enso-org/enso
1
1
Fork 0
mirror of https://github.com/enso-org/enso.git synced 2024-12-22 23:01:29 +03:00
Code Issues Projects Releases Wiki Activity

Improvements to Table.union (#9968)

Browse Source 
- Closes #9952
This commit is contained in:
Radosław Waśko 2024-05-22 11:38:10 +02:00 committed by GitHub
parent 517299bb09
commit 1e0649fda1
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
17 changed files with 772 additions and 515 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
CHANGELOG.md
View File

@ -666,6 +666,7 @@
- [Added `Text.cleanse` `Column.Text_Cleanse` and `Table.Text_Cleanse`][9879]
- [Added ability to save an existing Postgres connection as a Data Link in Enso
Cloud.][9957]
- [Improved `Table.union`.][9968]
[debug-shortcuts]:
https://github.com/enso-org/enso/blob/develop/app/gui/docs/product/shortcuts.md#debug
@ -978,6 +979,7 @@
[9873]: https://github.com/enso-org/enso/pull/9873
[9879]: https://github.com/enso-org/enso/pull/9879
[9957]: https://github.com/enso-org/enso/pull/9957
[9968]: https://github.com/enso-org/enso/pull/9968
#### Enso Compiler

181
distribution/lib/Standard/Database/0.0.0-dev/src/DB_Table.enso
View File

@ -18,6 +18,7 @@ from Standard.Base.Runtime import assert
from Standard.Base.Widget_Helpers import make_data_cleanse_vector_selector, make_delimiter_selector, make_format_chooser
import Standard.Table.Column_Operation.Column_Operation
import Standard.Table.Columns_To_Keep.Columns_To_Keep
import Standard.Table.Expression.Expression
import Standard.Table.Expression.Expression_Error
import Standard.Table.Internal.Add_Row_Number
@ -30,6 +31,7 @@ import Standard.Table.Internal.Problem_Builder.Problem_Builder
import Standard.Table.Internal.Replace_Helpers
import Standard.Table.Internal.Table_Helpers
import Standard.Table.Internal.Table_Helpers.Table_Column_Helper
import Standard.Table.Internal.Table_Helpers.Union_Result_Type
import Standard.Table.Internal.Table_Ref.Table_Ref
import Standard.Table.Internal.Unique_Name_Strategy.Unique_Name_Strategy
import Standard.Table.Internal.Value_Type_Helpers
@ -1719,90 +1721,77 @@ type DB_Table
- tables: A single table or a vector of tables to append to this one. The
tables are concatenated in the order they are specified, with `self`
being the first one.
- columns_to_keep: Specifies which columns to keep. Defaults to keeping
columns that are present in any of the tables, reporting a warning for
columns that are not present in all tables and adding `Nothing` values
for them.
- match_columns: Specifies how to match the columns.
- If `Match_Columns.By_Name` - the columns are matched by name across
all provided tables.
If unmatched columns are to be dropped, the resulting table will keep
only the set of columns that appear in all provided tables, in the
relative order that they appeared in the `self` table.
If unmatched columns are kept, they are added in the order of
appearance - i.e. first all columns from `self` will be added in the
original order, then any columns from the second table that were not
matched will be added at the end (preserving their relative order),
and so on for all the remaining tables.
- If `Match_Columns.By_Position` - the columns are mapped by position.
If unmatched columns are to be dropped, the resulting table will have
as many columns as the table that had the least columns and the
column names of the first table (`self`) will be used.
If unmatched columns are kept, the resulting table will have as many
columns as the table with the most columns. Since the first table may
not have all the necessary columns to provide column names for the
result, the result will have column names taken from the first table
that has the biggest number of columns.
- keep_unmatched_columns: If set to `True`, unmatched columns are kept
and are padded with `Nothing` for tables that did not have them.
If set to `False`, only the common subset of columns is kept - any
column that is not present in all tables is dropped. Defaults to
`Report_Unmatched`, which behaves like `True` - unmatched columns are
kept and padded with `Nothing`, but a problem is reported.
- allow_type_widening: Specifies if the resulting column type should be
adjusted to fit columns from all arguments. If `True`, a common type
will be chosen for each column (see "Unifying Column Types" below).
If `False`, the resulting column type will be the same as in the first
table containing the column. In this case, all columns that are
concatenated must have the same type as the first one (unless this
had a `Mixed` type - in which case it will accept any other types).
The names of each column come from the first table in which the given
column appears in.
The `List` option is not applicable when mapping columns by position.
Column names are taken from the first table if `In_All` and from the
first table that has the maximum number of columns if `In_Any`
- on_problems: Specifies how to handle problems if they occur, reporting
them as warnings by default.
- If `keep_unmatched_columns` is set to `Report_Unmatched` (the
default):
- If matching by name and there are columns that are not present in
all tables, `Unmatched_Columns` is reported.
- If matching by position and column counts of the merged tables
differ, then a `Column_Count_Mismatch` is reported. The error will
contain the greatest column count as its `expected` value and the
smallest one as its `actual` value.
- If `keep_unmatched_columns` is set to `False` and matching by name,
it is possible that there are no columns that are common to all
provided tables, in that case `No_Output_Columns` is thrown as a
dataflow error regardless of the `on_problems` setting, because there
are no columns to include in the resulting table.
- If type widening is disabled and one of corresponding columns has a
type that is incompatible with the type coming from the first table,
a `Column_Type_Mismatch` is reported. The problematic column will be
dropped from the resulting table. With type widening disabled, the
subsequent tables must have the same types as the first one, unless
the type of the first one was `Mixed` which will accept any other
type.
- If a common type coercion for a set of matched columns from
concatenated tables cannot be found, a `No_Common_Type` is reported.
In warning or ignore mode, the problematic column will be dropped
from the resulting table.
? Unifying Column Types
If `allow_type_widening` is set to `True`, then the following rules are
used to find a common type that will fit values from all merged tables.
Numeric columns are unified by finding the most general type that can
fit all of the columns. The biggest integer type will be chosen and if
Numeric columns are unified by finding the smallest type that can fit
all of the columns. The biggest integer type will be chosen and if
integers and decimals are mixed, the decimal type will be chosen.
If boolean columns are mixed with numeric columns, they will be coerced
to the numeric type (and converted to 0 and 1).
Text types will also be coerced according to the common rules - if
constant-length texts of different lengths are mixed, they will be
coerced to a varying-length type.
Text types will are also unified by finding the smallest type that can
fit all the values. If constant-length texts of different lengths are
mixed, they will be coerced to a varying-length type.
If date and date-time columns are unified, this yields a date-time
column. In-memory, the date is promoted by adding a time of 00:00 and
the system time-zone. In other backends that behaviour may differ.
If one of the matched columns has `Mixed` type, that type will be used
regardless of types of other columns. Mixing any other types will
result in a `No_Common_Type` problem. If columns of incompatible types
are meant to be mixed, at least one of them should be explicitly
retyped to the `Mixed` type to indicate that intention. Note that the
`Mixed` type may not be supported by most Database backends.
union : (DB_Table | Vector DB_Table) -> Match_Columns -> Boolean | Report_Unmatched -> Boolean -> Problem_Behavior -> DB_Table
union self tables:(DB_Table | Vector) match_columns=Match_Columns.By_Name keep_unmatched_columns=Report_Unmatched allow_type_widening=True on_problems=Report_Warning =
regardless of types of other columns. Note that the `Mixed` type may
not be supported by most Database backends.
Finally, if no common type is found using the rules above, everything
is converted to text.
? Problem Conditions
- If no common type is found and the text conversion fallback is used,
the `No_Common_Type` problem is reported.
- The `Float` type may not be able to exactly represent larger
integers, thus if such large integers are mixed with floats, the
resulting conversion to `Float` may cause a loss of precision.
In that case, a `Loss_Of_Integer_Precision` problem is reported.
This warning is only reported in the in-memory backend. Currently,
the Database backend proceeds without a warning about precision loss.
- If a column of dates is unified with a column of date-times, since
the assumption of using the midnight time-of-day is arbitrary,
a `Implicit_Date_As_Date_Time_Conversion` problem is reported.
- If an empty vector of tables is provided, an `Illegal_Argument` error
is raised.
- If `columns_to_keep` is set to `In_All` or `List` and an expected
column is missing in some of the tables, a `Unmatched_Columns`
problem is reported. If this causes the output to contain no columns,
a `No_Output_Columns` error is raised.
? Ordering of Columns in the result
When matching columns by name, it is possible that the ordering of
columns may vary between input tables. The ordering is determined as
following: columns that are kept from the first table are in the order
they appear in that table. If there are columns that do not appear in
the first table, they are appended to the end of the resulting table in
the order they appear in the input.
@tables (Widget.Vector_Editor item_editor=Widget.Code_Input item_default='_' display=Display.Always)
@columns_to_keep Columns_To_Keep.default_widget
union : (DB_Table | Vector DB_Table) -> Match_Columns -> Columns_To_Keep -> Problem_Behavior -> DB_Table
union self tables:(DB_Table | Vector) (columns_to_keep : Columns_To_Keep = ..In_Any_Warn_On_Missing) (match_columns : Match_Columns = Match_Columns.By_Name) (on_problems : Problem_Behavior = Report_Warning) =
all_tables = case tables of
v : Vector -> [self] + (v.map t-> DB_Table.from t)
single_table -> [self, single_table]
@ -1811,39 +1800,53 @@ type DB_Table
we only want to add a cause coming from unification; matching reports problems that would not fit this error.
problem_builder_for_matching = Problem_Builder.new
problem_builder_for_unification = Problem_Builder.new
matched_column_sets = Match_Columns_Helpers.match_columns all_tables match_columns keep_unmatched_columns problem_builder_for_matching
matched_column_sets = Match_Columns_Helpers.match_columns all_tables match_columns columns_to_keep problem_builder_for_matching
dialect = self.connection.dialect
type_mapping = dialect.get_type_mapping
merged_columns = matched_column_sets.map column_set->
case Table_Helpers.unify_result_type_for_union column_set all_tables allow_type_widening problem_builder_for_unification of
Nothing -> Nothing
result_type : Value_Type ->
sql_type = type_mapping.value_type_to_sql result_type Problem_Behavior.Report_Error
sql_type.catch Inexact_Type_Coercion error->
sql_type_from_value_type value_type =
type_mapping.value_type_to_sql value_type Problem_Behavior.Report_Error . catch Inexact_Type_Coercion error->
Panic.throw <|
Illegal_State.Error "Unexpected inexact type coercion in Union. The union logic should only operate in types supported by the given backend. This is a bug in the Database library. The coercion was: "+error.to_display_text cause=error
[column_set, sql_type, result_type]
good_columns = merged_columns.filter r-> r.is_nothing.not
case Table_Helpers.unify_result_type_for_union column_set all_tables problem_builder_for_unification of
Union_Result_Type.Common_Type common_type ->
[column_set, sql_type_from_value_type common_type, common_type]
Union_Result_Type.Fallback_To_Text ->
[column_set, sql_type_from_value_type Value_Type.Char, Value_Type.Char]
Union_Result_Type.No_Types_To_Unify ->
## If the column is all nulls, we still need to give it some type.
For DB `Mixed` is not available, so a portable type to use is `Char`.
[column_set, SQL_Type.null, Value_Type.Char]
problem_builder_for_matching.attach_problems_before on_problems <| problem_builder_for_unification.attach_problems_before on_problems <|
if good_columns.is_empty then problem_builder_for_unification.raise_no_output_columns_with_cause else
if merged_columns.is_empty then problem_builder_for_unification.raise_no_output_columns_with_cause else
queries = all_tables.map_with_index i-> t->
columns_to_select = good_columns.map description->
column_set = description.first
sql_type = description.second
columns_to_select = merged_columns.map description->
column_set = description.at 0
sql_type = description.at 1
result_type = description.at 2
column_name = column_set.name
## We assume that the type for this expression will never be queried - it is
just used internally to build the Union operation and never exposed externally.
infer_return_type _ = SQL_Type_Reference.null
case column_set.column_indices.at i of
corresponding_column_index : Integer ->
column = t.at corresponding_column_index
internal_named_column = column.as_internal.rename column_name
## We cast if the result type is different.
This is a bit on the safe side. In some cases the cast is not needed
(for example, most databases will allow union of int2 and int4 without casts; or SQLite does not need casts at all).
However, we do this for simplicity as determining the rules when the cast is needed or not is adding a lot of complication.
This is a possible future improvement to make queries lighter, but the benefit is unlikely to be worth it.
needs_cast = column.value_type != result_type
if needs_cast.not then internal_named_column else
dialect.make_cast internal_named_column sql_type infer_return_type
Nothing ->
typ = SQL_Type_Reference.from_constant SQL_Type.null
expr = SQL_Expression.Literal "NULL"
null_column = Internal_Column.Value column_name typ expr
## We assume that the type for this
expression will never be queried - it is
just used internally to build the Union
operation and never exposed externally.
infer_return_type _ = SQL_Type_Reference.null
if sql_type == SQL_Type.null then null_column else
dialect.make_cast null_column sql_type infer_return_type
corresponding_column_index : Integer ->
t.at corresponding_column_index . as_internal . rename column_name
pairs = columns_to_select.map c->
[c.name, c.expression]
Query.Select pairs t.context
@ -1860,7 +1863,7 @@ type DB_Table
See #6118.
infer_return_type expression =
SQL_Type_Reference.new self.connection new_ctx expression
new_columns = good_columns.map description->
new_columns = merged_columns.map description->
column_set = description.first
result_type = description.at 2
name = column_set.name

11
distribution/lib/Standard/Database/0.0.0-dev/src/Errors.enso
View File

@ -35,11 +35,18 @@ type SQL_Error
Convert the SQL error to a textual representation.
to_text : Text
to_text self =
query = if self.related_query.is_nothing.not then " [Query was: " + self.related_query.to_display_text + "]" else ""
query = if self.related_query.is_nothing then "" else
query_text = self.related_query.to_text
## Our generated queries tend to be very long, so to still be readable,
we don't shorten them too much. We impose an upper limit to avoid unbounded error message size.
max_length = 1000
shortened_query_text = if query_text.length <= max_length then query_text else
query_text.take (Index_Sub_Range.First (max_length.div 2)) + " (...) " + query_text.take (Index_Sub_Range.Last (max_length.div 2))
" [Query was: " + shortened_query_text + "]"
message = self.java_exception.getMessage
max_length = 300
short_message = if message.length < max_length then message else
message.take (Index_Sub_Range.First max_length/2) + " (...) " + message.take (Index_Sub_Range.Last max_length/2)
message.take (Index_Sub_Range.First (max_length.div 2)) + " (...) " + message.take (Index_Sub_Range.Last (max_length.div 2))
"There was an SQL error: " + short_message + "." + query
## PRIVATE

2
distribution/lib/Standard/Database/0.0.0-dev/src/Internal/Postgres/Postgres_Dialect.enso
View File

@ -140,7 +140,7 @@ type Postgres_Dialect
## PRIVATE
make_cast : Internal_Column -> SQL_Type -> (SQL_Expression -> SQL_Type_Reference) -> Internal_Column
make_cast self column target_type infer_result_type_from_database_callback =
make_cast self (column : Internal_Column) (target_type : SQL_Type) (infer_result_type_from_database_callback : SQL_Expression -> SQL_Type_Reference) =
mapping = self.get_type_mapping
source_type = mapping.sql_type_to_value_type column.sql_type_reference.get
target_value_type = mapping.sql_type_to_value_type target_type

2
distribution/lib/Standard/Database/0.0.0-dev/src/Internal/SQLite/SQLite_Dialect.enso
View File

@ -138,7 +138,7 @@ type SQLite_Dialect
## PRIVATE
make_cast : Internal_Column -> SQL_Type -> (SQL_Expression -> SQL_Type_Reference) -> Internal_Column
make_cast self column target_type infer_result_type_from_database_callback =
make_cast self (column : Internal_Column) (target_type : SQL_Type) (infer_result_type_from_database_callback : SQL_Expression -> SQL_Type_Reference) =
_ = [infer_result_type_from_database_callback]
mapping = self.get_type_mapping
target_value_type = mapping.sql_type_to_value_type target_type

35
distribution/lib/Standard/Table/0.0.0-dev/src/Columns_To_Keep.enso Normal file
View File

@ -0,0 +1,35 @@
from Standard.Base import Vector, Text
from Standard.Base.Metadata import make_single_choice, Widget
## Specifies which columns to keep in a union operation.
type Columns_To_Keep
## All columns are kept.
If a column is present only in some of the tables, it is padded with
`Nothing` for tables where it is missing.
In_Any
## Only columns that are present in all tables are kept.
If there are columns that are only present in some of the tables,
a problem is reported.
In_All
## Specific list of column names to keep.
If a table does not have a column that is specified in the list, it is
padded with `Nothing` and a problem is reported.
In_List (column_names : Vector Text)
## PRIVATE
Same as `In_Any`, but it will warn about columns that are not present in
all tables.
In_Any_Warn_On_Missing
## PRIVATE
The default widget for `Columns_To_Keep`.
It does not display the internal `In_Any_Warn_On_Missing` variant, since
that variant is only meant to be used as the default value.
default_widget -> Widget =
make_single_choice <|
["In_Any", "In_All", "In_List"].map c-> [c, ".."+c]

54
distribution/lib/Standard/Table/0.0.0-dev/src/Errors.enso
View File

@ -494,7 +494,8 @@ type Column_Type_Mismatch
type No_Common_Type
## PRIVATE
An error indicating that no common type could be found.
An error indicating that no common type could be found, and the operation
could not be performed.
Arguments:
- types: The types that were tried to be unified.
@ -502,6 +503,11 @@ type No_Common_Type
unified, if applicable.
Error (types : Vector Value_Type) (related_column_name : Nothing|Text)
## PRIVATE
A warning indicating that no common type could be found, so the operation
had to fall back to converting all values to text.
Warning_Convert_To_Text (types : Vector Value_Type) (related_column_name:Text)
## PRIVATE
Create a human-readable version of the error.
@ -509,11 +515,21 @@ type No_Common_Type
to_display_text self =
types = self.types.map .to_display_text . join ", "
prefix = "No common type was found for types: "+types
infix = case self.related_column_name of
column_name : Text -> " when unifying column ["+column_name+"]."
_ -> "."
suffix = " If you want to allow mixed types, please cast one of the columns to `Mixed` beforehand."
prefix + infix + suffix
location = case self.related_column_name of
column_name : Text -> " when unifying column ["+column_name+"]"
_ -> ""
suffix_type = case self of
No_Common_Type.Error _ _ -> "."
No_Common_Type.Warning_Convert_To_Text _ _ -> ", so the values were converted to text."
suffix_mixed = " If you want to have mixed types instead, please cast one of the columns to `Mixed` beforehand."
prefix + location + suffix_type + suffix_mixed
## PRIVATE
to_text self -> Text =
ctor = case self of
No_Common_Type.Error _ _ -> "Error"
No_Common_Type.Warning_Convert_To_Text _ _ -> "Warning_Convert_To_Text"
"No_Common_Type."+ctor+" "+self.types.to_text+" "+self.related_column_name.to_text
type Unmatched_Columns
## PRIVATE
@ -637,9 +653,11 @@ type Conversion_Failure
type Loss_Of_Integer_Precision
## PRIVATE
Indicates that an automatic conversion of an integer column to a decimal
Indicates that an automatic conversion of an Integer column to a Float
column is losing precision because some of the large integers cannot be
exactly represented by the `double` type.
exactly represented by the floating-point type.
Currently, this error is only reported in-memory.
Warning (affected_rows_count : Integer) (example_value : Integer) (example_value_converted : Float)
## PRIVATE
@ -834,3 +852,23 @@ type Nothing_Value_In_Filter_Condition
to_display_text : Text
to_display_text self =
"Using `Nothing` as an argument to a `"+self.filter_condition.to_text+"` cannot match anything."
## Indicates that different Date_Time (with or without timezone) or Date types
are mixed in the result, causing implicit coercions.
This is a warning, because using the `00:00` time and default time-zone may
not always be the expected choice, so the user should be aware of this.
type Mixing_Date_Time_Types
## PRIVATE
Date_To_Date_Time (related_column_name : Text | Nothing)
## PRIVATE
Implicit_Time_Zone (related_column_name : Text | Nothing)
to_display_text self -> Text =
location = if self.related_column_name.is_nothing then "" else " (in column ["+self.related_column_name+"])"
case self of
Mixing_Date_Time_Types.Date_To_Date_Time _ ->
"Mixing Date and Date_Time values"+location+": the Date values have been automatically converted to Date_Time by adding a time of 00:00 in the default time-zone."
Mixing_Date_Time_Types.Implicit_Time_Zone _ ->
"Mixing Date_Time values with and without timezone"+location+". A default timezone has been assumed where it was missing."

7
distribution/lib/Standard/Table/0.0.0-dev/src/Excel/Excel_Workbook.enso
View File

@ -12,6 +12,7 @@ from Standard.Base.Data.Filter_Condition import sql_like_to_regex
from Standard.Base.Metadata.Choice import Option
from Standard.Base.Metadata.Widget import Multiple_Choice, Single_Choice
import project.Columns_To_Keep.Columns_To_Keep
import project.Excel.Excel_Range.Excel_Range
import project.Headers.Headers
import project.Internal.Excel_Reader
@ -331,7 +332,7 @@ type Excel_Workbook
tables = sheet_names.map on_problems=on_problems address-> self.read address headers on_problems=on_problems
case return of
Return_As.Table_Of_Tables -> Table.new [["Sheet Name", sheet_names], ["Table", tables]]
Return_As.Merged_Table match ->
Return_As.Merged_Table columns_to_keep match ->
first_tbl = tables.find t-> t != Nothing
if first_tbl == Nothing then Error.throw (Illegal_Argument.Error "No valid sheets found.") else
unique = first_tbl.column_naming_helper.create_unique_name_strategy
@ -339,7 +340,7 @@ type Excel_Workbook
new_column_name = unique.make_unique "Sheet Name"
with_names = tables.zip sheet_names tbl->name-> if tbl == Nothing then Nothing else tbl.set name new_column_name . reorder_columns [new_column_name]
result = Table.from_union (with_names.filter Filter_Condition.Not_Nothing) match keep_unmatched_columns=True
result = Table.from_union (with_names.filter Filter_Condition.Not_Nothing) columns_to_keep=columns_to_keep match_columns=match
problem_builder = Problem_Builder.new
problem_builder.report_unique_name_strategy unique
@ -359,4 +360,4 @@ type Return_As
Table_Of_Tables
## All sheets are merged into a single table. A union operation is performed.
Merged_Table match:Match_Columns=Match_Columns.By_Name
Merged_Table (columns_to_keep : Columns_To_Keep = Columns_To_Keep.In_Any) (match : Match_Columns = Match_Columns.By_Name)

53
distribution/lib/Standard/Table/0.0.0-dev/src/Internal/Table_Helpers.enso
View File

