Did you know ... | Search Documentation: |
cql.pl -- CQL - Constraint Query Language |
Note that CQL is currently in a state of flux. Features may be dropped in future releases, and the generated SQL may change between releases. In particular, runtime mode is deprecated.
CQL is a Prolog interface to SQL databases. There are two modes: fully compiled and runtime. The fully compiled mode should be used if possible due to the far greater compile time checking it provides.
CQLv2 correctly compiles equality comparisons with NULL into the appropriate expression at runtime. In CQLv1, executing
A={null}, {[A], foo :: [a-A]}
would never succeed, regardless of the value of foo.a. This is no longer
the case: If A is {null}
then this will execute as SELECT .... WHERE
a IS NULL
and if A is not {null}, it will execute as SELECT ....
WHERE a = ?
See the section Removing null comparisions for the dealing with the common requirement to ignore comparisons with null.
It is generally not a good idea to wrap CQL inside a setof/3 or a bagof/3 ... unless you are prepared to declare all the CQL variables that are neither bound nor mentioned in the setof/bagof template. If you want to sort, use findall/3 followed by sort/2. Note that sort/2 (like setof/3) removes duplicates. If you don't want to remove duplicates, use msort/2.
In the course of executing a select query, the following rules are applied:
This is so we can handle outer joins. Consider this:
x :: [a-A] *== y :: [a-A]
Assume x.a binds A to a non-null value. If there is no matching row in
y
, then y.a = null
. If variable A was truly shared the query could
never succeed. By not binding the variable associated with y.a
the
query can succeed ( rule 1) and A will be bound to the value in x.a
.
Here is a simple example of a SQL SELECT from the table se_lt_x
test(A) :- format('About to call CQL with A=~w', [A]), {[], se_lt_x :: [a-A, b-B, c-C]}, format('B=~w, C=~w', [B, C]).
Comparisons can be done in-line e.g.
[a-'ELSTON_M']
or with the == operator e.g.
[a-A], A == 'ELSTON_M'.
The single = operator means unify, not compare. Use = for unification, not comparison
FIXME: Unification is deprecated.
The operators =:=
and \==
are also available for numerical value
comparisons (they just translate to SQL =
and <>
, so in fact you
could use them for string comparisons)
You can debug CQL using the meta-predicates ?/1, ??/2 and ???/3:
???{[], se_lt_x :: [a-A, b-_], A == 'ELSTON_M'}.
[main] CALL SELECT slx_2.b, slx_2.a FROM se_lt_x AS slx_2 WHERE slx_2.a = 'ELSTON_M' [main] EXIT SELECT slx_2.b, slx_2.a FROM se_lt_x AS slx_2 WHERE slx_2.a = 'ELSTON_M' (0.006963s, 0.01cpu, 3,899 inferences)
[main] CALL DECLARE @P0 VARCHAR(50); SET @P0 = 'ELSTON_M'; SELECT slx_450.b, slx_450.a FROM se_lt_x AS slx_450 WHERE slx_450.a = @P0 AND slx_450.a COLLATE Latin1_General_CS_AS = @P0 Result: se_lt_x.b = {null} se_lt_x.a = 'ELSTON_M' (0.003304s, 0.00cpu, 359 inferences)
[main] CALL SELECT slx_450.b, slx_450.a FROM se_lt_x AS slx_450 WHERE slx_450.a = 'ELSTON_M' Result: se_lt_x.b = {null} se_lt_x.a = 'ELSTON_M' (0.003304s, 0.00cpu, 359 inferences)
A Prolog variable can be simultaneously a SELECT variable, a JOIN variable and a WHERE variable as A is in the following example:
{[], se_lt_x :: [a-A, c-C] =*= se_lt_y :: [d-A, f-F], A == 'A4'}
which generates the following SQL
SELECT x_192.a, x_192.c, y_73.d, y_73.f FROM se_lt_x x_192 INNER JOIN se_lt_y y_73 ON y_73.d=x_192.a WHERE x_192.a = ? and y_73.d = ?
Note how all the variables referenced in the query are retrieved in the SELECT. This is done to make the query Prolog-like. This means the retrieved row should behave like a Prolog fact so that when a query succeeds all the variables become instantiated.
