def __init__(self, full_text, text_before_cursor):
self.identifier = None
self.word_before_cursor = word_before_cursor = last_word(
text_before_cursor, include='many_punctuations')
full_text = _strip_named_query(full_text)
text_before_cursor = _strip_named_query(text_before_cursor)
full_text, text_before_cursor, self.local_tables = \
isolate_query_ctes(full_text, text_before_cursor)
self.text_before_cursor_including_last_word = text_before_cursor
# If we've partially typed a word then word_before_cursor won't be an
# empty string. In that case we want to remove the partially typed
# string before sending it to the sqlparser. Otherwise the last token
# will always be the partially typed string which renders the smart
# completion useless because it will always return the list of
# keywords as completion.
if self.word_before_cursor:
if word_before_cursor[-1] == '(' or word_before_cursor[0] == '\\':
parsed = sqlparse.parse(text_before_cursor)
else:
text_before_cursor = text_before_cursor[:-len(
word_before_cursor)]
parsed = sqlparse.parse(text_before_cursor)
self.identifier = parse_partial_identifier(word_before_cursor)
else:
parsed = sqlparse.parse(text_before_cursor)
full_text, text_before_cursor, parsed = \
_split_multiple_statements(full_text, text_before_cursor, parsed)
self.full_text = full_text
self.text_before_cursor = text_before_cursor
self.parsed = parsed
self.last_token = parsed.token_prev(len(parsed.tokens))[1] \
if parsed and parsed.token_prev(len(parsed.tokens))[1] else ''
def get_column_matches(self, suggestion, word_before_cursor):
tables = suggestion.table_refs
do_qualify = suggestion.qualifiable and {'always': True, 'never': False,
'if_more_than_one_table': \
len(tables) > 1}[self.qualify_columns]
def qualify(col, tbl):
return (tbl + '.' +
self.case(col)) if do_qualify else self.case(col)
_logger.debug("Completion column scope: %r", tables)
scoped_cols = self.populate_scoped_cols(tables,
suggestion.local_tables)
def make_cand(name, ref):
synonyms = (name, generate_alias(self.case(name)))
return Candidate(qualify(name, ref), 0, 'column', synonyms)
def flat_cols():
return [
make_cand(c.name, t.ref) for t, cols in scoped_cols.items()
for c in cols
]
if suggestion.require_last_table:
# require_last_table is used for 'tb11 JOIN tbl2 USING (...' which should
# suggest only columns that appear in the last table and one more
ltbl = tables[-1].ref
other_tbl_cols = set(c.name for t, cs in scoped_cols.items()
if t.ref != ltbl for c in cs)
scoped_cols = {
t: [col for col in cols if col.name in other_tbl_cols]
for t, cols in scoped_cols.items() if t.ref == ltbl
}
lastword = last_word(word_before_cursor, include='most_punctuations')
if lastword == '*':
if suggestion.context == 'insert':
def filter_col(col):
if not col.has_default:
return True
return not any(
p.match(col.default)
for p in self.insert_col_skip_patterns)
scoped_cols = {
t: [col for col in cols if filter_col(col)]
for t, cols in scoped_cols.items()
}
if self.asterisk_column_order == 'alphabetic':
for cols in scoped_cols.values():
cols.sort(key=operator.attrgetter('name'))
if lastword != word_before_cursor \
and len(tables) == 1 \
and word_before_cursor[-len(lastword) - 1] == '.':
# User typed x.*; replicate "x." for all columns except the
# first, which gets the original (as we only replace the "*"")
sep = ', ' + word_before_cursor[:-1]
collist = sep.join(
self.case(c.completion) for c in flat_cols())
else:
collist = ', '.join(
qualify(c.name, t.ref) for t, cs in scoped_cols.items()
for c in cs)
return [
Match(completion=Completion(collist,
-1,
display_meta='columns',
display='*'),
priority=(1, 1, 1))
]
return self.find_matches(word_before_cursor,
flat_cols(),
meta='column')
def suggest_based_on_last_token(token, stmt):
if isinstance(token, string_types):
token_v = token.lower()
elif isinstance(token, Comparison):
# If 'token' is a Comparison type such as
# 'select * FROM abc a JOIN def d ON a.id = d.'. Then calling
# token.value on the comparison type will only return the lhs of the
# comparison. In this case a.id. So we need to do token.tokens to get
# both sides of the comparison and pick the last token out of that
# list.
token_v = token.tokens[-1].value.lower()
elif isinstance(token, Where):
# sqlparse groups all tokens from the where clause into a single token
# list. This means that token.value may be something like
# 'where foo > 5 and '. We need to look "inside" token.tokens to handle
# suggestions in complicated where clauses correctly
prev_keyword = stmt.reduce_to_prev_keyword()
return suggest_based_on_last_token(prev_keyword, stmt)
elif isinstance(token, Identifier):
# If the previous token is an identifier, we can suggest datatypes if
# we're in a parenthesized column/field list, e.g.:
# CREATE TABLE foo (Identifier <CURSOR>
# CREATE FUNCTION foo (Identifier <CURSOR>
# If we're not in a parenthesized list, the most likely scenario is the
# user is about to specify an alias, e.g.:
# SELECT Identifier <CURSOR>
# SELECT foo FROM Identifier <CURSOR>
prev_keyword, _ = find_prev_keyword(stmt.text_before_cursor)
if prev_keyword and prev_keyword.value == '(':
# Suggest datatypes
return suggest_based_on_last_token('type', stmt)
return (Keyword(), )
else:
token_v = token.value.lower()
if not token:
return (Keyword(), Special())
if token_v.endswith('('):
p = sqlparse.parse(stmt.text_before_cursor)[0]
if p.tokens and isinstance(p.tokens[-1], Where):
# Four possibilities:
# 1 - Parenthesized clause like "WHERE foo AND ("
# Suggest columns/functions
# 2 - Function call like "WHERE foo("
# Suggest columns/functions
# 3 - Subquery expression like "WHERE EXISTS ("
# Suggest keywords, in order to do a subquery
# 4 - Subquery OR array comparison like "WHERE foo = ANY("
# Suggest columns/functions AND keywords. (If we wanted to be
# really fancy, we could suggest only array-typed columns)
column_suggestions = suggest_based_on_last_token('where', stmt)
# Check for a subquery expression (cases 3 & 4)
where = p.tokens[-1]
prev_tok = where.token_prev(len(where.tokens) - 1)[1]
if isinstance(prev_tok, Comparison):
# e.g. "SELECT foo FROM bar WHERE foo = ANY("
prev_tok = prev_tok.tokens[-1]
prev_tok = prev_tok.value.lower()
if prev_tok == 'exists':
return (Keyword(), )
return column_suggestions
# Get the token before the parens
prev_tok = p.token_prev(len(p.tokens) - 1)[1]
if (prev_tok and prev_tok.value
and prev_tok.value.lower().split(' ')[-1] == 'using'):
# tbl1 INNER JOIN tbl2 USING (col1, col2)
tables = stmt.get_tables('before')
# suggest columns that are present in more than one table
return (Column(table_refs=tables,
require_last_table=True,
local_tables=stmt.local_tables), )
if p.token_first().value.lower() == 'select':
