SAT-2978 - SDC Series binops corrupt operands if fill_value is used (#891)

* SAT-2978 - SDC Series binops corrupt operands if fill_value is used

This PR solves multiple issues in implementation of Series arithmetic and
comparison methods:
* fixes wrong filling behavior (e.g. np.nan + np.nan was 2 * fill_value)
* removes inplace=True operations that corrupt operands
* fixes broken scalar + series use cases
* fixes a sub case of common case when series indexes are equal, and hence
no alignment should happen
* adds performance tests for all different implementations
* moves operators and binop methods unit tests to a separate test suite
* unskips many tests incorrectly decorated with skip_parallel

* Fixing PEP and other remarks

Author: kozlov-alexey

buildscripts/autogen_sources.py

@@ -117,9 +117,17 @@
     imports_start_line, import_end_line = min(imports_line_numbers), max(imports_line_numbers)
     import_section_text = ''.join(module_text_lines[imports_start_line: import_end_line + 1])
 
-    # read function templates for arithmetic and comparison operators from templates module
+    # templates for arithmetic methods
+    template_func_binop_def = inspect.getsource(templates_module.sdc_binop)
+    template_func_binop_ovld = inspect.getsource(templates_module.sdc_binop_ovld)
     template_series_binop = inspect.getsource(templates_module.sdc_pandas_series_binop)
+
+    # templates for comparison methods
+    template_func_comp_binop_def = inspect.getsource(templates_module.sdc_comp_binop)
+    template_func_comp_binop_ovld = inspect.getsource(templates_module.sdc_comp_binop_ovld)
     template_series_comp_binop = inspect.getsource(templates_module.sdc_pandas_series_comp_binop)
+
+    # templates for operators
     template_series_operator = inspect.getsource(templates_module.sdc_pandas_series_operator_binop)
     template_series_comp_operator = inspect.getsource(templates_module.sdc_pandas_series_operator_comp_binop)
     template_str_arr_comp_binop = inspect.getsource(templates_module.sdc_str_arr_operator_comp_binop)
@@ -136,27 +144,35 @@
             # certaing modifications are needed to be applied for templates, so
             # verify correctness of produced code manually
             for name in arithmetic_binops_symbols:
-                func_text = template_series_binop.replace('binop', name)
+                func_text = template_func_binop_def.replace('binop', name)
+                file.write(f'\n\n{func_text}')
+                func_text = template_func_binop_ovld.replace('def ', f"@sdc_overload(sdc_{name})\ndef ", 1)
+                func_text = func_text.replace('binop', name)
                 func_text = func_text.replace(' + ', f' {arithmetic_binops_symbols[name]} ')
+                file.write(f'\n\n{func_text}')
+                func_text = template_series_binop.replace('binop', name)
                 func_text = func_text.replace('def ', f"@sdc_overload_method(SeriesType, '{name}')\ndef ", 1)
                 file.write(f'\n\n{func_text}')
 
             for name in comparison_binops_symbols:
-                func_text = template_series_comp_binop.replace('comp_binop', name)
+                func_text = template_func_comp_binop_def.replace('comp_binop', name)
+                file.write(f'\n\n{func_text}')
+                func_text = template_func_comp_binop_ovld.replace('def ', f"@sdc_overload(sdc_{name})\ndef ", 1)
+                func_text = func_text.replace('comp_binop', name)
                 func_text = func_text.replace(' < ', f' {comparison_binops_symbols[name]} ')
+                file.write(f'\n\n{func_text}')
+                func_text = template_series_comp_binop.replace('comp_binop', name)
                 func_text = func_text.replace('def ', f"@sdc_overload_method(SeriesType, '{name}')\ndef ", 1)
                 file.write(f'\n\n{func_text}')
 
             for name in arithmetic_binops_symbols:
                 if name != "div":
                     func_text = template_series_operator.replace('binop', name)
-                    func_text = func_text.replace(' + ', f' {arithmetic_binops_symbols[name]} ')
                     func_text = func_text.replace('def ', f'@sdc_overload(operator.{name})\ndef ', 1)
                     file.write(f'\n\n{func_text}')
 
             for name in comparison_binops_symbols:
                 func_text = template_series_comp_operator.replace('comp_binop', name)
-                func_text = func_text.replace(' < ', f' {comparison_binops_symbols[name]} ')
                 func_text = func_text.replace('def ', f'@sdc_overload(operator.{name})\ndef ', 1)
                 file.write(f'\n\n{func_text}')
 

sdc/datatypes/hpat_pandas_series_functions.py

@@ -4137,8 +4137,6 @@ def hpat_pandas_series_fillna_impl(self, value=None, method=None, axis=None, inp
     else:
         # non inplace implementations, copy array, fill the NA/NaN and return a new Series
         if isinstance(self.dtype, types.UnicodeType):
-            # For StringArrayType implementation is taken from _series_fillna_str_alloc_impl
-            # (can be called directly when it's index handling is fixed)
             def hpat_pandas_series_str_fillna_impl(self, value=None, method=None, axis=None,
                                                    inplace=False, limit=None, downcast=None):
                 return pandas.Series(data=numpy_like.fillna(self._data, inplace=inplace, value=value),

sdc/functions/numpy_like.py

@@ -602,7 +602,8 @@ def sdc_fillna_str_impl(self, inplace=False, value=None):
                         num_chars += get_utf8_size(s)
 
                 filled_data = pre_alloc_string_array(n, num_chars)
-                for i in prange(n):
+                # StringArray doesn't support parallel setitem, thus no prange here
+                for i in numpy.arange(n):
                     if sdc.hiframes.api.isna(self, i):
                         filled_data[i] = value
                     else: