General boolean inference improvements
Some checks are pending
/ build (macos-latest, 3.8) (push) Waiting to run
/ build (ubuntu-latest, pypy-2.7) (push) Waiting to run
/ build (windows-latest, 3.14) (push) Waiting to run

This commit is contained in:
Miroslav Štampar 2026-07-06 12:02:22 +02:00
parent 6597415ab0
commit 50ff3debe5
13 changed files with 14506 additions and 1629 deletions

View file

@ -401,26 +401,39 @@ class Databases(object):
plusOne = Backend.getIdentifiedDbms() in PLUS_ONE_DBMSES
indexRange = getLimitRange(count, plusOne=plusOne)
for index in indexRange:
if Backend.isDbms(DBMS.SYBASE):
query = _query % (db, (kb.data.cachedTables[-1] if kb.data.cachedTables else " "))
elif Backend.getIdentifiedDbms() in (DBMS.MAXDB, DBMS.ACCESS, DBMS.MCKOI, DBMS.EXTREMEDB):
query = _query % (kb.data.cachedTables[-1] if kb.data.cachedTables else " ")
elif Backend.getIdentifiedDbms() in (DBMS.SQLITE, DBMS.FIREBIRD):
query = _query % index
elif Backend.getIdentifiedDbms() in (DBMS.HSQLDB, DBMS.INFORMIX, DBMS.FRONTBASE, DBMS.VIRTUOSO):
query = _query % (index, unsafeSQLIdentificatorNaming(db))
elif Backend.getIdentifiedDbms() in (DBMS.SPANNER,):
query = _query % (unsafeSQLIdentificatorNaming(db), unsafeSQLIdentificatorNaming(db), index)
else:
query = _query % (unsafeSQLIdentificatorNaming(db), index)
# Value-parallel, prediction-assisted name enumeration for the DBMSes using the
# generic "<db>, <index>" blind template. Retrieves whole names concurrently (one per
# worker, each decoded sequentially so wordlist prediction applies). Used only with
# '--threads' (like getColumns below); single-thread stays on the classic getValue loop,
# which still gets predictive inference via getPartRun. The special templates stay serial.
genericTemplate = Backend.getIdentifiedDbms() not in (DBMS.SYBASE, DBMS.MAXDB, DBMS.ACCESS, DBMS.MCKOI, DBMS.EXTREMEDB, DBMS.SQLITE, DBMS.FIREBIRD, DBMS.HSQLDB, DBMS.INFORMIX, DBMS.FRONTBASE, DBMS.VIRTUOSO, DBMS.SPANNER)
table = unArrayizeValue(inject.getValue(query, union=False, error=False))
if genericTemplate and conf.threads > 1 and isTechniqueAvailable(PAYLOAD.TECHNIQUE.BOOLEAN):
for table in (inject._threadedInferenceValues(lambda index: _query % (unsafeSQLIdentificatorNaming(db), index), indexRange, context="Tables") or []):
if not isNoneValue(table):
kb.hintValue = table
tables.append(safeSQLIdentificatorNaming(table, True))
else:
for index in indexRange:
if Backend.isDbms(DBMS.SYBASE):
query = _query % (db, (kb.data.cachedTables[-1] if kb.data.cachedTables else " "))
elif Backend.getIdentifiedDbms() in (DBMS.MAXDB, DBMS.ACCESS, DBMS.MCKOI, DBMS.EXTREMEDB):
query = _query % (kb.data.cachedTables[-1] if kb.data.cachedTables else " ")
elif Backend.getIdentifiedDbms() in (DBMS.SQLITE, DBMS.FIREBIRD):
query = _query % index
elif Backend.getIdentifiedDbms() in (DBMS.HSQLDB, DBMS.INFORMIX, DBMS.FRONTBASE, DBMS.VIRTUOSO):
query = _query % (index, unsafeSQLIdentificatorNaming(db))
elif Backend.getIdentifiedDbms() in (DBMS.SPANNER,):
query = _query % (unsafeSQLIdentificatorNaming(db), unsafeSQLIdentificatorNaming(db), index)
else:
query = _query % (unsafeSQLIdentificatorNaming(db), index)
if not isNoneValue(table):
kb.hintValue = table
table = safeSQLIdentificatorNaming(table, True)
tables.append(table)
table = unArrayizeValue(inject.getValue(query, union=False, error=False))
if not isNoneValue(table):
kb.hintValue = table
table = safeSQLIdentificatorNaming(table, True)
tables.append(table)
if tables:
kb.data.cachedTables[db] = tables
@ -841,7 +854,7 @@ class Databases(object):
logger.error(errMsg)
continue
for index in getLimitRange(count):
def columnNameQuery(index):
if Backend.getIdentifiedDbms() in (DBMS.MYSQL, DBMS.PGSQL, DBMS.HSQLDB, DBMS.VERTICA, DBMS.PRESTO, DBMS.CRATEDB, DBMS.CUBRID, DBMS.CACHE, DBMS.FRONTBASE, DBMS.VIRTUOSO):
query = rootQuery.blind.query % (unsafeSQLIdentificatorNaming(tbl), unsafeSQLIdentificatorNaming(conf.db))
query += condQuery
@ -880,8 +893,22 @@ class Databases(object):
query += condQuery
field = condition
query = agent.limitQuery(index, query, field, field)
column = unArrayizeValue(inject.getValue(query, union=False, error=False))
return agent.limitQuery(index, query, field, field)
indexList = list(getLimitRange(count))
# Value-parallel column-NAME enumeration: the same axis/mechanism as getTables (one name per
# worker, decoded sequentially - no length probe, predictive inference applies, names stream
# live). Serial fallback for single-thread and when also fetching per-column comments.
columnNames = None
if conf.threads > 1 and not conf.getComments and isTechniqueAvailable(PAYLOAD.TECHNIQUE.BOOLEAN):
columnNames = inject._threadedInferenceValues(columnNameQuery, indexList, context="Columns")
for position, index in enumerate(indexList):
if columnNames is not None:
column = unArrayizeValue(columnNames[position])
else:
column = unArrayizeValue(inject.getValue(columnNameQuery(index), union=False, error=False))
if not isNoneValue(column):
if conf.getComments:

