged_lines,
+          decimal_comma: decimal_comma,
+          glob: glob
         )
-      end
-    end
-
-    # Read into a DataFrame from a JSON file.
-    #
-    # @param source [Object]
-    #   Path to a file or a file-like object.
-    #
-    # @return [DataFrame]
-    def read_json(source)
-      DataFrame._read_json(source)
-    end
-
-    # Read into a DataFrame from a newline delimited JSON file.
-    #
-    # @param source [Object]
-    #   Path to a file or a file-like object.
-    #
-    # @return [DataFrame]
-    def read_ndjson(source)
-      DataFrame._read_ndjson(source)
-    end
-
-    # Read a SQL query into a DataFrame.
-    #
-    # @param query [Object]
-    #   ActiveRecord::Relation or ActiveRecord::Result.
-    # @param schema_overrides [Hash]
-    #   A hash mapping column names to dtypes, used to override the schema
-    #   inferred from the query.
-    #
-    # @return [DataFrame]
-    def read_database(query, schema_overrides: nil)
-      if !defined?(ActiveRecord)
-        raise Error, "Active Record not available"
-      end
-
-      result =
-        if query.is_a?(ActiveRecord::Result)
-          query
-        elsif query.is_a?(ActiveRecord::Relation)
-          query.connection.select_all(query.to_sql)
-        elsif query.is_a?(::String)
-          ActiveRecord::Base.connection.select_all(query)
+        if columns.nil?
+          return scan.collect
+        elsif is_str_sequence(columns, allow_str: false)
+          return scan.select(columns).collect
         else
-          raise ArgumentError, "Expected ActiveRecord::Relation, ActiveRecord::Result, or String"
+          raise ArgumentError, "cannot use glob patterns and integer based projection as `columns` argument; Use columns: List[str]"
         end
-
-      data = {}
-      schema_overrides = (schema_overrides || {}).transform_keys(&:to_s)
-
-      result.columns.each_with_index do |k, i|
-        column_type = result.column_types[i]
-
-        data[k] =
-          if column_type
-            result.rows.map { |r| column_type.deserialize(r[i]) }
-          else
-            result.rows.map { |r| r[i] }
-          end
-
-        polars_type =
-          case column_type&.type
-          when :binary
-            Binary
-          when :boolean
-            Boolean
-          when :date
-            Date
-          when :datetime, :timestamp
-            Datetime
-          when :decimal
-            Decimal
-          when :float
-            Float64
-          when :integer
-            Int64
-          when :string, :text
-            String
-          when :time
-            Time
-          # TODO fix issue with null
-          # when :json, :jsonb
-          #   Struct
-          end
-
-        schema_overrides[k] ||= polars_type if polars_type
       end
 
-      DataFrame.new(data, schema_overrides: schema_overrides)
-    end
-    alias_method :read_sql, :read_database
+      projection, columns = Utils.handle_projection_columns(columns)
 
-    # def read_excel
-    # end
+      rbdf =
+        RbDataFrame.read_csv(
+          file,
+          infer_schema_length,
+          batch_size,
+          has_header,
+          ignore_errors,
+          n_rows,
+          skip_rows,
+          projection,
+          sep,
+          rechunk,
+          columns,
+          encoding,
+          n_threads,
+          path,
+          dtype_list,
+          dtype_slice,
+          low_memory,
+          comment_char,
+          quote_char,
+          processed_null_values,
+          missing_utf8_is_empty_string,
+          parse_dates,
+          skip_rows_after_header,
+          Utils._prepare_row_count_args(row_count_name, row_count_offset),
+          sample_size,
+          eol_char,
+          raise_if_empty,
+          truncate_ragged_lines,
+          decimal_comma,
+          schema
+        )
+      Utils.wrap_df(rbdf)
+    end
 
