1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
|
from sqlglot import exp
from sqlglot.helper import ensure_list, subclasses
from sqlglot.optimizer.schema import ensure_schema
from sqlglot.optimizer.scope import Scope, traverse_scope
def annotate_types(expression, schema=None, annotators=None, coerces_to=None):
"""
Recursively infer & annotate types in an expression syntax tree against a schema.
Assumes that we've already executed the optimizer's qualify_columns step.
Example:
>>> import sqlglot
>>> schema = {"y": {"cola": "SMALLINT"}}
>>> sql = "SELECT x.cola + 2.5 AS cola FROM (SELECT y.cola AS cola FROM y AS y) AS x"
>>> annotated_expr = annotate_types(sqlglot.parse_one(sql), schema=schema)
>>> annotated_expr.expressions[0].type # Get the type of "x.cola + 2.5 AS cola"
<Type.DOUBLE: 'DOUBLE'>
Args:
expression (sqlglot.Expression): Expression to annotate.
schema (dict|sqlglot.optimizer.Schema): Database schema.
annotators (dict): Maps expression type to corresponding annotation function.
coerces_to (dict): Maps expression type to set of types that it can be coerced into.
Returns:
sqlglot.Expression: expression annotated with types
"""
schema = ensure_schema(schema)
return TypeAnnotator(schema, annotators, coerces_to).annotate(expression)
class TypeAnnotator:
ANNOTATORS = {
**{
expr_type: lambda self, expr: self._annotate_unary(expr)
for expr_type in subclasses(exp.__name__, exp.Unary)
},
**{
expr_type: lambda self, expr: self._annotate_binary(expr)
for expr_type in subclasses(exp.__name__, exp.Binary)
},
exp.Cast: lambda self, expr: self._annotate_with_type(expr, expr.args["to"].this),
exp.DataType: lambda self, expr: self._annotate_with_type(expr, expr.this),
exp.Alias: lambda self, expr: self._annotate_unary(expr),
exp.Literal: lambda self, expr: self._annotate_literal(expr),
exp.Boolean: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.BOOLEAN),
exp.Null: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.NULL),
exp.Anonymous: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.UNKNOWN),
exp.ApproxDistinct: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.BIGINT),
exp.Avg: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.DOUBLE),
exp.Ceil: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.INT),
exp.Count: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.BIGINT),
exp.CurrentDate: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.DATE),
exp.CurrentDatetime: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.DATETIME),
exp.CurrentTime: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.TIMESTAMP),
exp.CurrentTimestamp: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.TIMESTAMP),
exp.DateAdd: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.DATE),
exp.DateSub: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.DATE),
exp.DateDiff: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.INT),
exp.DatetimeAdd: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.DATETIME),
exp.DatetimeSub: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.DATETIME),
exp.DatetimeDiff: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.INT),
exp.Extract: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.INT),
exp.TimestampAdd: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.TIMESTAMP),
exp.TimestampSub: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.TIMESTAMP),
exp.TimestampDiff: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.INT),
exp.TimeAdd: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.TIMESTAMP),
exp.TimeSub: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.TIMESTAMP),
exp.TimeDiff: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.INT),
exp.DateStrToDate: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.DATE),
exp.DateToDateStr: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.VARCHAR),
exp.DateToDi: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.INT),
exp.Day: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.TINYINT),
exp.DiToDate: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.DATE),
exp.Exp: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.DOUBLE),
exp.Floor: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.INT),
exp.If: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.BOOLEAN),
exp.Initcap: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.VARCHAR),
exp.Length: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.BIGINT),
exp.Levenshtein: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.INT),
exp.Ln: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.DOUBLE),
exp.Log: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.DOUBLE),
exp.Log2: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.DOUBLE),
exp.Log10: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.DOUBLE),
exp.Lower: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.VARCHAR),
exp.Month: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.TINYINT),
exp.Pow: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.DOUBLE),
exp.Quantile: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.DOUBLE),
exp.ApproxQuantile: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.DOUBLE),
exp.RegexpLike: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.BOOLEAN),
exp.Round: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.DOUBLE),
exp.SafeDivide: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.DOUBLE),
exp.Substring: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.VARCHAR),
exp.StrPosition: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.INT),
exp.StrToDate: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.DATE),
exp.StrToTime: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.TIMESTAMP),
exp.Sqrt: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.DOUBLE),
exp.Stddev: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.DOUBLE),
exp.StddevPop: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.DOUBLE),
exp.StddevSamp: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.DOUBLE),
exp.TimeToStr: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.VARCHAR),
exp.TimeToTimeStr: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.VARCHAR),
exp.TimeStrToDate: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.DATE),
exp.TimeStrToTime: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.TIMESTAMP),
exp.Trim: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.VARCHAR),
exp.TsOrDsToDateStr: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.VARCHAR),
exp.TsOrDsToDate: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.DATE),
exp.TsOrDiToDi: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.INT),
exp.UnixToStr: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.VARCHAR),
exp.UnixToTime: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.TIMESTAMP),
exp.UnixToTimeStr: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.VARCHAR),
exp.Upper: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.VARCHAR),
exp.Variance: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.DOUBLE),
exp.VariancePop: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.DOUBLE),
exp.Week: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.TINYINT),
exp.Year: lambda self, expr: self._annotate_with_type(expr, exp.DataType.Type.TINYINT),
}
# Reference: https://spark.apache.org/docs/3.2.0/sql-ref-ansi-compliance.html
COERCES_TO = {
# CHAR < NCHAR < VARCHAR < NVARCHAR < TEXT
exp.DataType.Type.TEXT: set(),
exp.DataType.Type.NVARCHAR: {exp.DataType.Type.TEXT},
exp.DataType.Type.VARCHAR: {exp.DataType.Type.NVARCHAR, exp.DataType.Type.TEXT},
exp.DataType.Type.NCHAR: {exp.DataType.Type.VARCHAR, exp.DataType.Type.NVARCHAR, exp.DataType.Type.TEXT},
exp.DataType.Type.CHAR: {
exp.DataType.Type.NCHAR,
exp.DataType.Type.VARCHAR,
exp.DataType.Type.NVARCHAR,
exp.DataType.Type.TEXT,
},
# TINYINT < SMALLINT < INT < BIGINT < DECIMAL < FLOAT < DOUBLE
exp.DataType.Type.DOUBLE: set(),
exp.DataType.Type.FLOAT: {exp.DataType.Type.DOUBLE},
exp.DataType.Type.DECIMAL: {exp.DataType.Type.FLOAT, exp.DataType.Type.DOUBLE},
exp.DataType.Type.BIGINT: {exp.DataType.Type.DECIMAL, exp.DataType.Type.FLOAT, exp.DataType.Type.DOUBLE},
exp.DataType.Type.INT: {
exp.DataType.Type.BIGINT,