# If the lparen is preceeded by a space chances are we're about to
# do a sub-select.
if last_word(stmt.text_before_cursor,
'all_punctuations').startswith('('):
return (Keyword(), )
prev_prev_tok = prev_tok and p.token_prev(p.token_index(prev_tok))[1]
if prev_prev_tok and prev_prev_tok.normalized == 'INTO':
return (Column(table_refs=stmt.get_tables('insert'),
context='insert'), )
# We're probably in a function argument list
return (Column(table_refs=extract_tables(stmt.full_text),
local_tables=stmt.local_tables,
qualifiable=True), )
if token_v == 'set':
return (Column(table_refs=stmt.get_tables(),
local_tables=stmt.local_tables), )
if token_v in ('select', 'where', 'having', 'by', 'distinct'):
# Check for a table alias or schema qualification
parent = stmt.identifier.get_parent_name() \
if (stmt.identifier and stmt.identifier.get_parent_name()) else []
tables = stmt.get_tables()
if parent:
tables = tuple(t for t in tables if identifies(parent, t))
return (
Column(table_refs=tables, local_tables=stmt.local_tables),
Table(schema=parent),
View(schema=parent),
Function(schema=parent),
)
return (
Column(table_refs=tables,
local_tables=stmt.local_tables,
qualifiable=True),
Function(schema=None),
Keyword(token_v.upper()),
)
if token_v == 'as':
# Don't suggest anything for aliases
return ()
if (token_v.endswith('join') and token.is_keyword) or \
token_v in ('copy', 'from', 'update', 'into', 'describe', 'truncate'):
schema = stmt.get_identifier_schema()
tables = extract_tables(stmt.text_before_cursor)
is_join = token_v.endswith('join') and token.is_keyword
# Suggest tables from either the currently-selected schema or the
# public schema if no schema has been specified
suggest = []
if not schema:
# Suggest schemas
suggest.insert(0, Schema())
if token_v == 'from' or is_join:
suggest.append(
FromClauseItem(schema=schema,
table_refs=tables,
local_tables=stmt.local_tables))
elif token_v == 'truncate':
suggest.append(Table(schema))
else:
suggest.extend((Table(schema), View(schema)))
if is_join and _allow_join(stmt.parsed):
tables = stmt.get_tables('before')
suggest.append(Join(table_refs=tables, schema=schema))
return tuple(suggest)
if token_v == 'function':
schema = stmt.get_identifier_schema()
# stmt.get_previous_token will fail for e.g. `SELECT 1 FROM functions
# WHERE function:`
try:
prev = stmt.get_previous_token(token).value.lower()
if prev in ('drop', 'alter', 'create', 'create or replace'):
return (Function(schema=schema, usage='signature'), )
except ValueError:
pass
return tuple()
if token_v in ('table', 'view'):
# E.g. 'ALTER TABLE <tablname>'
rel_type = {
'table': Table,
'view': View,
'function': Function
}[token_v]
schema = stmt.get_identifier_schema()
if schema:
return (rel_type(schema=schema), )
return (Schema(), rel_type(schema=schema))
if token_v == 'column':
# E.g. 'ALTER TABLE foo ALTER COLUMN bar
return (Column(table_refs=stmt.get_tables()), )
if token_v == 'on':
tables = stmt.get_tables('before')
parent = stmt.identifier.get_parent_name() \
if (stmt.identifier and stmt.identifier.get_parent_name()) else None
if parent:
# "ON parent.<suggestion>"
# parent can be either a schema name or table alias
filteredtables = tuple(t for t in tables if identifies(parent, t))
sugs = [
Column(table_refs=filteredtables,
local_tables=stmt.local_tables),
Table(schema=parent),
View(schema=parent),
Function(schema=parent)
]
if filteredtables and _allow_join_condition(stmt.parsed):
sugs.append(
JoinCondition(table_refs=tables,
parent=filteredtables[-1]))
return tuple(sugs)
# ON <suggestion>
# Use table alias if there is one, otherwise the table name
aliases = tuple(t.ref for t in tables)
if _allow_join_condition(stmt.parsed):
return (Alias(aliases=aliases),
JoinCondition(table_refs=tables, parent=None))
return (Alias(aliases=aliases), )
if token_v in ('c', 'use', 'database', 'template'):
# "\c <db", "use <db>", "DROP DATABASE <db>",
# "CREATE DATABASE <newdb> WITH TEMPLATE <db>"
return (Database(), )
if token_v == 'schema':
# DROP SCHEMA schema_name, SET SCHEMA schema name
prev_keyword = stmt.reduce_to_prev_keyword(n_skip=2)
quoted = prev_keyword and prev_keyword.value.lower() == 'set'
return (Schema(quoted), )
if token_v.endswith(',') or token_v in ('=', 'and', 'or'):
prev_keyword = stmt.reduce_to_prev_keyword()
if prev_keyword:
return suggest_based_on_last_token(prev_keyword, stmt)
return ()
if token_v in ('type', '::'):
# ALTER TABLE foo SET DATA TYPE bar
# SELECT foo::bar
# Note that tables are a form of composite type in postgresql, so
# they're suggested here as well
schema = stmt.get_identifier_schema()
suggestions = [Datatype(schema=schema), Table(schema=schema)]
if not schema:
suggestions.append(Schema())
return tuple(suggestions)
if token_v in {'alter', 'create', 'drop'}:
return (Keyword(token_v.upper()), )
if token.is_keyword:
# token is a keyword we haven't implemented any special handling for
# go backwards in the query until we find one we do recognize
prev_keyword = stmt.reduce_to_prev_keyword(n_skip=1)
if prev_keyword:
return suggest_based_on_last_token(prev_keyword, stmt)
return (Keyword(token_v.upper()), )
return (Keyword(), )
def find_matches(self, text, collection, mode='fuzzy', meta=None):
"""Find completion matches for the given text.