View file

@ -418,41 +418,70 @@ class Entries(object):
debugMsg += "dumped as it appears to be empty"
logger.debug(debugMsg)
def cellQuery(column, index):
if Backend.getIdentifiedDbms() in (DBMS.MYSQL, DBMS.PGSQL, DBMS.HSQLDB, DBMS.H2, DBMS.VERTICA, DBMS.PRESTO, DBMS.CRATEDB, DBMS.CACHE, DBMS.CLICKHOUSE, DBMS.SNOWFLAKE, DBMS.SPANNER):
query = rootQuery.blind.query % (agent.preprocessField(tbl, column), conf.db, conf.tbl, prioritySortColumns(colList)[0], index)
elif Backend.getIdentifiedDbms() in (DBMS.ORACLE, DBMS.DB2, DBMS.DERBY, DBMS.ALTIBASE,):
query = rootQuery.blind.query % (agent.preprocessField(tbl, column), tbl.upper() if not conf.db else ("%s.%s" % (conf.db.upper(), tbl.upper())), index)
elif Backend.getIdentifiedDbms() in (DBMS.MIMERSQL,):
query = rootQuery.blind.query % (agent.preprocessField(tbl, column), tbl.upper() if not conf.db else ("%s.%s" % (conf.db.upper(), tbl.upper())), prioritySortColumns(colList)[0], index)
elif Backend.getIdentifiedDbms() in (DBMS.SQLITE, DBMS.EXTREMEDB):
query = rootQuery.blind.query % (agent.preprocessField(tbl, column), tbl, index)
elif Backend.isDbms(DBMS.FIREBIRD):
query = rootQuery.blind.query % (index, agent.preprocessField(tbl, column), tbl)
elif Backend.getIdentifiedDbms() in (DBMS.INFORMIX, DBMS.VIRTUOSO):
query = rootQuery.blind.query % (index, agent.preprocessField(tbl, column), conf.db, tbl, prioritySortColumns(colList)[0])
elif Backend.isDbms(DBMS.FRONTBASE):
query = rootQuery.blind.query % (index, agent.preprocessField(tbl, column), conf.db, tbl)
else:
query = rootQuery.blind.query % (agent.preprocessField(tbl, column), conf.db, tbl, index)
return agent.whereQuery(query)
try:
for index in indexRange:
# Value-parallel dumping: one whole cell per worker, decoded sequentially, so there
# is NO per-cell LENGTH() probe (the position-parallel path needs one to split a
# value's characters across threads) and the per-column Huffman model + low-cardinality
# guessing engage under concurrency. Used for the boolean channel with '--threads'; the
# classic per-character-parallel loop stays for single-thread and time-based.
if conf.threads > 1 and not conf.dnsDomain and isTechniqueAvailable(PAYLOAD.TECHNIQUE.BOOLEAN):
# One value-parallel pass over every (non-empty) cell, so there is a single
# thread pool and values stream live as they complete - out of order, exactly
# like the error/union dumps - instead of a silent progress counter.
nonEmpty = [_ for _ in colList if _ not in emptyColumns]
tasks = [(column, index) for column in nonEmpty for index in indexRange]
retrieved = inject._threadedInferenceValues(lambda pair: cellQuery(pair[0], pair[1]), tasks, charsetType=None, dump=True) if tasks else []
retrieved = retrieved if retrieved is not None else [None] * len(tasks)
offset = 0
for column in colList:
value = ""