@ -10,6 +10,7 @@ import project.Expression.Expression
import project.Internal.Column_Naming_Helper.Column_Naming_Helper
import project.Internal.Problem_Builder.Problem_Builder
import project.Internal.Value_Type_Helpers
import project.Match_Columns.Column_Set
import project.Position.Position
import project.Set_Mode.Set_Mode
import project.Sort_Column.Sort_Column
@ -530,30 +531,38 @@ is_column obj =
## PRIVATE
A helper method that resolves what should be the result type of a particular
column set based on the union settings.
unify_result_type_for_union column_set all_tables allow_type_widening problem_builder =
unify_result_type_for_union (column_set : Column_Set) (all_tables : Vector) (problem_builder : Problem_Builder) -> Union_Result_Type =
columns = column_set.resolve_columns all_tables
case allow_type_widening of
True ->
types = columns.filter Filter_Condition.Not_Nothing . map .value_type
. filter Filter_Condition.Not_Nothing
types = columns.map .value_type
if types.is_empty then Union_Result_Type.No_Types_To_Unify else
## First we check if we can find a generic common type.
This includes widening numeric column sizes, or converting Integer to Float.
common_type = Value_Type_Helpers.find_common_type types strict=True
if common_type.is_nothing then
problem_builder.report_other_warning (No_Common_Type.Error types related_column_name=column_set.name)
common_type
False ->
is_not_nothing c = case c of
Nothing -> False
_ -> True
first_column = columns.find is_not_nothing
first_type = first_column.value_type
if first_type == Value_Type.Mixed then Value_Type.Mixed else
first_wrong_column = columns.find if_missing=Nothing col->
is_not_nothing col && col.value_type != first_type
case first_wrong_column of
Nothing -> first_type
_ ->
got_type = first_wrong_column.value_type
problem_builder.report_other_warning (Column_Type_Mismatch.Error column_set.name first_type got_type)
Nothing
if common_type.is_nothing.not then Union_Result_Type.Common_Type common_type else
## Union has less strict requirements than other operations relying on `find_common_type`,
so if the common type was not found, we still check some fallbacks.
common_numeric_boolean = Value_Type_Helpers.find_common_numeric_boolean_type types
if common_numeric_boolean.is_nothing.not then Union_Result_Type.Common_Type common_numeric_boolean else
common_date_type = Value_Type_Helpers.find_common_date_types types column_set.name problem_builder
if common_date_type.is_nothing.not then Union_Result_Type.Common_Type common_date_type else
# Lastly, we fall back to text, reporting a warning.
problem_builder.report_other_warning (No_Common_Type.Warning_Convert_To_Text types column_set.name)
Union_Result_Type.Fallback_To_Text
## PRIVATE
type Union_Result_Type
## PRIVATE
Common_Type (value_type : Value_Type)
## PRIVATE
Fallback_To_Text
## PRIVATE
This case is returned if the requested column was missing from _all_ tables,
so there were no types to unify. An all-null column should be created.
No_Types_To_Unify
## PRIVATE
Replace a set of columns in the table with a new set of columns. The old

79
distribution/lib/Standard/Table/0.0.0-dev/src/Internal/Value_Type_Helpers.enso
View File

@ -1,13 +1,15 @@
from Standard.Base import all
import Standard.Base.Data.Vector.No_Wrap
import Standard.Base.Errors.Illegal_Argument.Illegal_Argument
from Standard.Base.Runtime import assert
import project.Column.Column
import project.Internal.Problem_Builder.Problem_Builder
import project.Internal.Storage
import project.Value_Type.Auto
import project.Value_Type.Bits
import project.Value_Type.Value_Type
from project.Errors import Invalid_Value_Type, No_Common_Type
from project.Errors import Invalid_Value_Type, Mixing_Date_Time_Types, No_Common_Type
from project.Internal.Table_Helpers import is_column
polyglot java import org.enso.base.polyglot.NumericConverter
@ -60,7 +62,7 @@ reconcile_types current new = case current of
Value_Type.Integer size -> case new of
Value_Type.Integer new_size ->
Value_Type.Integer (max_size size new_size)
Value_Type.Byte -> Value_Type.Integer size
Value_Type.Byte -> current
# If we unify integers with floats, we select the default Float 64 regardless of the input sizes.
Value_Type.Float _ -> Value_Type.Float
Value_Type.Decimal _ _ -> new
@ -75,8 +77,7 @@ reconcile_types current new = case current of
_ -> Value_Type.Mixed
Value_Type.Byte -> case new of
Value_Type.Byte -> Value_Type.Byte
Value_Type.Integer size ->
Value_Type.Integer size
Value_Type.Integer _ -> new
Value_Type.Float _ -> Value_Type.Float
Value_Type.Decimal _ _ -> new
_ -> Value_Type.Mixed
@ -89,9 +90,6 @@ reconcile_types current new = case current of
Value_Type.Byte -> Value_Type.Decimal precision scale
Value_Type.Float _ -> Value_Type.Float
_ -> Value_Type.Mixed
Value_Type.Boolean -> case new of
Value_Type.Boolean -> Value_Type.Boolean
_ -> Value_Type.Mixed
Value_Type.Char current_size current_variable -> case new of
Value_Type.Char new_size new_variable ->
result_variable = current_variable || new_variable || current_size != new_size
@ -118,19 +116,70 @@ max_size a b =
## PRIVATE
Finds the most specific value type that will fit all the provided types.
If `strict` is `True`, it is implemented as specified in the note
"Unifying Column Types" in `Table.union`. In that case, if no common type
is found, `Nothing` is returned.
It assumes that the `types` vector is not empty.
find_common_type : Vector Value_Type -> Boolean -> Value_Type | Nothing
find_common_type types strict =
Arguments:
- types: a vector of types to unify. It must not be empty.
- strict: A flag determining how strict the unification is.
If `False`, if no common type can be found, `Mixed` is used as a generic fallback.
If `True`, `Nothing` is returned if no common type can be found and `Mixed`
is only returned if any of the input types was already `Mixed`.
find_common_type (types : Vector Value_Type) (strict : Boolean) -> Value_Type | Nothing =
assert types.not_empty
most_generic_type = (types.drop 1).fold types.first reconcile_types
if strict.not || most_generic_type != Value_Type.Mixed then most_generic_type else
# Double check if Mixed was really allowed to come out.
## We return the Mixed type only if the input contained Mixed.
Otherwise we report failure to find common type.
if types.contains Value_Type.Mixed then Value_Type.Mixed else
Nothing
## PRIVATE
An extra helper function that reconciles Date_Time types with varying timezone
setting, as well as Date type, reporting any warnings.
It can be used as a fallback after `find_common_type` does not find a simple common type.
If non-date types are provided, it will fail by returning `Nothing`. It will not report any warnings in that case.
It assumes that the list of `types` is not empty.
find_common_date_types (types : Vector Value_Type) (related_column_name : Text | Nothing) (problem_builder : Problem_Builder) -> Value_Type | Nothing =
assert types.not_empty
all_date = types.all typ-> case typ of
Value_Type.Date -> True
Value_Type.Date_Time _ -> True
_ -> False
if all_date.not then Nothing else
has_date = types.contains Value_Type.Date
has_date_time_with_tz = types.contains (Value_Type.Date_Time True)
has_date_time_without_tz = types.contains (Value_Type.Date_Time False)
# The common type is the 'largest' one.
common_type = if has_date_time_with_tz then Value_Type.Date_Time True else
if has_date_time_without_tz then Value_Type.Date_Time False else
Value_Type.Date
if has_date && (common_type != Value_Type.Date) then
problem_builder.report_other_warning (Mixing_Date_Time_Types.Date_To_Date_Time related_column_name)
if has_date_time_without_tz && (common_type != Value_Type.Date_Time False) then
problem_builder.report_other_warning (Mixing_Date_Time_Types.Implicit_Time_Zone related_column_name)
common_type
## PRIVATE
An extra helper function that reconciles numeric and boolean types.
Unifying Boolean and numeric types is not expected by all operations, but
some may want to opt-in to it. This method allows to do so.
If non-numeric or non-boolean types are provided, it will fail by returning
`Nothing`.
No warnings are reported, as coercing boolean to integer is harmless, it was
just chosen not to be done by default.
find_common_numeric_boolean_type (types : Vector Value_Type) -> Value_Type | Nothing =
assert types.not_empty
all_numeric_or_boolean = types.all typ-> typ.is_numeric || (typ == Value_Type.Boolean)
if all_numeric_or_boolean.not then Nothing else
## We just find a common type again, ignoring the boolean types: the
boolean will fit any numeric type that we get out of this.
without_boolean = types.filter typ-> typ != Value_Type.Boolean
find_common_type without_boolean strict=True
## PRIVATE
Finds the type of an argument to a column operation.

87
distribution/lib/Standard/Table/0.0.0-dev/src/Match_Columns.enso
View File

@ -1,6 +1,9 @@
from Standard.Base import all
import Standard.Base.Errors.Illegal_Argument.Illegal_Argument
import Standard.Base.Runtime.State
from Standard.Base.Runtime import assert
import project.Columns_To_Keep.Columns_To_Keep
import project.Constants.Report_Unmatched
from project.Errors import Column_Count_Mismatch, No_Output_Columns, Unmatched_Columns
@ -31,21 +34,43 @@ type Match_Columns
columns should appear in the resulting table.
The method assumes at least one table is provided in its input.
match_columns tables matching_mode keep_unmatched_columns problem_builder = case matching_mode of
Match_Columns.By_Name -> case keep_unmatched_columns of
False ->
match_columns tables matching_mode columns_to_keep problem_builder =
assert tables.not_empty
case matching_mode of
Match_Columns.By_Name -> match_columns_by_name tables columns_to_keep problem_builder
Match_Columns.By_Position -> match_columns_by_position tables columns_to_keep problem_builder
## PRIVATE
match_columns_by_name tables columns_to_keep problem_builder = case columns_to_keep of
Columns_To_Keep.In_List list -> if list.is_empty then Error.throw (Illegal_Argument.Error "The list of columns to keep cannot be empty.") else
output_column_names = list.distinct
column_counts = find_column_counts tables
all_tables_count = tables.length
unmatched_column_names = output_column_names.filter name->
column_counts.get name 0 < all_tables_count
if unmatched_column_names.not_empty then
problem_builder.report_other_warning (Unmatched_Columns.Error unmatched_column_names)
build_column_set_by_name tables output_column_names
Columns_To_Keep.In_All ->
column_counts = find_column_counts tables
# This will only include columns that were present in all tables.
all_tables_count = tables.length
common_column_names = tables.first.column_names.filter name->
column_counts.at name == tables.length
column_counts.at name == all_tables_count
if common_column_names.is_empty then Error.throw (No_Output_Columns.Error "Unmatched columns are set to be dropped, but no common column names were found.") else
common_column_names.map name->
column_indices = tables.map table->
table.column_names.index_of name
Column_Set.Value name column_indices
dropped_column_names = tables.map .column_names
. flatten
. filter (name-> column_counts.at name < all_tables_count)
. distinct
if dropped_column_names.not_empty then
problem_builder.report_other_warning (Unmatched_Columns.Error dropped_column_names)
build_column_set_by_name tables common_column_names
_ ->
output_column_names = distinct_columns_in_appearance_order tables
if keep_unmatched_columns == Report_Unmatched then
report_missing = case columns_to_keep of
Columns_To_Keep.In_Any -> False
Columns_To_Keep.In_Any_Warn_On_Missing -> True
if report_missing then
column_counts = find_column_counts tables
all_tables_count = tables.length
## We iterate over output column names to get deterministic
@ -54,18 +79,48 @@ match_columns tables matching_mode keep_unmatched_columns problem_builder = case
column_counts.get name 0 < all_tables_count
if unmatched_column_names.not_empty then
problem_builder.report_other_warning (Unmatched_Columns.Error unmatched_column_names)
build_column_set_by_name tables output_column_names
## PRIVATE
Common logic for computing the final result of by-name matching.
Once the set of output column names is determined, we compute the
`Column_Set` by finding the corresponding column indices in each table (if found).
build_column_set_by_name tables output_column_names =
output_column_names.map name->
column_indices = tables.map table->
table.columns.index_of col-> col.name==name
# TODO this gets O(N^2), we should optimize
table.column_names.index_of name
Column_Set.Value name column_indices
Match_Columns.By_Position ->
## PRIVATE
match_columns_by_position tables columns_to_keep problem_builder = case columns_to_keep of
Columns_To_Keep.In_List _ ->
Error.throw (Illegal_Argument.Error "The In_List option for `columns_to_keep` cannot be used together with `By_Position` matching.")
_ ->
column_counts = tables.map table-> table.columns.length
minmax = column_counts.compute_bulk [Statistic.Minimum, Statistic.Maximum]
columns_to_take = if keep_unmatched_columns == False then minmax.first else minmax.second
if (minmax.first != minmax.second) && (keep_unmatched_columns == Report_Unmatched) then
problem_builder.report_other_warning (Column_Count_Mismatch.Error minmax.second minmax.first)
name_source = if keep_unmatched_columns == False then tables.first else
min = minmax.first
max = minmax.second
columns_to_take = case columns_to_keep of
Columns_To_Keep.In_All -> min
Columns_To_Keep.In_Any -> max
Columns_To_Keep.In_Any_Warn_On_Missing -> max
has_unmatched_columns = min != max
if has_unmatched_columns then
should_report_unmatched = case columns_to_keep of
Columns_To_Keep.In_All -> True
Columns_To_Keep.In_Any -> False
Columns_To_Keep.In_Any_Warn_On_Missing -> True
# TODO should we rephrase the wording of the error? should it depend on In_Any_Warn_On_Missing vs In_All?
if should_report_unmatched then
problem_builder.report_other_warning (Column_Count_Mismatch.Error max min)
name_source = case columns_to_keep of
Columns_To_Keep.In_All -> tables.first
_ ->
# We find the first table that has all the columns present.
tables.find table-> table.columns.length == columns_to_take
column_sets = Vector.new columns_to_take i->
name = name_source.at i . name
column_ids = tables.map table->
@ -79,7 +134,7 @@ type Column_Set
Value (name : Text) (column_indices : Vector Integer)
## PRIVATE
resolve_columns self all_tables = self.column_indices.zip all_tables i-> parent_table->
resolve_columns self (all_tables : Vector) = self.column_indices.zip all_tables i-> parent_table->
case i of
Nothing -> Nothing
_ : Integer -> parent_table.at i