There is one notable exception however: WHERE variables and JOIN variables are not bound in aggregation selections
FIXME: Is this still the case?
sum_test :- {[], #se_lt_x :: [a-ValueA, sum(b)-Summation] =*= #se_lt_y :: [e-ValueB], ValueA == ValueB, % Explicit join point group_by([ValueA])}, writeln(ValueA-ValueB-Summation).
'ELSTON_M'-_G375971-99450 true ;
The following attributes are automatically provided i.e if the attribute is present in the table, CQL will automatically fill in the value:
All the special attributes can be overridden by supplying the attribute-value pair explicitly.
Rather than provide an abstract description of CQL syntax here is a set of examples that show how to use it.
{[], insert(se_lt_x, [a-'A', b-'B', c-100])}
{[], insert(se_lt_x, [a-'A', b-'B', c-100]), identity(I)}
{[], delete(se_lt_x, [x_pk-I])}
Note that the WHERE clause is part of the delete/2 term unlike update where the WHERE clause is defined outside the update/2 term. I could have made delete consisent with update, but this would have required the @ alias in the delete WHERE clause to identify the table where the rows are to be deleted). This seems like overkill because a delete can in fact refer to only one table anyway i.e. you can't identify rows to delete via a JOIN.
{[], se_lt_x :: [a-A, b-B]}
This query will either:
se_lt_x.b
where se_lt_x.a = A
se_lt_x.a
where se_lt_x.b = B
se_lt_x
{[], update(se_lt_x, [c-100]), @ :: [a-'A1'], row_count(N)}
This corresponds to UPDATE se_lt_x SET c=100 WHERE se_lt_x.a='A1'
.
The '@' is a special alias referring to the table that is being updated.
The row_count/1 term gives the number or rows updated.
{[], se_lt_x :: [a-A, c-C], C > 10}
{[], se_lt_x :: [a-J1, c-C] =*= se_lt_y :: [d-J1, f-F]}
The join is se_lt_x.a = se_lt_y.d
because of the shared variable J1.
se_lt_x.c
will be returned in C and se_lt_y.f
will be returned in
F
{[], update(se_lt_x, [c-(C + 2 * F + 20)]), @ :: [a-A, c-C] =*= se_lt_y :: [d-A, f-F], C < 100}
This joins the table being updated (table se_lt_x
) on table se_lt_y
where se_lt_x.a = se_lt_y.a
and where se_lt_x.c < 200
then updates
each identified row se_lt_x.c
with the specified expression.
\+ exists {[], se_lt_x :: [a-'Z']}
{[], se_lt_x :: [count(c)-C]}
This will count the rows in table se_lt_x
{[], se_lt_x :: [sum(c)-C]}
Sum the values of attribute c in table se_lt_x
{[], se_lt_x :: [avg(c)-C]}
Calculate the mean of the values of attribute c in table se_lt_x
{[], se_lt_x :: [max(c)-C]}
Calculate the maximum of the values of attribute c in table se_lt_x
{[], se_lt_x :: [min(c)-C]}
Calculate the minimum of the values of attribute c in table se_lt_x
{[], se_lt_z :: [g-G, sum(i)-I], group_by([G])}
This will generate the GROUP BY SQL
and sum se_lt_z.i
for each
value of se_lt_z.g
{[], se_lt_x :: [b-J1, a-A] =*= se_lt_z :: [h-J1, i-I, g-Z], I > min(Y, se_lt_y :: [f-Y, d-Z])}
The main query and the sub-query share variable Z. The generated SQL is:
SELECT x37.a, z4.i, z4.g FROM se_lt_x x37 INNER JOIN se_lt_z z4 ON x37.b=z4.h and z4.h=x37.b WHERE z4.i > (SELECT min(y11.f) FROM se_lt_y y11 WHERE z4.g=y11.d)
{[], se_lt_y :: [d-D,f-F], F < sum(I, se_lt_x :: [b-J1] =*= se_lt_z :: [h-J1, i-I])}
{[], se_lt_x :: [a-A, b-B], \+ exists se_lt_y :: [d-A]}
The generated SQL is:
SELECT x39.a, x39.b FROM se_lt_x x39 WHERE NOT EXISTS (SELECT * FROM se_lt_y y13 WHERE x39.a = y13.d)
An exists restriction translates to a WHERE
sub-query and is used to
say that "each row returned in the main query must satisfy some
condition expressed by another query".
Example
{[], se_lt_x :: [a-A, b-B], exists se_lt_y :: [d-A]}
compiles to:
SELECT x.b, x.a FROM se_lt_x x WHERE EXISTS (SELECT * FROM se_lt_y WHERE x.a = y.d)
se_lt_x :: [a-J1, b-B] *== se_lt_y :: [d-J1, e-E]}
CQL supports query restrictions based on lists. Note that in both cases \== [] and == [] are equivalent despite the obvious logical inconsistency.
FIXME: Can we make this behaviour be controlled by a flag? It IS quite useful, even if it is completely illogical
{[], se_lt_x :: [a-Bar], Bar == []}
and
{[], se_lt_x :: [a-Bar], Bar \== []}
both do exactly the same thing - they will not restrict the query based on Bar. The second case seems to be logically consistent - all things are not in the empty list.
If your list is bound at compile-time, you can simply use it as the attribute value in CQL, for example:
{[], se_lt_x :: [a-['ELSTON_M', 'LILLEY_N']]}
This does not require the list to be ground, merely bound. For example, this is not precluded:
foo(V1, V2):- {[], se_lt_x :: [a-[V1, V2]]}.
If, however, your list is not bound at compile-time, you must wrap the variable in list/1:
Bar = [a,b,c], {[], se_lt_x :: [bar-list(Bar)]}
If you write
foo(V1):- {[], se_lt_x :: [a-V1]}.
and at runtime call foo([value1])
, you will get a type error.
Remember: If the list of IN values is empty then no restriction is generated i.e.
{[], se_lt_x :: [a-[], b-B} is the exactly the same as {[], se_lt_x :: [b-B}
{[], se_lt_x :: [a-A, b-B, c-C], (C == 10 ; B == 'B2', C < 4)}
The generated SQL is:
SELECT x.a, x.b, x.c FROM se_lt_x x WHERE ((x.b = ? AND x.c < ?) OR x.c = ?)
{[], se_lt_x :: [a-A, c-C] =*= (se_lt_y :: [d-A] ; se_lt_z :: [g-A])}
The generated SQL is:
SELECT x43.c FROM (se_lt_x x43 INNER JOIN se_lt_z z6 ON x43.a=z6.g AND z6.g=x43.a) UNION SELECT x44.c FROM (se_lt_x x44 INNER JOIN se_lt_y y16 ON x44.a=y16.d AND y16.d=x44.a)
{[], (se_lt_x :: [a-A, c-10] ; se_lt_y :: [d-A, f-25])}
The output variable A is bound to the value from two different attributes and so the query is implemented as two separate ODBC queries
{[], se_lt_z :: [g-G, h-H], order_by([-G])}
The order_by specification is a list of "signed" variables. The example above will order by se_lt_z.g descending
Use distinct(ListOfVars)
to specify which attributes you want to be
distinct:
test_distinct :- findall(UserName, {[], se_lt_x :: [a-UserName, c-Key], Key >= 7, distinct([UserName])}, L), length(L, N), format('~w solutions~n', [N]). CALL : user:test_distinct/0 26 solutions EXIT : user:test_distinct/0 (0.098133s, 0.00cpu, 1,488 inferences)
{[], se_lt_z :: [i-I, j-J], J \== {null}}
{[], N = 3, se_lt_z :: [i-I], top(N), order_by([+I])}
This generates a TOP clause in SQL Server, and LIMIT clauses for PostgreSQL and SQLite
{[], se_lt_z :: [h-H, i-I1] =*= se_lt_z :: [h-H, i-I2], I1 \== I2}
Use the ignore_if_null wrapper in your CQL to 'filter out' null input values. This is a useful extension for creating user-designed searches.
{[], se_lt_x :: [a-UserName, b-ignore_if_null(SearchKey), ...]}
At runtime, if SearchKey is bound to a value other than {null} then the
query will contain WHERE ... b = ?
. If, however, SearchKey is bound
to {null}
, then this comparison will be omitted.
Disjunctions
In general, don't use ignore_if_null in disjunctions. Consider this query:
SearchKey = '%ELSTON%', {[], se_lt_x :: [a-UserName, b-RealName], ( RealName =~ SearchKey ; UserName =~ SearchKey)}
The query means "find a user where the UserName contains ELSTON OR the RealName contain ELSTON". If !SearchKey is {null} then RealName=~ {null} will fail, which is correct. If ignore_if_null was used, the test would succeed, which means the disjunction would always succeed i.e. the query would contain no restriction, which is clearly not the intended result. FIXME: Mike, what is this all about?
{[], se_lt_x :: [a-A, c-C] =*= se_lt_y :: [d-A, f-F] =*= se_lt_z :: [i-F, g-G]}
The shared variable A joins se_lt_x
and se_lt_y
; the shared variable
F joins se_lt_y
and se_lt_z
{[], se_lt_x :: [a-A, c-C] =*= #se_lt_y :: [d-A, f-F] =*= #se_lt_z :: [i-F, g-G]}
The hash operator indicates the table that should be accessed WITH (NOLOCK)
{[], se_lt_z :: [g-G, i-I], G =~ 'A_'}
The operator =~ means LIKE. If you are using PostgreSQL, it means ILIKE.
You can write an exception term directly to a varchar-type column in the database. Note that it will be rendered as text using ~p, and truncated if necessary - so you certainly can't read it out again and expect to get an exception! Example code:
catch(process_message(Message), Exception, {[], update(some_table, [status-'ERROR', status_comment-Exception]), @ :: [some_table_primary_key-PrimaryKey]}).
FIXME: This code is specific to my usage of CQL
You can pass the "N" is TOP N as a parameter (Subject to DBMS compatibility. This works in SQL Server 2005 and later, and PostgreSQL 9 (possibly earlier versions) and SQLite3.
N = 3, findall(I, {[], se_lt_z :: [i-I], top(N), order_by([+I])}, L)
You can include compile_time_goal(Goal)
in your CQL. If you specify a
module, it will be used, otherwise the goal will be called in the
current module. Note that the goal is executed in-order - if you want to
use the bindings in your CQL, you must put the compile_time_goal before
them.
Example 1
{[], se_lt_x :: [a-UserName, b-RealName, d-FavouriteColour], compile_time_goal(standard_batch_size_for_search(StandardBatchSize)), top(StandardBatchSize), order_by([+UserName]}
Example 2
excellent_colours(['RED', 'BLUE']). {[], se_lt_x :: [a-UserName, b-RealName, d-FavouriteColour], compile_time_goal(excellent_colours(Colours)), FavouriteColour == Colours}
CQL supports both constant and shared variable join specifications. This is particularly useful when specifying outer joins.
Example
{[], se_lt_x :: [a-UserNameA, b-RealName, d-FavouriteColour] *== se_lt_x :: [a-UserNameB, e-FavouriteFood] on( UserNameA == UserNameB, FavouriteColour == FavouriteFood, FavouriteFood == 'ORANGE')}
All the CQL comparison operators, <, =<, ==, =~, \=~, \==, >=, >
can
be used in ON specifications.
For example:
{[], se_lt_z :: [i-J1, k-K] *== se_lt_x :: [c-J1, a-A, b-B] on A \== 'A1'},
Expressions in WHERE restrictions are supported, for example:
{[], se_lt_n :: [i-I, j-J, k-K], J > 10 * (K / I) + 15},
To avoid accidentally deleting or updating all rows in a table CQL raises an exception if there is no WHERE restriction.
Sometimes however you really do need to delete or update all rows in a table.
To support this requirement in a disciplined way (and to avoid the creation of "dummy" WHERE restrictions) the keyword absence_of_where_restriction_is_deliberate has been added. For example:
{[], update(se_lt_x, [c-10]), @ :: [], absence_of_where_restriction_is_deliberate}
HAVING restrictions can be specified. For example:
{[], se_lt_z :: [sum(i)-I, g-G], group_by([G]), having(I > 30)}
For a description of HAVING see http://en.wikipedia.org/wiki/Having_(SQL)
There is one important difference between SQL HAVING and SQL WHERE clauses. The SQL WHERE clause condition is tested against each and every row of data, while the SQL HAVING clause condition is tested against the groups and/or aggregates specified in the SQL GROUP BY clause and/or the SQL SELECT column list.
INSERT and UPDATE values can be formatted in-line at runtime. For example:
Suffix = 'NOGG', cql_transaction(Schema, UserId, {[], insert(se_lt_x, [a-'A', b-'B', c-100, d-format('EGG_~w', [Suffix])])}),
will insert 'EGG_NOGG' into attribute 'd'.
You can specify negations in CQL WHERE clauses e.g.
{[], se_lt_z :: [g-G, h-H, i-I], \+((G == 'A1', H == 'B1' ; G == 'D1', H == 'B3'))},
Note that, just like in Prolog, \+ is a unary operator hence the "double" brackets in the example above.
It is possible to generate compile time attribute values by specifying a predicate which is executed when the CQL statement is compiled.
The predicate must return the value you want as its last argument. You specify the predicate where you would normally put the attribute value. The predicate is specified with its output argument missing.
Example - Using domain allowed values in a query.
In the following CQL statement the predicate cql_domain_allowed_value/3 is called within findall/3 at compile time to generate a list of domain values that restrict favourite_colour to be 'ORANGE' or 'PINK' or 'BLUE', or 'GREEN'.
colour('ORANGE'). colour('PINK'). colour('BLUE'). colour('GREEN'). {[], se_lt_x :: [d-findall(Value, permissible_colour(Value)), a-UserName]}
Note how findall/3 is actually called by specifying findall/2.
There is not much point using predicate-generated attribute values in compile-at-runtime CQL as you can always call the predicate to generate the required values outside the CQL statement.
INSERT from SELECT is supported:
Constant = 'MIKE', {[], insert(se_lt_x1, [x_pk-Pk, a-A, b-B, c-C, d-Constant]), se_lt_x :: [x_pk-Pk, a-A, b-B, c-C, as(d)-Constant]}
which generates the following SQL:
INSERT INTO se_lt_x1 (x_pk, a, b, c, d) SELECT se_lt_x_955.x_pk, se_lt_x_955.a, se_lt_x_955.b, se_lt_x_955.c, ? AS d FROM se_lt_x lt_x_955
Note the use of the as(d)
construct in the SELECT part of the CQL to
make the constant 'MIKE' appear to come from the SELECT thus setting
lt_x1.d
to 'MIKE' in every row inserted.
CQL provides a large number of hooks to fine-tune behaviour and allow for customization. These are:
cql_dependency_hook(+EntitySet, +Module)
can be defined to be
notified when a given Module references a list of database
entities. This can be used to manage metadata/code dependencycql_generated_sql_hook(+Filename, +LineNumber, +Goals)
can be
defined to examine generated SQL. Use cql_sql_clause(+Goals, -SQL,
-Parameters)
to examine the goalscql_index_suggestion_hook(+Index)
can be defined if you are
interested in proposed indices for your schema. Note that this is
not very mature (yet)cql_atomic_value_check_hook(+Value)
can be defined to declare
new 'atomic' types (That is, types which can be written directly to
the database), such as a representation like boolean(true)
for 1.cql_check_value_hook(+Value)
can be used to check that a value
is legalapplication_value_to_odbc_value_hook(+OdbcDataType, +Schema,
+TableName, +ColumnName, +Qualifiers, +ApplicationValue,
-OdbcValue)
.odbc_value_to_application_value_hook(+OdbcDataType, +Schema,
+TableName, +ColumnName, +Domain, +OdbcValue, -ApplicationValue)
.cql_access_token_hook(+AccessToken, -UserId)
can be defined to
map the generic 'AccessToken' passed to cql_transaction/3 to a user
ID. If not defined, the AccessToken is assumed to be the user ID.
This UserID is used in logging.cql_execution_hook(+Statement, +OdbcParameters,
+OdbcParameterDataTypes, -Row)
can be defined if you want to
implement the exeuction yourself (for example, to add extra
debugging)cql_log_hook(+Topics, +Level, +Format, +Args)
can be defined
to redirect CQL logging.
debug(deadlocks)
]sql_gripe_hook(+Level, +Format, +Args)
is called when
suspect SQL is found by the SQL parsercql_normalize_atom_hook(+DBMS, +ApplciationAtom, -DBMSAtom)
can be used to create a map for atoms in a specific DBMS. For
example, your schema may have arbitrarily long table names, but
your DBMS may only allow names up to 64 bytes long. In this case,
you can create a scheme for mapping the application-level atom to
the DBMS. Other uses include deleting or normalizing illegal
characters in namescql_error_hook(+ErrorId, +Format, +Args)
can be defined to
generate a specific exception term from the given arguments. If not
defined (or if it does not throw an exception, or fails), you will
get cql_error(ErrorId, FormattedMessage)
.cql_max_db_connections_hook(-Max)
can be defined to limit the number of simultaneous connections each
thread will attempt to haveodbc_connection_complete_hook(+Schema, +Details, +Connection)
can be hooked if you want to know every time a connection is madecql_transaction_info_hook(+AccessToken, +Connection, +DBMS, +Goal, -Info)
can be defined if you want to define any application-defined
information on a per-transaction level. This can be recovered via
database_transaction_query_info(?ThreadId, ?Goal, ?Info).These define the schema. You MUST either define them, or include library(cql/cql_autoschema) and add two directives to build the schema automatically:
register_database_connection_details(+Schema, +ConnectionInfo)
.build_schema(+Schema)
.Otherwise, you need to define at least default_schema/1 and cql:dbms/2, and then as many of the other facts as needed for your schema.
default_schema(-Schema)
MUST be defined. CQL autoschema will define this for you if you use
it.dbms(+Schema, -DBMS)
MUST be defined for every schema you use. CQL autoschema will define
this for you if you use it. DBMS must be one of the following:
odbc_data_type(+Schema, +TableName, +ColumnName, +OdbcDataType)
.primary_column_name(+Schema, +Tablename, +ColumnName)
.allows_nulls(true/false)
, +IsIdentity:is_identity(true/false)
, +ColumnDefault).database_domain(+DomainName, +OdbcDataType)
.routine_return_type(+Schema, +RoutineName, +OdbcDataType)
.database_constraint(+Schema, +EntityName, +ConstraintName, +Constraint)
.CQL provides hooks for maintaining detailed history of data in the database.
The hook predicates are:
cql_event_notification_table(+Schema, +TableName)
cql_history_attribute(+Schema, +TableName, +ColumnName)
cql_update_history_hook(+Schema, +TableName, +ColumnName, +PrimaryKeyColumnName, +PrimaryKeyValue, +ApplicationValueBefore, +ApplicationValueAfter, +AccessToken, +TransactionId, +TransactionTimestamp, +ThreadId, +Connection, +Goal)
.process_database_events(+Events)
Event Processing and History recording can be suppressed for a particular update/insert/delete statement by including the _no_state_change_actions_9 directive.
For example
{[], update(se_lt_x, [f-'LILAC'] @ :: [a-'ELSTON_M'], no_state_change_actions, % Don't want history to record this change row_count(RowCount)}
CQL has hooks to enable in-memory statistics to be tracked for database tables. Using this hook, it's possible to monitor the number of rows in a table with a particular value in a particular column.
Often the kind of statistics of interest are 'how many rows in this table are in ERROR' or 'how many in this table are at NEW'? While it may be possible to maintain these directly in any code which updates tables, it can be difficult to ensure all cases are accounted for, and requires developers to remember which attributes are tracked.
To ensure that all (CQL-originated) updates to statuses are captured, it's possible to use the CQL hook system to update them automatically. Define add a fact like:
cql_statistic_monitored_attribute_hook(my_schema, my_table, my_table_status_column).
This will examine the domain for the column 'my_table_status_column',
and generate a statistic for each of
my_table::my_table_status_column(xxx)
, where xxx is each possible
allowed value for the domain. Code will be automatically generated to
trap updates to this specific column, and maintain the state. This way,
if you are interested in the number of rows in my_table which have a
status of 'NEW', you can look at
my_table::my_table_status_column('NEW')
, without having to manage the
state directly. CQL update statements which affect the status will
automatically maintain the statistics.
The calculations are vastly simpler than the history mechanism, so as to keep performance as high as possible. For inserts, there is no cost to monitoring the table (the insert simply increments the statistic if the transaction completes). For deletes, the delete query is first run as a select, aggregating on the monitored columns to find the number of deletes for each domain allowed value. This means that a delete of millions of rows might requires a select returning only a single row for statistics purposes. For updates, the delete code is run, then the insert calculation is done, multiplied by the number of rows affected by the update.
In all cases, CQL ends up calling cql_statistic_monitored_attribute_change_hook/5, where the last argument is a signed value indicating the number of changes to that particular statistic.
Expand at runtime if the first term is compile_at_runtime
varchar(30)
or decimal(10, 5)
Can be autoconfigured.primary_key(ColumnNames:list)
foreign_key(ForeignTableName:atom, ForeignColumnNames:list, ColumnNames:list)
unique(ColumnNames:list)
check(CheckClause)
In theory this can be autoconfigured too, but I have not written the code for it yet
The following predicates are exported from this file while their implementation is defined in imported modules or non-module files loaded by this module.
The following predicates are exported, but not or incorrectly documented.