     # Read a CSV file in batches.
     #
@@ -755,6 +397,8 @@ module Polars
     #   allocation needed.
     # @param eol_char [String]
     #   Single byte end of line character.
+    # @param truncate_ragged_lines [Boolean]
+    #   Truncate lines that are longer than the schema.
     #
     # @return [BatchedCsvReader]
     #
@@ -774,6 +418,7 @@ module Polars
       skip_rows: 0,
       dtypes: nil,
       null_values: nil,
+      missing_utf8_is_empty_string: false,
       ignore_errors: false,
       parse_dates: false,
       n_threads: nil,
@@ -787,7 +432,10 @@ module Polars
       row_count_name: nil,
       row_count_offset: 0,
       sample_size: 1024,
-      eol_char: "\n"
+      eol_char: "\n",
+      raise_if_empty: true,
+      truncate_ragged_lines: false,
+      decimal_comma: false
     )
       projection, columns = Utils.handle_projection_columns(columns)
 
@@ -813,6 +461,7 @@ module Polars
         skip_rows: skip_rows,
         dtypes: dtypes,
         null_values: null_values,
+        missing_utf8_is_empty_string: missing_utf8_is_empty_string,
         ignore_errors: ignore_errors,
         parse_dates: parse_dates,
         n_threads: n_threads,
@@ -827,36 +476,205 @@ module Polars
         row_count_offset: row_count_offset,
         sample_size: sample_size,
         eol_char: eol_char,
-        new_columns: new_columns
+        new_columns: new_columns,
+        raise_if_empty: raise_if_empty,
+        truncate_ragged_lines: truncate_ragged_lines,
+        decimal_comma: decimal_comma
       )
     end
 
-    # Get a schema of the IPC file without reading data.
+    # Lazily read from a CSV file or multiple files via glob patterns.
+    #
+    # This allows the query optimizer to push down predicates and
+    # projections to the scan level, thereby potentially reducing
+    # memory overhead.
     #
     # @param source [Object]
-    #   Path to a file or a file-like object.
+    #   Path to a file.
+    # @param has_header [Boolean]
+    #   Indicate if the first row of dataset is a header or not.
+    #   If set to false, column names will be autogenerated in the
+    #   following format: `column_x`, with `x` being an
+    #   enumeration over every column in the dataset starting at 1.
+    # @param sep [String]
+    #   Single byte character to use as delimiter in the file.
+    # @param comment_char [String]
+    #   Single byte character that indicates the start of a comment line,
+    #   for instance `#`.
+    # @param quote_char [String]
+    #   Single byte character used for csv quoting.
+    #   Set to None to turn off special handling and escaping of quotes.
+    # @param skip_rows [Integer]
+    #   Start reading after `skip_rows` lines. The header will be parsed at this
+    #   offset.
+    # @param dtypes [Object]
+    #   Overwrite dtypes during inference.
+    # @param null_values [Object]
+    #   Values to interpret as null values. You can provide a:
+    #
+    #   - `String`: All values equal to this string will be null.
+    #   - `Array`: All values equal to any string in this array will be null.
+    #   - `Hash`: A hash that maps column name to a null value string.
+    # @param ignore_errors [Boolean]
+    #   Try to keep reading lines if some lines yield errors.
+    #   First try `infer_schema_length: 0` to read all columns as
+    #   `:str` to check which values might cause an issue.
+    # @param cache [Boolean]
+    #   Cache the result after reading.
+    # @param with_column_names [Object]
+    #   Apply a function over the column names.
+    #   This can be used to update a schema just in time, thus before
+    #   scanning.
+    # @param infer_schema_length [Integer]
+    #   Maximum number of lines to read to infer schema.
+    #   If set to 0, all columns will be read as `:str`.
+    #   If set to `nil`, a full table scan will be done (slow).
+    # @param n_rows [Integer]
+    #   Stop reading from CSV file after reading `n_rows`.
+    # @param encoding ["utf8", "utf8-lossy"]
+    #   Lossy means that invalid utf8 values are replaced with `�`
+    #   characters.
+    # @param low_memory [Boolean]
+    #   Reduce memory usage in expense of performance.
+    # @param rechunk [Boolean]
+    #   Reallocate to contiguous memory when all chunks/ files are parsed.
+    # @param skip_rows_after_header [Integer]
+    #   Skip this number of rows when the header is parsed.
+    # @param row_count_name [String]
+    #   If not nil, this will insert a row count column with the given name into
+    #   the DataFrame.
+    # @param row_count_offset [Integer]
+    #   Offset to start the row_count column (only used if the name is set).
+    # @param parse_dates [Boolean]
+    #   Try to automatically parse dates. If this does not succeed,
+    #   the column remains of data type `:str`.
+    # @param eol_char [String]
+    #   Single byte end of line character.
+    # @param truncate_ragged_lines [Boolean]
+    #   Truncate lines that are longer than the schema.
     #
-    # @return [Hash]
-    def read_ipc_schema(source)
+    # @return [LazyFrame]
+    def scan_csv(
+      source,
+      has_header: true,
+      sep: ",",
+      comment_char: nil,
+      quote_char: '"',
+      skip_rows: 0,
+      dtypes: nil,
+      null_values: nil,
+      missing_utf8_is_empty_string: false,
+      ignore_errors: false,
+      cache: true,
+      with_column_names: nil,
+      infer_schema_length: 100,
+      n_rows: nil,
+      encoding: "utf8",
+      low_memory: false,
+      rechunk: true,
+      skip_rows_after_header: 0,
+      row_count_name: nil,
+      row_count_offset: 0,
+      parse_dates: false,
+      eol_char: "\n",
+      raise_if_empty: true,
+      truncate_ragged_lines: false,
+      decimal_comma: false,
+      glob: true
+    )
+      Utils._check_arg_is_1byte("sep", sep, false)
+      Utils._check_arg_is_1byte("comment_char", comment_char, false)
+      Utils._check_arg_is_1byte("quote_char", quote_char, true)
+
       if Utils.pathlike?(source)
-        source = Utils.normalise_filepath(source)
+        source = Utils.normalize_filepath(source)
       end
 