281
distribution/lib/Standard/Table/0.0.0-dev/src/Table.enso
View File

@ -23,6 +23,7 @@ import project.Aggregate_Column.Aggregate_Column
import project.Blank_Selector.Blank_Selector
import project.Column.Column
import project.Column_Ref.Column_Ref
import project.Columns_To_Keep.Columns_To_Keep
import project.Constants.Previous_Value
import project.Constants.Report_Unmatched
import project.Data_Formatter.Data_Formatter
@ -49,7 +50,9 @@ import project.Internal.Replace_Helpers
import project.Internal.Split_Tokenize
import project.Internal.Table_Helpers
import project.Internal.Table_Helpers.Table_Column_Helper
import project.Internal.Table_Helpers.Union_Result_Type
import project.Internal.Table_Ref.Table_Ref
import project.Internal.Value_Type_Helpers
import project.Internal.Widget_Helpers
import project.Join_Condition.Join_Condition
import project.Join_Kind.Join_Kind
@ -2245,91 +2248,78 @@ type Table
- tables: A single table or a vector of tables to append to this one. The
tables are concatenated in the order they are specified, with `self`
being the first one.
- columns_to_keep: Specifies which columns to keep. Defaults to keeping
columns that are present in any of the tables, reporting a warning for
columns that are not present in all tables and adding `Nothing` values
for them.
- match_columns: Specifies how to match the columns.
- If `Match_Columns.By_Name` - the columns are matched by name across
all provided tables.
If unmatched columns are to be dropped, the resulting table will keep
only the set of columns that appear in all provided tables, in the
relative order that they appeared in the `self` table.
If unmatched columns are kept, they are added in the order of
appearance - i.e. first all columns from `self` will be added in the
original order, then any columns from the second table that were not
matched will be added at the end (preserving their relative order),
and so on for all the remaining tables.
- If `Match_Columns.By_Position` - the columns are mapped by position.
If unmatched columns are to be dropped, the resulting table will have
as many columns as the table that had the least columns and the
column names of the first table (self) will be used.
If unmatched columns are kept, the resulting table will have as many
columns as the table with the most columns. Since the first table may
not have all the necessary columns to provide column names for the
result, the result will have column names taken from the first table
that has the biggest number of columns.
- keep_unmatched_columns: If set to `True`, unmatched columns are kept
and are padded with `Nothing` for tables that did not have them.
If set to `False`, only the common subset of columns is kept - any
column that is not present in all tables is dropped. Defaults to
`Report_Unmatched`, which behaves like `True` - unmatched columns are
kept and padded with `Nothing`, but a problem is reported.
- allow_type_widening: Specifies if the resulting column type should be
adjusted to fit columns from all arguments. If `True`, a common type
will be chosen for each column (see "Unifying Column Types" below).
If `False`, the resulting column type will be the same as in the first
table containing the column. In this case, all columns that are
concatenated must have the same type as the first one (unless this
had a `Mixed` type - in which case it will accept any other types).
The names of each column come from the first table in which the given
column appears in.
The `List` option is not applicable when mapping columns by position.
Column names are taken from the first table if `In_All` and from the
first table that has the maximum number of columns if `In_Any`
- on_problems: Specifies how to handle problems if they occur, reporting
them as warnings by default.
- If `keep_unmatched_columns` is set to `Report_Unmatched` (the
default):
- If matching by name and there are columns that are not present in
all tables, `Unmatched_Columns` is reported.
- If matching by position and column counts of the merged tables
differ, then a `Column_Count_Mismatch` is reported. The error will
contain the greatest column count as its `expected` value and the
smallest one as its `actual` value.
- If `keep_unmatched_columns` is set to `False` and matching by name,
it is possible that there are no columns that are common to all
provided tables, in that case `No_Output_Columns` is thrown as a
dataflow error regardless of the `on_problems` setting, because there
are no columns to include in the resulting table.
- If type widening is disabled and one of corresponding columns has a
type that is incompatible with the type coming from the first table,
a `Column_Type_Mismatch` is reported. The problematic column will be
dropped from the resulting table. With type widening disabled, the
subsequent tables must have the same types as the first one, unless
the type of the first one was `Mixed` which will accept any other
type.
- If a common type coercion for a set of matched columns from
concatenated tables cannot be found, a `No_Common_Type` is reported.
In warning or ignore mode, the problematic column will be dropped
from the resulting table.
? Unifying Column Types
If `allow_type_widening` is set to `True`, then the following rules are
used to find a common type that will fit values from all merged tables.
Numeric columns are unified by finding the most general type that can
fit all of the columns. The biggest integer type will be chosen and if
Numeric columns are unified by finding the smallest type that can fit
all of the columns. The biggest integer type will be chosen and if
integers and decimals are mixed, the decimal type will be chosen.
If boolean columns are mixed with numeric columns, they will be coerced
to the numeric type (and converted to 0 and 1).
Text types will also be coerced according to the common rules - if
constant-length texts of different lengths are mixed, they will be
coerced to a varying-length type.
Text types will are also unified by finding the smallest type that can
fit all the values. If constant-length texts of different lengths are
mixed, they will be coerced to a varying-length type.
If date and date-time columns are unified, this yields a date-time
column. In-memory, the date is promoted by adding a time of 00:00 and
the system time-zone. In other backends that behaviour may differ.
If one of the matched columns has `Mixed` type, that type will be used
regardless of types of other columns. Mixing any other types will
result in a `No_Common_Type` problem. If columns of incompatible types
are meant to be mixed, at least one of them should be explicitly
retyped to the `Mixed` type to indicate that intention. Note that the
`Mixed` type may not be supported by most Database backends.
union : (Table | Vector Table) -> Match_Columns -> Boolean | Report_Unmatched -> Boolean -> Problem_Behavior -> Table
union self tables:(Table | Vector) match_columns=Match_Columns.By_Name keep_unmatched_columns=Report_Unmatched allow_type_widening=True on_problems=Report_Warning =
Table.from_union ([self] + Vector.unify_vector_or_element tables) match_columns keep_unmatched_columns allow_type_widening on_problems
regardless of types of other columns. Note that the `Mixed` type may
not be supported by most Database backends.
Finally, if no common type is found using the rules above, everything
is converted to text.
? Problem Conditions
- If no common type is found and the text conversion fallback is used,
the `No_Common_Type` problem is reported.
- The `Float` type may not be able to exactly represent larger
integers, thus if such large integers are mixed with floats, the
resulting conversion to `Float` may cause a loss of precision.
In that case, a `Loss_Of_Integer_Precision` problem is reported.
This warning is only reported in the in-memory backend. Currently,
the Database backend proceeds without a warning about precision loss.
- If a column of dates is unified with a column of date-times, since
the assumption of using the midnight time-of-day is arbitrary,
a `Implicit_Date_As_Date_Time_Conversion` problem is reported.
- If an empty vector of tables is provided, an `Illegal_Argument` error
is raised.
- If `columns_to_keep` is set to `In_All` or `List` and an expected
column is missing in some of the tables, a `Unmatched_Columns`
problem is reported. If this causes the output to contain no columns,
a `No_Output_Columns` error is raised.
? Ordering of Columns in the result
When matching columns by name, it is possible that the ordering of
columns may vary between input tables. The ordering is determined as
following: columns that are kept from the first table are in the order
they appear in that table. If there are columns that do not appear in
the first table, they are appended to the end of the resulting table in
the order they appear in the input.
@tables (Widget.Vector_Editor item_editor=Widget.Code_Input item_default='_' display=Display.Always)
@columns_to_keep Columns_To_Keep.default_widget
union : (Table | Vector Table) -> Columns_To_Keep -> Match_Columns -> Problem_Behavior -> Table
union self tables:(Table | Vector) (columns_to_keep : Columns_To_Keep = ..In_Any_Warn_On_Missing) (match_columns : Match_Columns = Match_Columns.By_Name) (on_problems : Problem_Behavior = Report_Warning) =
Table.from_union ([self] + Vector.unify_vector_or_element tables) columns_to_keep match_columns on_problems
## ALIAS drop_missing_rows, dropna
GROUP Standard.Base.Selections
@ -2953,108 +2943,97 @@ type Table
Arguments:
- tables: A vector of tables to union together. The
tables are concatenated in the order they are specified.
- columns_to_keep: Specifies which columns to keep. Defaults to keeping
columns that are present in any of the tables, reporting a warning for
columns that are not present in all tables and adding `Nothing` values
for them.
- match_columns: Specifies how to match the columns.
- If `Match_Columns.By_Name` - the columns are matched by name across
all provided tables.
If unmatched columns are to be dropped, the resulting table will keep
only the set of columns that appear in all provided tables, in the
relative order that they appeared in the `self` table.
If unmatched columns are kept, they are added in the order of
appearance - i.e. first all columns from `self` will be added in the
original order, then any columns from the second table that were not
matched will be added at the end (preserving their relative order),
and so on for all the remaining tables.
- If `Match_Columns.By_Position` - the columns are mapped by position.
If unmatched columns are to be dropped, the resulting table will have
as many columns as the table that had the least columns and the
column names of the first table (self) will be used.
If unmatched columns are kept, the resulting table will have as many
columns as the table with the most columns. Since the first table may
not have all the necessary columns to provide column names for the
result, the result will have column names taken from the first table
that has the biggest number of columns.
- keep_unmatched_columns: If set to `True`, unmatched columns are kept
and are padded with `Nothing` for tables that did not have them.
If set to `False`, only the common subset of columns is kept - any
column that is not present in all tables is dropped. Defaults to
`Report_Unmatched`, which behaves like `True` - unmatched columns are
kept and padded with `Nothing`, but a problem is reported.
- allow_type_widening: Specifies if the resulting column type should be
adjusted to fit columns from all arguments. If `True`, a common type
will be chosen for each column (see "Unifying Column Types" below).
If `False`, the resulting column type will be the same as in the first
table containing the column. In this case, all columns that are
concatenated must have the same type as the first one (unless this
had a `Mixed` type - in which case it will accept any other types).
The names of each column come from the first table in which the given
column appears in.
The `List` option is not applicable when mapping columns by position.
Column names are taken from the first table if `In_All` and from the
first table that has the maximum number of columns if `In_Any`.
- on_problems: Specifies how to handle problems if they occur, reporting
them as warnings by default.
- If `keep_unmatched_columns` is set to `Report_Unmatched` (the
default):
- If matching by name and there are columns that are not present in
all tables, `Unmatched_Columns` is reported.
- If matching by position and column counts of the merged tables
differ, then a `Column_Count_Mismatch` is reported. The error will
contain the greatest column count as its `expected` value and the
smallest one as its `actual` value.
- If `keep_unmatched_columns` is set to `False` and matching by name,
it is possible that there are no columns that are common to all
provided tables, in that case `No_Output_Columns` is thrown as a
dataflow error regardless of the `on_problems` setting, because there
are no columns to include in the resulting table.
- If type widening is disabled and one of corresponding columns has a
type that is incompatible with the type coming from the first table,
a `Column_Type_Mismatch` is reported. The problematic column will be
dropped from the resulting table. With type widening disabled, the
subsequent tables must have the same types as the first one, unless
the type of the first one was `Mixed` which will accept any other
type.
- If a common type coercion for a set of matched columns from
concatenated tables cannot be found, a `No_Common_Type` is reported.
In warning or ignore mode, the problematic column will be dropped
from the resulting table.
? Unifying Column Types
If `allow_type_widening` is set to `True`, then the following rules are
used to find a common type that will fit values from all merged tables.
Numeric columns are unified by finding the most general type that can
fit all of the columns. The biggest integer type will be chosen and if
Numeric columns are unified by finding the smallest type that can fit
all of the columns. The biggest integer type will be chosen and if
integers and decimals are mixed, the decimal type will be chosen.
If boolean columns are mixed with numeric columns, they will be coerced
to the numeric type (and converted to 0 and 1).
Text types will also be coerced according to the common rules - if
constant-length texts of different lengths are mixed, they will be
coerced to a varying-length type.
Text types will are also unified by finding the smallest type that can
fit all the values. If constant-length texts of different lengths are
mixed, they will be coerced to a varying-length type.
If date and date-time columns are unified, this yields a date-time
column. In-memory, the date is promoted by adding a time of 00:00 and
the system time-zone. In other backends that behaviour may differ.
If one of the matched columns has `Mixed` type, that type will be used
regardless of types of other columns. Mixing any other types will
result in a `No_Common_Type` problem. If columns of incompatible types
are meant to be mixed, at least one of them should be explicitly
retyped to the `Mixed` type to indicate that intention. Note that the
`Mixed` type may not be supported by most Database backends.
from_union : (Vector Table) -> Match_Columns -> Boolean | Report_Unmatched -> Boolean -> Problem_Behavior -> Table
from_union tables:(Vector) match_columns=Match_Columns.By_Name keep_unmatched_columns=Report_Unmatched allow_type_widening=True on_problems=Report_Warning =
regardless of types of other columns. Note that the `Mixed` type may
not be supported by most Database backends.
Finally, if no common type is found using the rules above, everything
is converted to text.
? Problem Conditions
- If no common type is found and the text conversion fallback is used,
the `No_Common_Type` problem is reported.
- The `Float` type may not be able to exactly represent larger
integers, thus if such large integers are mixed with floats, the
resulting conversion to `Float` may cause a loss of precision.
In that case, a `Loss_Of_Integer_Precision` problem is reported.
This warning is only reported in the in-memory backend. Currently,
the Database backend proceeds without a warning about precision loss.
- If a column of dates is unified with a column of date-times, since
the assumption of using the midnight time-of-day is arbitrary,
a `Implicit_Date_As_Date_Time_Conversion` problem is reported.
- If an empty vector of tables is provided, an `Illegal_Argument` error
is raised.
- If `columns_to_keep` is set to `In_All` or `List` and an expected
column is missing in some of the tables, a `Unmatched_Columns`
problem is reported. If this causes the output to contain no columns,
a `No_Output_Columns` error is raised.
? Ordering of Columns in the result
When matching columns by name, it is possible that the ordering of
columns may vary between input tables. The ordering is determined as
following: columns that are kept from the first table are in the order
they appear in that table. If there are columns that do not appear in
the first table, they are appended to the end of the resulting table in
the order they appear in the input.
@tables (Widget.Vector_Editor item_editor=Widget.Code_Input item_default='_' display=Display.Always)
@columns_to_keep Columns_To_Keep.default_widget
from_union : (Vector Table) -> Columns_To_Keep -> Match_Columns -> Problem_Behavior -> Table ! No_Output_Columns | Illegal_Argument
from_union (tables : Vector) (columns_to_keep : Columns_To_Keep = ..In_Any_Warn_On_Missing) (match_columns : Match_Columns = Match_Columns.By_Name) (on_problems : Problem_Behavior = Report_Warning) =
all_tables = (tables.map t-> Table.from t)
all_tables.if_not_error <|
if all_tables.is_empty then Error.throw (Illegal_Argument.Error "`Table.from_union` needs at least 1 input table.") else
## We keep separate problem builders, because if we are reporting `No_Output_Columns`,
we only want to add a cause coming from unification; matching reports problems that would not fit this error.
problem_builder_for_matching = Problem_Builder.new
problem_builder_for_unification = Problem_Builder.new
matched_column_sets = Match_Columns_Helpers.match_columns all_tables match_columns keep_unmatched_columns problem_builder_for_matching
matched_column_sets = Match_Columns_Helpers.match_columns all_tables match_columns columns_to_keep problem_builder_for_matching
result_row_count = all_tables.fold 0 c-> t-> c + t.row_count
merged_columns = matched_column_sets.map column_set->
case Table_Helpers.unify_result_type_for_union column_set all_tables allow_type_widening problem_builder_for_unification of
Nothing -> Nothing
result_type : Value_Type ->
concat_columns column_set all_tables result_type result_row_count on_problems
good_columns = merged_columns.filter Filter_Condition.Not_Nothing
case Table_Helpers.unify_result_type_for_union column_set all_tables problem_builder_for_unification of
Union_Result_Type.Common_Type common_type ->
concat_columns column_set all_tables common_type result_row_count needs_cast=False on_problems
Union_Result_Type.Fallback_To_Text ->
concat_columns column_set all_tables Value_Type.Char result_row_count needs_cast=True on_problems
Union_Result_Type.No_Types_To_Unify ->
Column.from_repeated_item column_set.name Nothing result_row_count
problem_builder_for_matching.attach_problems_before on_problems <|
problem_builder_for_unification.attach_problems_before on_problems <|
if good_columns.is_empty then problem_builder_for_unification.raise_no_output_columns_with_cause else
Table.new good_columns
if merged_columns.is_empty then problem_builder_for_unification.raise_no_output_columns_with_cause else
Table.new merged_columns
## PRIVATE
A helper to create a new table consisting of slices of the original table.
@ -3073,7 +3052,7 @@ make_join_helpers left_table right_table =
## PRIVATE
A helper that efficiently concatenates storages of in-memory columns.
concat_columns column_set all_tables result_type result_row_count on_problems =
concat_columns column_set all_tables result_type result_row_count needs_cast on_problems =
Java_Problems.with_problem_aggregator on_problems java_problem_aggregator->
storage_builder = make_storage_builder_for_type result_type on_problems initial_size=result_row_count java_problem_aggregator
column_set.column_indices.zip all_tables i-> parent_table->
@ -3082,7 +3061,9 @@ concat_columns column_set all_tables result_type result_row_count on_problems =
null_row_count = parent_table.row_count
storage_builder.appendNulls null_row_count
_ : Integer ->
storage = parent_table.at i . java_column . getStorage
column = parent_table.at i
converted = if needs_cast then column.cast result_type on_problems=Report_Error else column
storage = converted.java_column.getStorage
storage_builder.appendBulkStorage storage
sealed_storage = storage_builder.seal
Column.from_storage column_set.name sealed_storage