Given the user's input text and a collection of available
completions, find completions matching the last word of the
text.
`collection` can be either a list of strings or a list of Candidate
namedtuples.
`mode` can be either 'fuzzy', or 'strict'
'fuzzy': fuzzy matching, ties broken by name prevalance
`keyword`: start only matching, ties broken by keyword prevalance
yields prompt_toolkit Completion instances for any matches found
in the collection of available completions.
"""
if not collection:
return []
prio_order = [
'keyword', 'function', 'view', 'table', 'datatype', 'database',
'schema', 'column', 'table alias', 'join', 'name join', 'fk join'
]
type_priority = prio_order.index(meta) if meta in prio_order else -1
text = last_word(text, include='most_punctuations').lower()
text_len = len(text)
if text and text[0] == '"':
# text starts with double quote; user is manually escaping a name
# Match on everything that follows the double-quote. Note that
# text_len is calculated before removing the quote, so the
# Completion.position value is correct
text = text[1:]
if mode == 'fuzzy':
fuzzy = True
priority_func = self.prioritizer.name_count
else:
fuzzy = False
priority_func = self.prioritizer.keyword_count
# Construct a `_match` function for either fuzzy or non-fuzzy matching
# The match function returns a 2-tuple used for sorting the matches,
# or None if the item doesn't match
# Note: higher priority values mean more important, so use negative
# signs to flip the direction of the tuple
if fuzzy:
regex = '.*?'.join(map(re.escape, text))
pat = re.compile('(%s)' % regex)
def _match(item):
if item.lower()[:len(text) + 1] in (text, text + ' '):
# Exact match of first word in suggestion
# This is to get exact alias matches to the top
# E.g. for input `e`, 'Entries E' should be on top
# (before e.g. `EndUsers EU`)
return float('Infinity'), -1
r = pat.search(self.unescape_name(item.lower()))
if r:
return -len(r.group()), -r.start()
else:
match_end_limit = len(text)
def _match(item):
match_point = item.lower().find(text, 0, match_end_limit)
if match_point >= 0:
# Use negative infinity to force keywords to sort after all
# fuzzy matches
return -float('Infinity'), -match_point
matches = []
for cand in collection:
if isinstance(cand, _Candidate):
item, prio, display_meta, synonyms, prio2, display = cand
if display_meta is None:
display_meta = meta
syn_matches = (_match(x) for x in synonyms)
# Nones need to be removed to avoid max() crashing in Python 3
syn_matches = [m for m in syn_matches if m]
sort_key = max(syn_matches) if syn_matches else None
else:
item, display_meta, prio, prio2, display = cand, meta, 0, 0, cand
sort_key = _match(cand)
if sort_key:
if display_meta and len(display_meta) > 50:
# Truncate meta-text to 50 characters, if necessary
display_meta = display_meta[:47] + u'...'
# Lexical order of items in the collection, used for
# tiebreaking items with the same match group length and start
# position. Since we use *higher* priority to mean "more
# important," we use -ord(c) to prioritize "aa" > "ab" and end
# with 1 to prioritize shorter strings (ie "user" > "users").
# We first do a case-insensitive sort and then a
# case-sensitive one as a tie breaker.
# We also use the unescape_name to make sure quoted names have
# the same priority as unquoted names.
lexical_priority = (tuple(0 if c in(' _') else -ord(c) \
for c in self.unescape_name(item.lower())) +
(1,) + tuple(c for c in item))
item = self.case(item)
display = self.case(display)
priority = (sort_key, type_priority, prio, priority_func(item),
prio2, lexical_priority)
item = decode(item)
display_meta = decode(display_meta)
display = decode(display)
matches.append(
Match(completion=Completion(text=item,
start_position=-text_len,
display_meta=display_meta,
display=display),
priority=priority))
return matches