exp.DataType.Type.DECIMAL,
exp.DataType.Type.FLOAT,
exp.DataType.Type.DOUBLE,
},
exp.DataType.Type.SMALLINT: {
exp.DataType.Type.INT,
exp.DataType.Type.BIGINT,
exp.DataType.Type.DECIMAL,
exp.DataType.Type.FLOAT,
exp.DataType.Type.DOUBLE,
},
exp.DataType.Type.TINYINT: {
exp.DataType.Type.SMALLINT,
exp.DataType.Type.INT,
exp.DataType.Type.BIGINT,
exp.DataType.Type.DECIMAL,
exp.DataType.Type.FLOAT,
exp.DataType.Type.DOUBLE,
},
# DATE < DATETIME < TIMESTAMP < TIMESTAMPTZ < TIMESTAMPLTZ
exp.DataType.Type.TIMESTAMPLTZ: set(),
exp.DataType.Type.TIMESTAMPTZ: {exp.DataType.Type.TIMESTAMPLTZ},
exp.DataType.Type.TIMESTAMP: {exp.DataType.Type.TIMESTAMPTZ, exp.DataType.Type.TIMESTAMPLTZ},
exp.DataType.Type.DATETIME: {
exp.DataType.Type.TIMESTAMP,
exp.DataType.Type.TIMESTAMPTZ,
exp.DataType.Type.TIMESTAMPLTZ,
},
exp.DataType.Type.DATE: {
exp.DataType.Type.DATETIME,
exp.DataType.Type.TIMESTAMP,
exp.DataType.Type.TIMESTAMPTZ,
exp.DataType.Type.TIMESTAMPLTZ,
},
}
TRAVERSABLES = (exp.Select, exp.Union, exp.UDTF, exp.Subquery)
def __init__(self, schema=None, annotators=None, coerces_to=None):
self.schema = schema
self.annotators = annotators or self.ANNOTATORS
self.coerces_to = coerces_to or self.COERCES_TO
def annotate(self, expression):
if isinstance(expression, self.TRAVERSABLES):
for scope in traverse_scope(expression):
subscope_selects = {
name: {select.alias_or_name: select for select in source.selects}
for name, source in scope.sources.items()
if isinstance(source, Scope)
}
# First annotate the current scope's column references
for col in scope.columns:
source = scope.sources[col.table]
if isinstance(source, exp.Table):
col.type = self.schema.get_column_type(source, col)
else:
col.type = subscope_selects[col.table][col.name].type
# Then (possibly) annotate the remaining expressions in the scope
self._maybe_annotate(scope.expression)
return self._maybe_annotate(expression) # This takes care of non-traversable expressions
def _maybe_annotate(self, expression):
if not isinstance(expression, exp.Expression):
return None
if expression.type:
return expression # We've already inferred the expression's type
annotator = self.annotators.get(expression.__class__)
return (
annotator(self, expression)
if annotator
else self._annotate_with_type(expression, exp.DataType.Type.UNKNOWN)
)
def _annotate_args(self, expression):
for value in expression.args.values():
for v in ensure_list(value):
self._maybe_annotate(v)
return expression
def _maybe_coerce(self, type1, type2):
# We propagate the NULL / UNKNOWN types upwards if found
if exp.DataType.Type.NULL in (type1, type2):
return exp.DataType.Type.NULL
if exp.DataType.Type.UNKNOWN in (type1, type2):
return exp.DataType.Type.UNKNOWN
return type2 if type2 in self.coerces_to[type1] else type1
def _annotate_binary(self, expression):
self._annotate_args(expression)
left_type = expression.left.type
right_type = expression.right.type
if isinstance(expression, (exp.And, exp.Or)):
if left_type == exp.DataType.Type.NULL and right_type == exp.DataType.Type.NULL:
expression.type = exp.DataType.Type.NULL
elif exp.DataType.Type.NULL in (left_type, right_type):
expression.type = exp.DataType.build("NULLABLE", expressions=exp.DataType.build("BOOLEAN"))
else:
expression.type = exp.DataType.Type.BOOLEAN
elif isinstance(expression, (exp.Condition, exp.Predicate)):
expression.type = exp.DataType.Type.BOOLEAN
else:
expression.type = self._maybe_coerce(left_type, right_type)
return expression
def _annotate_unary(self, expression):
self._annotate_args(expression)
if isinstance(expression, exp.Condition) and not isinstance(expression, exp.Paren):
expression.type = exp.DataType.Type.BOOLEAN
else:
expression.type = expression.this.type
return expression
def _annotate_literal(self, expression):
if expression.is_string:
expression.type = exp.DataType.Type.VARCHAR
elif expression.is_int:
expression.type = exp.DataType.Type.INT
else:
expression.type = exp.DataType.Type.DOUBLE
return expression
def _annotate_with_type(self, expression, target_type):
expression.type = target_type
return self._annotate_args(expression)
|