41
std-bits/table/src/main/java/org/enso/table/data/column/builder/DateTimeBuilder.java
View File

@ -1,11 +1,16 @@
package org.enso.table.data.column.builder;
import java.time.LocalDate;
import java.time.ZoneId;
import java.time.ZonedDateTime;
import org.enso.table.data.column.storage.Storage;
import org.enso.table.data.column.storage.datetime.DateStorage;
import org.enso.table.data.column.storage.datetime.DateTimeStorage;
import org.enso.table.data.column.storage.type.DateTimeType;
import org.enso.table.data.column.storage.type.DateType;
import org.enso.table.data.column.storage.type.StorageType;
import org.enso.table.error.ValueTypeMismatchException;
import org.graalvm.polyglot.Context;
/** A builder for string columns. */
public class DateTimeBuilder extends TypedBuilderImpl<ZonedDateTime> {
@ -23,6 +28,14 @@ public class DateTimeBuilder extends TypedBuilderImpl<ZonedDateTime> {
return DateTimeType.INSTANCE;
}
/**
* TODO DRY {@link org.enso.table.data.column.operation.cast.ToDateTimeStorageConverter}
* convertDate.
*/
private ZonedDateTime convertDate(LocalDate date) {
return date.atStartOfDay().atZone(ZoneId.systemDefault());
}
@Override
public void appendNoGrow(Object o) {
try {
@ -32,6 +45,34 @@ public class DateTimeBuilder extends TypedBuilderImpl<ZonedDateTime> {
}
}
@Override
public void appendBulkStorage(Storage<?> storage) {
if (storage.getType() instanceof DateType) {
if (storage instanceof DateStorage dateStorage) {
Context context = Context.getCurrent();
for (int i = 0; i < dateStorage.size(); ++i) {
LocalDate date = dateStorage.getItemBoxed(i);
if (date == null) {
data[currentSize++] = null;
} else {
data[currentSize++] = convertDate(date);
}
context.safepoint();
}
} else {
throw new IllegalStateException(
"Unexpected storage implementation for type "
+ storage.getType()
+ ": "
+ storage
+ ". This is a bug in the Table library.");
}
} else {
super.appendBulkStorage(storage);
}
}
@Override
public boolean accepts(Object o) {
return o instanceof ZonedDateTime;

11
std-bits/table/src/main/java/org/enso/table/data/column/builder/TypedBuilder.java
View File

@ -28,6 +28,15 @@ public abstract class TypedBuilder extends Builder {
*/
public abstract TypedBuilder retypeTo(StorageType type);
/** Specifies if the following object will be accepted by this builder's append* methods. */
/**
* Specifies if the following object will be accepted by this builder's append* methods.
*
* <p>This is used to determine if a given value can be appended to the current builder, or if it
* needs to be retyped to a more general one.
*
* <p>Note that the {@code appendBulkStorage} method may still accept more types than {@code
* accept}. This is exploited by operations like Union where more flexibility in merging column
* types is allowed than in building new columns from scratch.
*/
public abstract boolean accepts(Object o);
}

11
std-bits/table/src/main/java/org/enso/table/data/table/Column.java
View File

@ -171,16 +171,15 @@ public class Column {
Object converted = Polyglot_Utils.convertPolyglotValue(item);
Builder builder;
if (converted == null) {
builder = new MixedBuilder(repeat);
} else {
StorageType storageType = StorageType.forBoxedItem(converted);
builder = Builder.getForType(storageType, repeat, problemAggregator);
Builder builder = new MixedBuilder(repeat);
builder.appendNulls(repeat);
return new Column(name, builder.seal());
}
StorageType storageType = StorageType.forBoxedItem(converted);
Builder builder = Builder.getForType(storageType, repeat, problemAggregator);
Context context = Context.getCurrent();
for (int i = 0; i < repeat; i++) {
builder.appendNoGrow(converted);
context.safepoint();

348
test/Table_Tests/src/Common_Table_Operations/Join/Union_Spec.enso
View File

@ -9,7 +9,7 @@ from Standard.Database.Errors import Unsupported_Database_Operation, Integrity_E
from Standard.Test import all
from project.Common_Table_Operations.Util import expect_column_names, run_default_backend, within_table
from project.Common_Table_Operations.Util import all
import project.Util
main filter=Nothing = run_default_backend add_specs filter
@ -42,6 +42,11 @@ add_specs suite_builder setup =
suite_builder.group prefix+"Table.from_union" pending=db_pending group_builder->
run_union_tests group_builder setup call_static_union
group_builder.specify "should fail if no tables are provided" <|
r = Table.from_union []
r.should_fail_with Illegal_Argument
r.catch.to_display_text . should_contain "at least 1"
run_union_tests group_builder setup call_union =
create_connection_fn = setup.create_connection_func
data = Data.setup create_connection_fn
@ -90,37 +95,70 @@ run_union_tests group_builder setup call_union =
problems2 = [Unmatched_Columns.Error ["A", "D"]]
Problems.test_problem_handling action2 problems2 tester2
group_builder.specify "should drop unmatched columns if asked to" <|
group_builder.specify "should fill unmatched columns with nulls with no warning, if In_Any is explicitly chosen" <|
t1 = table_builder [["A", [1, 2, 3]], ["B", ["a", "b", "c"]]]
t2 = table_builder [["C", ["d", "e", "f"]], ["A", [4, 5, 6]]]
t3 = table_builder [["D", [Nothing, Nothing, 0]], ["C", ["g", "h", "i"]]]
table = call_union [t1, t2, t3] columns_to_keep=..In_Any on_problems=..Report_Error
Problems.assume_no_problems table
expect_column_names ["A", "B", "C", "D"] table
table.at "A" . to_vector . should_equal [1, 2, 3, 4, 5, 6, Nothing, Nothing, Nothing]
table.at "B" . to_vector . should_equal ["a", "b", "c", Nothing, Nothing, Nothing, Nothing, Nothing, Nothing]
table.at "C" . to_vector . should_equal [Nothing, Nothing, Nothing, "d", "e", "f", "g", "h", "i"]
table.at "D" . to_vector . should_equal [Nothing, Nothing, Nothing, Nothing, Nothing, Nothing, Nothing, Nothing, 0]
group_builder.specify "should drop unmatched columns and warn, if In_All is selected" <|
t1 = table_builder [["A", [1, 2, 3]], ["B", ["a", "b", "c"]]]
t2 = table_builder [["C", ["d", "e", "f"]], ["A", [4, 5, 6]]]
t3 = table_builder [["A", [Nothing, Nothing, 0]], ["C", ["g", "h", "i"]]]
t4 = call_union[t1, t2, t3] keep_unmatched_columns=False on_problems=Problem_Behavior.Report_Error
Problems.assume_no_problems t4
t4 = call_union [t1, t2, t3] columns_to_keep=..In_All
w = Problems.expect_only_warning Unmatched_Columns t4
w.column_names.should_equal ["B", "C"]
expect_column_names ["A"] t4
t4.at "A" . to_vector . should_equal [1, 2, 3, 4, 5, 6, Nothing, Nothing, 0]
group_builder.specify "should keep unmatched columns without errors if asked to" <|
t1 = table_builder [["A", [1, 2, 3]], ["B", ["a", "b", "c"]]]
t2 = table_builder [["C", ["d", "e", "f"]], ["A", [4, 5, 6]]]
t3 = table_builder [["A", [Nothing, Nothing, 0]], ["C", ["g", "h", "i"]]]
t4 = call_union [t1, t2, t3] keep_unmatched_columns=True on_problems=Problem_Behavior.Report_Error
Problems.assume_no_problems t4
expect_column_names ["A", "B", "C"] t4
t4.at "A" . to_vector . should_equal [1, 2, 3, 4, 5, 6, Nothing, Nothing, 0]
t4.at "B" . to_vector . should_equal ["a", "b", "c", Nothing, Nothing, Nothing, Nothing, Nothing, Nothing]
t4.at "C" . to_vector . should_equal [Nothing, Nothing, Nothing, "d", "e", "f", "g", "h", "i"]
group_builder.specify "should fail if asked to drop unmatched columns but the set of common columns is empty" <|
t1 = table_builder [["A", [1, 2, 3]], ["B", ["a", "b", "c"]]]
t2 = table_builder [["C", ["d", "e", "f"]], ["A", [4, 5, 6]]]
t3 = table_builder [["D", [Nothing, Nothing, 0]], ["C", ["g", "h", "i"]]]
t4 = call_union [t1, t2, t3] keep_unmatched_columns=False on_problems=Problem_Behavior.Ignore
t4 = call_union [t1, t2, t3] columns_to_keep=..In_All on_problems=..Ignore
t4.should_fail_with No_Output_Columns
t4.catch.to_display_text . should_equal "No columns in the result, because of another problem: Unmatched columns are set to be dropped, but no common column names were found."
group_builder.specify "should allow to select specified columns for union by In_List, using the ordering from the list" <|
t1 = table_builder [["A", [1]], ["X", [2]], ["B", ["a"]], ["Y", [3]]]
t2 = table_builder [["A", [4]], ["Z", [5]], ["B", ["b"]], ["X", [6]]]
t3 = call_union [t1, t2] columns_to_keep=(..In_List ["B", "A"])
expect_column_names ["B", "A"] t3
t3.at "B" . to_vector . should_equal ["a", "b"]
t3.at "A" . to_vector . should_equal [1, 4]
group_builder.specify "should add a Null column for unmatched columns from In_List" <|
t1 = table_builder [["A", [1]], ["X", [2]]]
t2 = table_builder [["Z", [4]], ["A", [5]]]
t3 = call_union [t1, t2] columns_to_keep=(..In_List ["B", "A"])
expect_column_names ["B", "A"] t3
t3.at "B" . to_vector . should_equal [Nothing, Nothing]
t3.at "A" . to_vector . should_equal [1, 5]
group_builder.specify "does not allow an empty list in In_List" <|
t1 = table_builder [["A", [1]], ["X", [2]]]
t2 = table_builder [["Z", [4]], ["A", [5]]]
r = call_union [t1, t2] columns_to_keep=(..In_List [])
r.should_fail_with Illegal_Argument
group_builder.specify "does not error if duplicate entries appear in the In_List" <|
t1 = table_builder [["A", [1]], ["X", [2]], ["B", ["a"]], ["Y", [3]]]
t2 = table_builder [["A", [4]], ["Z", [5]], ["B", ["b"]], ["X", [6]]]
t3 = call_union [t1, t2] columns_to_keep=(..In_List ["B", "B", "A", "A", "B"])
expect_column_names ["B", "A"] t3
group_builder.specify "should ignore column names when matching by position" <|
t1 = table_builder [["A", [1, 2, 3]], ["Y", ["a", "b", "c"]]]
t2 = table_builder [["X", [4, 5, 6]], ["A", ["d", "e", "f"]]]
@ -144,28 +182,36 @@ run_union_tests group_builder setup call_union =
problems = [Column_Count_Mismatch.Error 3 1]
Problems.test_problem_handling action problems tester
group_builder.specify "should keep the least number of columns with positional matching if asked to drop unmatched ones" <|
group_builder.specify "should keep the least number of columns with positional matching if In_All" <|
t1 = table_builder [["A", [1, 2, 3]], ["B", ["a", "b", "c"]]]
t2 = table_builder [["A1", [4, 5, 6]], ["B1", ["d", "e", "f"]], ["C", [7, 8, 9]]]
t3 = table_builder [["A2", [10, 20, 30]]]
t4 = call_union [t1, t2, t3] keep_unmatched_columns=False match_columns=Match_Columns.By_Position on_problems=Problem_Behavior.Report_Error
Problems.assume_no_problems t4
t4 = call_union [t1, t2, t3] columns_to_keep=..In_All match_columns=..By_Position
expect_column_names ["A"] t4
t4.at "A" . to_vector . should_equal [1, 2, 3, 4, 5, 6, 10, 20, 30]
w = Problems.expect_only_warning Column_Count_Mismatch t4
w.expected.should_equal 3
w.actual.should_equal 1
group_builder.specify "should keep the greatest number of columns with positional matching if asked to keep unmatched ones, filling missing values with null and reporting no problems" <|
group_builder.specify "should keep the greatest number of columns with positional matching if In_Any, reporting no problems" <|
t1 = table_builder [["A", [1, 2, 3]], ["B", ["a", "b", "c"]]]
t2 = table_builder [["A1", [4, 5, 6]], ["B1", ["d", "e", "f"]], ["C", [7, 8, 9]]]
t3 = table_builder [["A2", [10, 20, 30]]]
t4 = call_union [t1, t2, t3] match_columns=Match_Columns.By_Position keep_unmatched_columns=True on_problems=Problem_Behavior.Ignore
t4 = call_union [t1, t2, t3] columns_to_keep=..In_Any match_columns=..By_Position on_problems=..Report_Error
Problems.assume_no_problems t4
expect_column_names ["A1", "B1", "C"] t4
t4.at "A1" . to_vector . should_equal [1, 2, 3, 4, 5, 6, 10, 20, 30]
t4.at "B1" . to_vector . should_equal ["a", "b", "c", "d", "e", "f", Nothing, Nothing, Nothing]
t4.at "C" . to_vector . should_equal [Nothing, Nothing, Nothing, 7, 8, 9, Nothing, Nothing, Nothing]
group_builder.specify "does not allow In_List with positional matching" <|
t1 = table_builder [["A", [1, 2, 3]], ["B", ["a", "b", "c"]]]
t2 = table_builder [["A1", [4, 5, 6]], ["B1", ["d", "e", "f"]], ["C", [7, 8, 9]]]
r = call_union [t1, t2] columns_to_keep=(..In_List ["A", "B"]) match_columns=Match_Columns.By_Position
r.should_fail_with Illegal_Argument
group_builder.specify "should use column names from the first table that has enough columns in positional matching mode" <|
t1 = table_builder [["A", [1, 2, 3]]]
t2 = table_builder [["X", [4, 5, 6]], ["A", ["a", "b", "c"]]]
@ -180,10 +226,6 @@ run_union_tests group_builder setup call_union =
check t3
Problems.get_attached_warnings t3 . should_equal [Column_Count_Mismatch.Error 2 1]
t4 = call_union [t1, t2] match_columns=Match_Columns.By_Position keep_unmatched_columns=True
within_table t4 <|
check t4
t5 = table_builder [["Y", [7, 8, 9]], ["A", ["d", "e", "f"]], ["Z", [10, 11, 12]]]
t6 = table_builder [["W", [0]]]
t7 = table_builder [["X", [7, 8, 9]], ["Y", ["d", "e", "f"]], ["Z", [10, 11, 12]]]
@ -216,11 +258,13 @@ run_union_tests group_builder setup call_union =
check_same <| call_union [t1]
check_same <| call_union [t1] match_columns=Match_Columns.By_Position
check_same <| call_union [t1] keep_unmatched_columns=False
check_same <| call_union [t1] match_columns=Match_Columns.By_Position keep_unmatched_columns=False
check_same <| call_union [t1] columns_to_keep=..In_All
check_same <| call_union [t1] match_columns=Match_Columns.By_Position columns_to_keep=..In_All
check_same <| call_union [t1] keep_unmatched_columns=True
check_same <| call_union [t1] match_columns=Match_Columns.By_Position keep_unmatched_columns=True
check_same <| call_union [t1] columns_to_keep=..In_Any
check_same <| call_union [t1] match_columns=Match_Columns.By_Position columns_to_keep=..In_Any
check_same <| call_union [t1] columns_to_keep=(..In_List ["A", "B"])
group_builder.specify "should correctly unify text columns of various lengths" pending=(if setup.test_selection.fixed_length_text_columns.not then "Fixed-length Char columns are not supported by this backend.") <|
t1 = (table_builder [["A", ["a", "b", "c"]]]) . cast "A" (Value_Type.Char size=1 variable_length=False)
@ -231,12 +275,13 @@ run_union_tests group_builder setup call_union =
t3 = call_union [t1, t2]
expect_column_names ["A"] t3
Problems.assume_no_problems t3
t3.at "A" . to_vector . should_equal ["a", "b", "c", "xyz", "abc", "def"]
t3.at "A" . value_type . is_text . should_be_true
Test.with_clue "t3[A].value_type="+(t3.at "A").value_type.to_display_text+": " <|
t3.at "A" . value_type . variable_length . should_be_true
group_builder.specify "should find a common type that will fit the merged columns" <|
group_builder.specify "should find a common type that will fit the merged columns (Integer + Float)" <|
t1 = table_builder [["A", [0, 1, 2]]]
t2 = table_builder [["A", [1.0, 2.0, 2.5]]]
@ -245,29 +290,77 @@ run_union_tests group_builder setup call_union =
t3 = call_union [t1, t2]
expect_column_names ["A"] t3
Problems.assume_no_problems t3
t3.at "A" . value_type . is_floating_point . should_be_true
t3.at "A" . to_vector . should_equal [0, 1, 2, 1.0, 2.0, 2.5]
# Specific type tests that apply to in-memory. Database behaviour is up to implementation.
if setup.is_database.not then
t4 = table_builder [["A", [2^100, 2^10, 2]]]
t4.at "A" . value_type . should_be_a (Value_Type.Decimal ...)
group_builder.specify "should find a common type that will fit the merged columns (numeric + Boolean)" <|
t1 = table_builder [["A", [0, 1, 20]]]
t2 = table_builder [["A", [True, False, True]]]
t1.at "A" . value_type . is_integer . should_be_true
t2.at "A" . value_type . should_equal Value_Type.Boolean
t3 = call_union [t1, t2]
expect_column_names ["A"] t3
Problems.assume_no_problems t3
t3.at "A" . value_type . is_integer . should_be_true
t3.at "A" . to_vector . should_equal [0, 1, 20, 1, 0, 1]
t4 = table_builder [["A", [1.5, 0.0, 2.0]]]
t5 = call_union [t2, t4]
expect_column_names ["A"] t5
Problems.assume_no_problems t5
t5.at "A" . value_type . is_floating_point . should_be_true
t5.at "A" . to_vector . should_equal [1.0, 2.0, 2.5, 2^100, 2^10, 2]
t5.at "A" . to_vector . should_equal [1.0, 0.0, 1.0, 1.5, 0.0, 2.0]
t6 = call_union [t1, t4]
expect_column_names ["A"] t6
t6.at "A" . value_type . should_be_a (Value_Type.Decimal ...)
t6.at "A" . to_vector . should_equal [0, 1, 2, 2^100, 2^10, 2]
group_builder.specify "should warn about loss of precision when converting large Integer to Float" pending=(if setup.is_database then "Loss_Of_Integer_Precision not yet supported in DB.") <|
# 2^70 is not exactly representable as a Float.
t1 = table_builder [["A", [2^70, 2^10, 2]]]
t2 = table_builder [["A", [1.5, 2.0, 2.5]]]
t1.at "A" . value_type . is_decimal . should_be_true
t2.at "A" . value_type . is_floating_point . should_be_true
t3 = call_union [t1, t2]
expect_column_names ["A"] t3
w = Problems.expect_only_warning Loss_Of_Integer_Precision t3
# TODO should we try to include column name here for context? may be worth it...
w.affected_rows_count.should_equal 1
t3.at "A" . value_type . is_floating_point . should_be_true
t3.at "A" . to_vector . should_equal [(2^70).to_float, 2^10, 2, 1.5, 2.0, 2.5]
group_builder.specify "should find a common type (Integer and Char of different sizes)" <|
t1 = (table_builder [["X", [0, 1, 2]], ["Y", ['aa', 'bb', 'cc']]]) . cast "X" (Value_Type.Integer Bits.Bits_16) . cast "Y" (Value_Type.Char size=2 variable_length=False)
t2 = (table_builder [["X", [3, 4, 5]], ["Y", ['x', 'y', 'z']]]) . cast "X" (Value_Type.Integer Bits.Bits_32) . cast "Y" (Value_Type.Char size=1 variable_length=False)
supports_complex_types = (t1.is_error || t2.is_error || Problems.get_attached_warnings t1 . not_empty).not
case supports_complex_types of
False -> Nothing
True ->
t12 = call_union [t1, t2]
# No warnings are expected
Problems.assume_no_problems t12
t12.at "X" . value_type . should_equal (Value_Type.Integer Bits.Bits_32)
t12.at "Y" . value_type . should_equal (Value_Type.Char size=2 variable_length=True)
t12.at "X" . to_vector . should_equal [0, 1, 2, 3, 4, 5]
t12.at "Y" . to_vector . should_equal ['aa', 'bb', 'cc', 'x', 'y', 'z']
date_time_pending = if setup.test_selection.date_time.not then "Date/Time operations are not supported."
group_builder.specify "should warn when converting a Date to Date_Time" pending=date_time_pending <|
t1 = table_builder [["D", [Date_Time.new 2024 5 16 16 48 23]]]
t2 = table_builder [["D", [Date.new 2019 10 23, Date.new 2020]]]
action = call_union [t1, t2] on_problems=_
tester table =
expect_column_names ["D"] table
table.at "D" . value_type . should_equal Value_Type.Date_Time
table.at "D" . to_vector . should_equal_tz_agnostic [Date_Time.new 2024 5 16 16 48 23, Date_Time.new 2019 10 23 0 0 0, Date_Time.new 2020 1 1 0 0 0]
problems = [Mixing_Date_Time_Types.Date_To_Date_Time "D"]
problems.first.to_display_text . should_contain "[D]"
Problems.test_problem_handling action problems tester
# Database backends are not required to support Mixed types.
if setup.is_database.not then
group_builder.specify "should resort to Mixed value type only if at least one column is already Mixed" <|
## TODO currently no way to retype a column to Mixed, so we are
using a custom object
t1 = table_builder [["A", [1, 2, 3]], ["mixed", ["a", My_Type.Value 1 2, Nothing]]]
t2 = table_builder [["A", [4, 5, 6]], ["mixed", [1, 2, 3]]]
t1.at "mixed" . value_type . should_equal Value_Type.Mixed
@ -291,145 +384,68 @@ run_union_tests group_builder setup call_union =
t6.at "mixed" . to_vector . should_equal ["X", "y", "a", My_Type.Value 1 2, Nothing, 1, 2, 3, True, False]
t6.at "mixed" . value_type . should_equal Value_Type.Mixed
group_builder.specify "if no common type can be found, should report error and drop the problematic column" <|
group_builder.specify "if no common type can be found, will fall back to converting all types to text and warn" <|
t1 = table_builder [["A", [1, 2, 3]], ["B", ["a", "b", "c"]], ["C", [True, False, Nothing]]]
t2 = table_builder [["C", ["x", "Y", "Z"]], ["A", [4, 5, 6]], ["B", [1, 2, 3]]]
r1 = call_union [t1, t2] on_problems=Problem_Behavior.Report_Error
r1.should_fail_with No_Common_Type
r1.catch.to_display_text . should_contain "converted to text"
r2 = call_union [t1, t2] on_problems=Problem_Behavior.Ignore
Problems.assume_no_problems r2
action = call_union [t1, t2] on_problems=_
result_checker table =
expect_column_names ["A", "B", "C"] table
# If type was matched - the columns are merged as is:
table.at "A" . to_vector . should_equal [1, 2, 3, 4, 5, 6]
table.at "A" . value_type . is_integer . should_be_true
# If mixed, they are converted to text representation:
table.at "B" . to_vector . should_equal ["a", "b", "c", "1", "2", "3"]
table.at "B" . value_type . is_text . should_be_true
r3 = call_union [t1, t2] on_problems=Problem_Behavior.Report_Warning
w3 = Problems.get_attached_warnings r3
w3.each w-> w.should_be_a No_Common_Type
w3.map w->
v = table.at "C" . to_vector
# The check needs to be case insensitive because various database backends may represent Booleans with lower or uppercase.
v.take 2 . map (t -> t.to_case Case.Lower) . should_equal ["true", "false"]
# Nothing is preserved, not converted to text because we want to preserve the meaning of 'missing value':
v.drop 2 . should_equal [Nothing, "x", "Y", "Z"]
table.at "C" . value_type . is_text . should_be_true
error_checker result = result.should_fail_with No_Common_Type
warnings_checker warnings =
warnings.map w->
w.should_be_a No_Common_Type
w.to_display_text . should_contain "converted to text"
["B", "C"].should_contain w.related_column_name
## We look just at names of the Value_Type constructors, as
different database backends may choose integers of different
sizes and have differing settings for text types.
types = w.types.map value_type->
Meta.meta value_type . constructor . name
(types == ["Char", "Integer"]) || (types == ["Boolean", "Char"]) . should_be_true
Test.with_clue "(should be one of...) " <|
[["Char", "Integer"], ["Boolean", "Char"]].should_contain types
Problems.test_advanced_problem_handling action error_checker warnings_checker result_checker
# A boolean column cannot be merged with integers.
t3 = t1.select_columns ["C", "A"] reorder=True
t4 = t2.select_columns ["B", "A"] reorder=True
r4 = call_union [t3, t4] match_columns=Match_Columns.By_Position on_problems=Problem_Behavior.Report_Error
r4.should_fail_with No_Common_Type
group_builder.specify "if no common type can be found, will fall back to converting all types to text and warn (Date+Time)" pending=date_time_pending <|
t1 = table_builder [["D", [Time_Of_Day.new 12, Time_Of_Day.new 13, Time_Of_Day.new 14]]]
t2 = table_builder [["D", [Date.new 2019, Date.new 2020, Date.new 2021]]]
group_builder.specify "if type widening is not allowed, should use the type from first table that contained the given column" <|
t1 = table_builder [["A", [1, 2, 3]]]
t2 = table_builder [["A", [4, 5, 6]], ["B", [1.2, 2.2, 3.1]]]
t3 = call_union [t1, t2] allow_type_widening=False keep_unmatched_columns=True
within_table t3 <|
Problems.assume_no_problems t3
expect_column_names ["A", "B"] t3
t3.at "A" . to_vector . should_equal [1, 2, 3, 4, 5, 6]
t3.at "B" . to_vector . should_equal [Nothing, Nothing, Nothing, 1.2, 2.2, 3.1]
t3.at "A" . value_type . is_integer . should_be_true
t2.at "B" . value_type . is_floating_point . should_be_true
t3.at "B" . value_type . is_floating_point . should_be_true
group_builder.specify "if type widening is not allowed and types do not match, should report error and drop the problematic column" <|
t1 = table_builder [["A", [1, 2, 3]], ["B", [1, 2, 3]], ["E", [1.1, 2.5, 3.2]]]
t2 = table_builder [["A", [4, 5, 6]], ["B", [1.5, 2.5, 3.5]], ["E", [1, 2, 3]]]
t1.at "B" . value_type . is_integer . should_be_true
t1.at "E" . value_type . is_floating_point . should_be_true
t2.at "B" . value_type . is_floating_point . should_be_true
t2.at "E" . value_type . is_integer . should_be_true
action = call_union [t1, t2] allow_type_widening=False on_problems=_
action = call_union [t1, t2] on_problems=_
tester table =
expect_column_names ["A"] table
table.at "A" . to_vector . should_equal [1, 2, 3, 4, 5, 6]
expect_column_names ["D"] table
table.at "D" . to_vector . should_equal ["12:00:00", "13:00:00", "14:00:00", "2019-01-01", "2020-01-01", "2021-01-01"]
table.at "D" . value_type . is_text . should_be_true
problems = [No_Common_Type.Warning_Convert_To_Text [Value_Type.Time, Value_Type.Date] "D"]
Problems.test_problem_handling action problems tester
problem_checker problem =
problem.should_be_a Column_Type_Mismatch
True
err_checker err =
problem_checker err.catch
warn_checker warnings =
warnings.all problem_checker
Problems.test_advanced_problem_handling action err_checker warn_checker tester
# Database backends are not required to support Mixed types.
if setup.is_database.not then
group_builder.specify "even if type widening is not allowed, if the first column is mixed, it should accept any column to be concatenated to it" <|
t1 = table_builder [["X", ["a", 1, Nothing]]]
t2 = table_builder [["X", [1]]]
t3 = table_builder [["X", [1.2, 2.3, 3.4]]]
t4 = table_builder [["X", ["a", "b"]]]
t5 = table_builder [["X", [True, False]]]
t1.at "X" . value_type . should_equal Value_Type.Mixed
t2.at "X" . value_type . should_equal Value_Type.Integer
t6 = call_union [t1, t2, t3, t4, t5] allow_type_widening=False
Problems.assume_no_problems t6
t6.at "X" . value_type . should_equal Value_Type.Mixed
t6.at "X" . to_vector . should_equal ["a", 1, Nothing, 1, 1.2, 2.3, 3.4, "a", "b", True, False]
group_builder.specify "when finding a common type for numeric columns to be Float, any precision loss should be reported" <|
t1 = table_builder [["X", [1, (2^62)-1, 3]]]
t2 = table_builder [["X", [1.5, 2.5, 3.5]]]
t3 = table_builder [["X", [(2^100)+1, 2^10, 2]]]
t1.at "X" . value_type . should_equal Value_Type.Integer
t2.at "X" . value_type . should_equal Value_Type.Float
t3.at "X" . value_type . should_be_a (Value_Type.Decimal ...)
t4 = call_union [t2, t1, t3] allow_type_widening=True
t4.at "X" . value_type . should_equal Value_Type.Float
t4.at "X" . to_vector . should_equal [1.5, 2.5, 3.5, 1, (2^62)-1, 3, (2^100)+1 . to_float, 2^10, 2]
w = Problems.expect_only_warning Loss_Of_Integer_Precision t4
# Losing precision on (2^62)-1 and 2^100+1.
w.affected_rows_count . should_equal 2
group_builder.specify "if type mismatches cause all columns to be dropped, fail with No_Output_Columns" <|
t1 = table_builder [["A", [1, 2, 3]]]
t2 = table_builder [["A", ['x']]]
e3 = call_union [t1, t2] allow_type_widening=True on_problems=Problem_Behavior.Ignore
e3.should_fail_with No_Output_Columns
t4 = table_builder [["A", [1.5]]]
e5 = call_union [t1, t4] allow_type_widening=False on_problems=Problem_Behavior.Ignore
e5.should_fail_with No_Output_Columns
group_builder.specify "should find a common type (2)" <|
t1 = (table_builder [["X", [0, 1, 2]], ["Y", ['aa', 'bb', 'cc']]]) . cast "X" (Value_Type.Integer Bits.Bits_16) . cast "Y" (Value_Type.Char size=2 variable_length=False)
t2 = (table_builder [["X", [3, 4, 5]], ["Y", ['x', 'y', 'z']]]) . cast "X" (Value_Type.Integer Bits.Bits_32) . cast "Y" (Value_Type.Char size=1 variable_length=False)
supports_complex_types = (t1.is_error || t2.is_error || Problems.get_attached_warnings t1 . not_empty).not
case supports_complex_types of
False -> Nothing
True ->
t12 = call_union [t1, t2]
Problems.assume_no_problems t12
t12.at "X" . value_type . should_equal (Value_Type.Integer Bits.Bits_32)
t12.at "Y" . value_type . should_equal (Value_Type.Char size=2 variable_length=True)
t12.at "X" . to_vector . should_equal [0, 1, 2, 3, 4, 5]
t12.at "Y" . to_vector . should_equal ['aa', 'bb', 'cc', 'x', 'y', 'z']
group_builder.specify "should fail to find a common type if widening is not allowed (2)" <|
t1 = (table_builder [["X", [0, 1, 2]], ["Y", ['aa', 'bb', 'cc']]]) . cast "X" (Value_Type.Integer Bits.Bits_16) . cast "Y" (Value_Type.Char size=2 variable_length=False)
t2 = (table_builder [["X", [3, 4, 5]], ["Y", ['x', 'y', 'z']]]) . cast "X" (Value_Type.Integer Bits.Bits_32) . cast "Y" (Value_Type.Char size=1 variable_length=False)
supports_complex_types = (t1.is_error || t2.is_error || Problems.get_attached_warnings t1 . not_empty).not
case supports_complex_types of
False -> Nothing
True ->
r1 = call_union [t1, t2] allow_type_widening=False
r1.should_fail_with No_Output_Columns
r1.catch.cause . should_be_a Column_Type_Mismatch
r1.catch.to_display_text . should_equal "No columns in the result, because of another problem: The column [X] expects type Integer (16 bits) but one of the provided tables had type Integer (32 bits) which is not compatible with it."
# And this should report Column_Type_Mismatch as the more important error too:
call_union [t1, t2] allow_type_widening=False on_problems=Problem_Behavior.Report_Error . should_fail_with Column_Type_Mismatch
group_builder.specify "will use the _output_ column name in the warnings when matching by position (so input names may differ)" <|
t1 = table_builder [["A", [1]]]
t2 = table_builder [["B", ["a"]]]
r1 = call_union [t1, t2] match_columns=Match_Columns.By_Position
expect_column_names ["A"] r1
r1.at "A" . value_type . is_text . should_be_true
r1.at "A" . to_vector . should_equal ["1", "a"]
w = Problems.expect_only_warning No_Common_Type r1
w.related_column_name.should_equal "A"
group_builder.specify "should gracefully handle tables from different backends" <|
t1 = table_builder [["A", [1, 2, 3]], ["B", ["a", "b", "c"]]]

20
test/Table_Tests/src/In_Memory/Table_Spec.enso
View File

@ -822,13 +822,21 @@ add_specs suite_builder =
int = ["int", [1, 2, 3, 3]]
int2 = ["int2", [1, Nothing, Nothing, 1]]
dbl = ["dbl", [0.0, 0.0, Nothing, Nothing]]
dates = ["dates", [Date.new 2000, Date.new 1999 1 1, Date.new 1999 1 1, Date_Time.new 2022 8 20]]
dts = ["dts", [Date_Time.new 2022 8 27 11 22 25, Nothing, Date_Time.new 2030, Date.new 2000]]
tod = ["tod", [Time_Of_Day.new 18 00, Time_Of_Day.new 18 19, Date_Time.new 2000 1 1, Time_Of_Day.new 18 19]]
dates = ["dates", [Date.new 2000, Date.new 1999 1 1, Date.new 1999 1 1, Nothing]]
dts = ["dts", [Date_Time.new 2022 8 27 11 22 25, Nothing, Nothing, Date_Time.new 2030]]
dts_mixed = ["dts_mixed", [Date_Time.new 2022 8 27 11 22 25, Nothing, Date_Time.new 2030, Date.new 2000]]
tod = ["tod", [Time_Of_Day.new 18 00, Time_Of_Day.new 18 19, Nothing, Time_Of_Day.new 18 19]]
tod_mixed = ["tod_mixed", [Time_Of_Day.new 18 00, Time_Of_Day.new 18 19, Date_Time.new 2000 1 1, Time_Of_Day.new 18 19]]
mix = ["mix", [42, Date_Time.new 2022 8 27, 1, 1]]
nulls = ["nulls", [Nothing, Nothing, Nothing, 0]]
custom = ["custom", [2, My.Data 2 1, Nothing, Nothing]]
[str, int, int2, dbl, dates, dts, tod, mix, nulls, custom]
[str, int, int2, dbl, dates, dts, dts_mixed, tod, tod_mixed, mix, nulls, custom]
ins.at "dates" . value_type . should_equal Value_Type.Date
ins.at "dts" . value_type . is_date_time . should_be_true
ins.at "dts_mixed" . value_type . should_equal Value_Type.Mixed
ins.at "tod" . value_type . should_equal Value_Type.Time
ins.at "tod_mixed" . value_type . should_equal Value_Type.Mixed
data.varied_type_table.filter "strs" (Filter_Condition.Is_In (ins.at "str")) . at "strs" . to_vector . should_equal ["b", "c"]
data.varied_type_table.filter "strs" (Filter_Condition.Is_In (ins.at "str" . to_vector)) . at "strs" . to_vector . should_equal ["b", "c"]
data.varied_type_table.filter "ints" (Filter_Condition.Is_In (ins.at "int")) . at "ints" . to_vector . should_equal [1, 2]
@ -841,7 +849,11 @@ add_specs suite_builder =
data.varied_type_table.filter "dates" (Filter_Condition.Is_In (ins.at "dates" . to_vector)) . at "dates" . to_vector . should_equal [Date.new 2000, Date.new 1999 1 1]
data.varied_type_table.filter "datetimes" (Filter_Condition.Is_In (ins.at "dts")) . at "datetimes" . to_vector . should_equal [Date_Time.new 2022 8 27 11 22 25]
data.varied_type_table.filter "datetimes" (Filter_Condition.Is_In (ins.at "dts" . to_vector)) . at "datetimes" . to_vector . should_equal [Date_Time.new 2022 8 27 11 22 25]
# The Date_Time.new 2000 should not match with Date.new 2000 because the types are different:
data.varied_type_table.filter "datetimes" (Filter_Condition.Is_In (ins.at "dts_mixed")) . at "datetimes" . to_vector . should_equal [Date_Time.new 2022 8 27 11 22 25]
data.varied_type_table.filter "dates" (Filter_Condition.Is_In (ins.at "dts_mixed")) . at "dates" . to_vector . should_equal [Date.new 2000]
data.varied_type_table.filter "times" (Filter_Condition.Is_In (ins.at "tod")) . at "times" . to_vector . should_equal [Time_Of_Day.new 18 00]
data.varied_type_table.filter "times" (Filter_Condition.Is_In (ins.at "tod_mixed")) . at "times" . to_vector . should_equal [Time_Of_Day.new 18 00]
data.varied_type_table.filter "times" (Filter_Condition.Is_In (ins.at "tod" . to_vector)) . at "times" . to_vector . should_equal [Time_Of_Day.new 18 00]
data.varied_type_table.filter "mixed" (Filter_Condition.Is_In [42, "a", 1, Nothing, Date.new 2022 8 27, Date_Time.new 2022 8 27]) . at "mixed" . to_vector . should_equal [1, "a", Date.new 2022 8 27]
data.varied_type_table.filter "mixed" (Filter_Condition.Is_In (ins.at "mix")) . at "mixed" . to_vector . should_equal [1]