entries[column] = BigArray()
lengths.setdefault(column, 0)
if column not in lengths:
lengths[column] = 0
if column not in entries:
entries[column] = BigArray()
if Backend.getIdentifiedDbms() in (DBMS.MYSQL, DBMS.PGSQL, DBMS.HSQLDB, DBMS.H2, DBMS.VERTICA, DBMS.PRESTO, DBMS.CRATEDB, DBMS.CACHE, DBMS.CLICKHOUSE, DBMS.SNOWFLAKE, DBMS.SPANNER):
query = rootQuery.blind.query % (agent.preprocessField(tbl, column), conf.db, conf.tbl, prioritySortColumns(colList)[0], index)
elif Backend.getIdentifiedDbms() in (DBMS.ORACLE, DBMS.DB2, DBMS.DERBY, DBMS.ALTIBASE,):
query = rootQuery.blind.query % (agent.preprocessField(tbl, column), tbl.upper() if not conf.db else ("%s.%s" % (conf.db.upper(), tbl.upper())), index)
elif Backend.getIdentifiedDbms() in (DBMS.MIMERSQL,):
query = rootQuery.blind.query % (agent.preprocessField(tbl, column), tbl.upper() if not conf.db else ("%s.%s" % (conf.db.upper(), tbl.upper())), prioritySortColumns(colList)[0], index)
elif Backend.getIdentifiedDbms() in (DBMS.SQLITE, DBMS.EXTREMEDB):
query = rootQuery.blind.query % (agent.preprocessField(tbl, column), tbl, index)
elif Backend.isDbms(DBMS.FIREBIRD):
query = rootQuery.blind.query % (index, agent.preprocessField(tbl, column), tbl)
elif Backend.getIdentifiedDbms() in (DBMS.INFORMIX, DBMS.VIRTUOSO):
query = rootQuery.blind.query % (index, agent.preprocessField(tbl, column), conf.db, tbl, prioritySortColumns(colList)[0])
elif Backend.isDbms(DBMS.FRONTBASE):
query = rootQuery.blind.query % (index, agent.preprocessField(tbl, column), conf.db, tbl)
if column in emptyColumns:
values = [NULL] * len(indexRange)
else:
query = rootQuery.blind.query % (agent.preprocessField(tbl, column), conf.db, tbl, index)
values = retrieved[offset:offset + len(indexRange)]
offset += len(indexRange)
query = agent.whereQuery(query)
for value in values:
value = '' if value is None else value
lengths[column] = max(lengths[column], getConsoleLength(DUMP_REPLACEMENTS.get(getUnicode(value), getUnicode(value))))
entries[column].append(value)
else:
for index in indexRange:
for column in colList:
if column not in lengths:
lengths[column] = 0
value = NULL if column in emptyColumns else inject.getValue(query, union=False, error=False, dump=True)
value = '' if value is None else value
if column not in entries:
entries[column] = BigArray()
lengths[column] = max(lengths[column], getConsoleLength(DUMP_REPLACEMENTS.get(getUnicode(value), getUnicode(value))))
entries[column].append(value)
value = NULL if column in emptyColumns else inject.getValue(cellQuery(column, index), union=False, error=False, dump=True)
value = '' if value is None else value
lengths[column] = max(lengths[column], getConsoleLength(DUMP_REPLACEMENTS.get(getUnicode(value), getUnicode(value))))
entries[column].append(value)
except KeyboardInterrupt:
kb.dumpKeyboardInterrupt = True