-      Plr.ipc_schema(source)
+      _scan_csv_impl(
+        source,
+        has_header: has_header,
+        sep: sep,
+        comment_char: comment_char,
+        quote_char: quote_char,
+        skip_rows: skip_rows,
+        dtypes: dtypes,
+        null_values: null_values,
+        ignore_errors: ignore_errors,
+        cache: cache,
+        with_column_names: with_column_names,
+        infer_schema_length: infer_schema_length,
+        n_rows: n_rows,
+        low_memory: low_memory,
+        rechunk: rechunk,
+        skip_rows_after_header: skip_rows_after_header,
+        encoding: encoding,
+        row_count_name: row_count_name,
+        row_count_offset: row_count_offset,
+        parse_dates: parse_dates,
+        eol_char: eol_char,
+        truncate_ragged_lines: truncate_ragged_lines
+      )
     end
 
-    # Get a schema of the Parquet file without reading data.
-    #
-    # @param source [Object]
-    #   Path to a file or a file-like object.
-    #
-    # @return [Hash]
-    def read_parquet_schema(source)
-      if Utils.pathlike?(source)
-        source = Utils.normalise_filepath(source)
+    # @private
+    def _scan_csv_impl(
+      file,
+      has_header: true,
+      sep: ",",
+      comment_char: nil,
+      quote_char: '"',
+      skip_rows: 0,
+      dtypes: nil,
+      null_values: nil,
+      ignore_errors: false,
+      cache: true,
+      with_column_names: nil,
+      infer_schema_length: 100,
+      n_rows: nil,
+      encoding: "utf8",
+      low_memory: false,
+      rechunk: true,
+      skip_rows_after_header: 0,
+      row_count_name: nil,
+      row_count_offset: 0,
+      parse_dates: false,
+      eol_char: "\n",
+      truncate_ragged_lines: true
+    )
+      dtype_list = nil
+      if !dtypes.nil?
+        dtype_list = []
+        dtypes.each do |k, v|
+          dtype_list << [k, Utils.rb_type_to_dtype(v)]
+        end
       end
-
-      Plr.parquet_schema(source)
+      processed_null_values = Utils._process_null_values(null_values)
+
+      rblf =
+        RbLazyFrame.new_from_csv(
+          file,
+          sep,
+          has_header,
+          ignore_errors,
+          skip_rows,
+          n_rows,
+          cache,
+          dtype_list,
+          low_memory,
+          comment_char,
+          quote_char,
+          processed_null_values,
+          infer_schema_length,
+          with_column_names,
+          rechunk,
+          skip_rows_after_header,
+          encoding,
+          Utils._prepare_row_count_args(row_count_name, row_count_offset),
+          parse_dates,
+          eol_char,
+          truncate_ragged_lines
+        )
+      Utils.wrap_ldf(rblf)
     end
 
     private