Oracles contain pre-computed expression on a given grammar. A grammar is composed of variables, and operators (unary/binary) that can be applied on these variables. An oracle object is instanciated with such grammar class and an input vector.

class qsynthesis.grammar.grammar.TritonGrammar(vars: List[Tuple[qsynthesis.types.Char, qsynthesis.types.BitSize]], ops: List[BvOp])[source]

Triton Grammar class. It represent a set of operators, and variables of a given size (only 64 bits at the moment).

Constructor taking a set of variables (name and size) and a set of operators.

static from_dict(g_dict: Dict) TritonGrammar[source]

Static method instanciating a TritonGrammar from its representation as a dictionnary.


g_dict – dictionarry representation of the grammar


TritonGrammar object

gen_test_inputs(n: int) List[qsynthesis.types.Input][source]

Generate a list of n input. Thus it generate a random valuation for each variables of the grammar and that n times.


n (int) – Number of Input to generate (size of the list)


list of inputs

Return type:


property non_terminal_operators: List[Operator]

Return the list of non-terminal operators. All unary and binary operators are non terminal as they can be derived.


list of operators namedtuples

str_to_expr(s: str, *args) TritonAst[source]

Convert a string in the format of the grammar into a TritonAst. In practice an args[0] should be a TritonAst from which to spawn a new TritonAst. That is required to get the same mapping of normalized variables than the one used by expr.


s – expression string to convert to TritonAst


the TritonAst representing the expressions string


NameError, TypeError

to_dict() Dict[source]

Return a dictionnary representation of the grammar. This is used for serialization in database etc.


QSynthesis support any types of oracles as long as they implement InputOutputOracle base class which interface is given below. At the moment two children classes implement this interface:

  • InputOutputOracleLevelDB: Main table object where entries are stored in a key-value database made by called Level-DB made by Google, that guarantee logarithmic read in database.

  • InputOutputOracleREST: Class interfacing a InputOutputOracleLevelDB being served on a REST API (cf. Oracle REST server). This table avoids having a whole table locally.

class qsynthesis.tables.base.InputOutputOracle(gr: TritonGrammar, inputs: List[qsynthesis.types.Input], f_name: Path | str = '')[source]

Base Lookup table class. Specify the interface that child class have to implement to be interoperable with other the synthesizer.

Constructor making a I/O oracle from a grammar a set of inputs and a hash type.

  • gr – triton grammar

  • inputs – List of inputs

  • f_name – file name of the table (when being loaded)

add_entries(worklist: List[Tuple[qsynthesis.types.Hash, str]]) None[source]

Add the given list of entries in the database.


worklist (List[Tuple[qsynthesis.types.Hash, str]]) – list of entries to add



add_entry(hash: qsynthesis.types.Hash, value: str) None[source]

Abstract function to add an entry in the lookuptable.

  • hash – already computed hash to add

  • value (str) – expression value to add in the table

property bitsize: qsynthesis.types.BitSize

Size of expression in bit

Return type:


static create(filename: str | Path, grammar: TritonGrammar, inputs: List[qsynthesis.types.Input], constants: List[int] = []) InputOutputOracle[source]

Create a new empty lookup table with the given initial parameters, grammars, inputs and hash_mode.

  • filename – filename of the table to create

  • grammar – TritonGrammar object representing variables and operators

  • inputs (List[qsynthesis.types.Input]) – list of inputs on which to perform evaluation

  • constants – list of constants used


lookuptable instance object

generate(bitsize: int, constants: List[int] = [], do_watch: bool = False, watchdog_threshold: int | float = 90, linearize: bool = False, do_use_blacklist: bool = False, limit: int = 0) None[source]

Generate a new lookup table from scratch with the variables and operators set in the constructor of the table.

  • bitsize – Bitsize of expressions to generate

  • constants – List of constants to use in the generation

  • do_watch – Enable RAM watching thread to monitor memory

  • watchdog_threshold – threshold to be sent to the memory watchdog

  • linearize – whether or not to apply linearization on expressions

  • do_use_blacklist – enable blacklist mechanism on commutative operators. Slower but less memory consuming

  • limit – Maximum number of entries to generate



hash(outs: List[qsynthesis.types.Output]) qsynthesis.types.Hash[source]

Main hashing method that convert outputs to an hash. The hash used is MD5. Note that hashed values are systematically casted in an array of 64bit integers.


outs (List[qsynthesis.types.Output]) – list of outputs to hash


Hash type (bytes, int ..) of the outputs

Return type:


property input_number: int

Number of inputs used in this table

Return type:


is_expr_compatible(expr: TritonAst) bool[source]

Check the compatibility of the given expression with the table. The function checks sizes of expr variables against the one of its own grammar.


expr – TritonAst expression to check


True if the table can decide on this expression

property is_writable: bool

Whether the table enable being written (with new expressions)

Return type:


static load(file: Path | str) InputOutputOracle[source]

Load the given lookup table and returns an instance object.


file – Database file to load


InputOutputOracle object

lookup(outputs: List[qsynthesis.types.Output], *args, use_cache: bool = True) TritonAst | None[source]

Perform a lookup in the table with a given set of outputs corresponding to the evaluation of an AST against the Input of this exact same table. If an entry is found a TritonAst is created and returned.

  • outputs – list of output result of evaluating an ast against the inputs of this table

  • args – args forwarded to grammar and ultimately to the tritonAst in charge of building a new TritonAst

  • use_cache – Boolean enabling caching the the hash of outputs. A second call if the same outputs (which is common) will not trigger a lookup in the database




optional TritonAst corresponding of the expression found in the table

lookup_hash(h: qsynthesis.types.Hash) str | None[source]

Raw lookup for a given key in database.


h (qsynthesis.types.Hash) – hash key to look for in database


string of the expression if found

Return type:


property name: str

Name of the table

Return type:


property operator_number: int

Number of operators used in this table

Return type:


property size: int

Size of the table (number of entries)

Return type:


property var_number: int

Maximum number of variables contained in the table

Return type: