Skip to content

symbolic

BasicBlock

Linear sequence of statements with a single entry and exit.

Successor/predecessor lists encode the CFG edges. Statements execute in order; no internal branching occurs within a BasicBlock.

Source code in src/finch/symbolic/dataflow.py 
 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
class BasicBlock:
    """Linear sequence of statements with a single entry and exit.

    Successor/predecessor lists encode the CFG edges. Statements execute in
    order; no internal branching occurs within a BasicBlock.
    """

    def __init__(self, id: str) -> None:
        self.id = id
        self.statements: list = []
        self.successors: list[BasicBlock] = []
        self.predecessors: list[BasicBlock] = []

    def add_statement(self, statement) -> None:
        self.statements.append(statement)

    def add_successor(self, successor: "BasicBlock") -> None:
        if successor not in self.successors:
            self.successors.append(successor)

        if self not in successor.predecessors:
            successor.predecessors.append(self)

    def __str__(self) -> str:
        """String representation of BasicBlock in LLVM style."""
        lines = []

        if self.successors:
            succ_names = [succ.id for succ in self.successors]
            succ_str = f" #succs=[{', '.join(succ_names)}]"
        else:
            succ_str = " #succs=[]"

        # Block header
        lines.append(f"{self.id}:{succ_str}")

        # Block statements
        lines.extend(f"    {stmt}" for stmt in self.statements)

        return "\n".join(lines)

__str__()

String representation of BasicBlock in LLVM style.

Source code in src/finch/symbolic/dataflow.py 
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
def __str__(self) -> str:
    """String representation of BasicBlock in LLVM style."""
    lines = []

    if self.successors:
        succ_names = [succ.id for succ in self.successors]
        succ_str = f" #succs=[{', '.join(succ_names)}]"
    else:
        succ_str = " #succs=[]"

    # Block header
    lines.append(f"{self.id}:{succ_str}")

    # Block statements
    lines.extend(f"    {stmt}" for stmt in self.statements)

    return "\n".join(lines)

CallTerm dataclass

Bases: TermTree, ABC

A tree term which applies the operator computed by the expression op to args.

Source code in src/finch/symbolic/term.py 
 98
 99
100
101
102
103
104
class CallTerm(TermTree, ABC):
    """
    A tree term which applies the operator computed by the expression `op` to `args`.
    """

    op: ExpressionTerm
    args: tuple[Term, ...]

Chain

Bases: Generic[T]

A rewriter which rewrites using each rewriter in itr. If all rewriters return nothing, return nothing.

Attributes:

Name Type Description
rws Iterable[RwCallable]

A collection of rewriter functions to apply.
Source code in src/finch/symbolic/rewriters.py 
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
class Chain(Generic[T]):
    """
    A rewriter which rewrites using each rewriter in `itr`. If all rewriters
    return `nothing`, return `nothing`.

    Attributes:
        rws (Iterable[RwCallable]): A collection of rewriter functions to apply.
    """

    def __init__(self, rws: Iterable[RwCallable]):
        self.rws = rws

    def __call__(self, x: T) -> T | None:
        is_success = False
        for rw in self.rws:
            y = rw(x)
            if y is not None:
                is_success = True
                x = y
        if is_success:
            return x
        return None

CompilerMode dataclass

Safe mode checks tensor extents; debug also checks buffer accesses.

Source code in src/finch/symbolic/stage.py 
 5
 6
 7
 8
 9
10
@dataclass(frozen=True)
class CompilerMode:
    """Safe mode checks tensor extents; debug also checks buffer accesses."""

    safe: bool = False
    debug: bool = False

Context

Bases: ABC

Source code in src/finch/symbolic/environment.py 
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
class Context(ABC):
    def __init__(self, namespace=None, preamble=None, epilogue=None, mode=None):
        self.namespace = namespace if namespace is not None else Namespace()
        self.preamble = preamble if preamble is not None else []
        self.epilogue = epilogue if epilogue is not None else []
        self.mode = mode if mode is not None else CompilerMode()

    def exec(self, thunk: Any):
        self.preamble.append(thunk)

    def post(self, thunk: Any):
        self.epilogue.append(thunk)

    def freshen(self, *tags):
        return self.namespace.freshen(*tags)

    def resolve(self, *names: str):
        return self.namespace.resolve(*names)

    def block(self):
        """
        Create a new block. Preambles and epilogues will stay within this block.
        """
        blk = self.__class__()
        blk.namespace = self.namespace
        blk.preamble = []
        blk.epilogue = []
        blk.mode = self.mode
        return blk

    @abstractmethod
    def emit(self):
        """
        Emit the code in this context.
        """
        ...

block()

Create a new block. Preambles and epilogues will stay within this block.

Source code in src/finch/symbolic/environment.py 
149
150
151
152
153
154
155
156
157
158
def block(self):
    """
    Create a new block. Preambles and epilogues will stay within this block.
    """
    blk = self.__class__()
    blk.namespace = self.namespace
    blk.preamble = []
    blk.epilogue = []
    blk.mode = self.mode
    return blk

emit() abstractmethod

Emit the code in this context.

Source code in src/finch/symbolic/environment.py 
160
161
162
163
164
165
@abstractmethod
def emit(self):
    """
    Emit the code in this context.
    """
    ...

ControlFlowGraph

Collection of BasicBlocks plus explicit ENTRY and EXIT nodes.

Provides helpers to allocate uniquely named blocks and holds the global block mapping used by dataflow analyses.

Source code in src/finch/symbolic/dataflow.py 
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
class ControlFlowGraph:
    """Collection of BasicBlocks plus explicit ENTRY and EXIT nodes.

    Provides helpers to allocate uniquely named blocks and holds the global
    block mapping used by dataflow analyses.
    """

    def __init__(self) -> None:
        self.block_counter = 0
        self.block_name = ""
        self.blocks: dict[str, BasicBlock] = {}

        self.entry_block = self.new_block_custom("ENTRY")
        self.exit_block = self.new_block_custom("EXIT")

    def new_block(self) -> BasicBlock:
        bid = f"{self.block_name}_{self.block_counter}"
        self.block_counter += 1
        block = BasicBlock(bid)
        self.blocks[bid] = block
        return block

    def new_block_custom(self, name: str) -> BasicBlock:
        bid = f"{name}_{self.block_counter}"
        self.block_counter += 1
        block = BasicBlock(bid)
        self.blocks[bid] = block
        return block

    def __str__(self) -> str:
        """Print the CFG in LLVM style format."""
        blocks = list(self.blocks.values())

        # Use list comprehension with join for better performance
        block_strings = [str(block) for block in blocks]
        return "\n\n".join(block_strings)

__str__()

Print the CFG in LLVM style format.

Source code in src/finch/symbolic/dataflow.py 
76
77
78
79
80
81
82
def __str__(self) -> str:
    """Print the CFG in LLVM style format."""
    blocks = list(self.blocks.values())

    # Use list comprehension with join for better performance
    block_strings = [str(block) for block in blocks]
    return "\n\n".join(block_strings)

DataFlowAnalysis

Bases: ABC

Base class for performing data flow analyses over a ControlFlowGraph.

This abstract class implements a generic work-list algorithm that can run either in a "forward" or "backward" direction. Subclasses must provide IR-specific semantics by implementing the abstract methods: transfer, join, stmt_str and direction.

Key attributes - cfg: the ControlFlowGraph to analyze. - input_states: a mapping from basic block id to the lattice state that holds at the entry of the block. - output_states: a mapping from basic block id to the lattice state that holds at the exit of the block.

Source code in src/finch/symbolic/dataflow.py 
 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
class DataFlowAnalysis(ABC):
    """
    Base class for performing data flow analyses over a ControlFlowGraph.

    This abstract class implements a generic work-list algorithm that can run either
    in a "forward" or "backward" direction. Subclasses must provide IR-specific
    semantics by implementing the abstract methods: transfer, join, stmt_str and
    direction.

    Key attributes
    - cfg: the ControlFlowGraph to analyze.
    - input_states: a mapping from basic block id to the lattice state that holds
        at the entry of the block.
    - output_states: a mapping from basic block id to the lattice state that holds
        at the exit of the block.
    """

    def __init__(self, cfg: ControlFlowGraph):
        self.cfg: ControlFlowGraph = cfg
        self.input_states: dict[str, dict] = {
            block.id: {} for block in cfg.blocks.values()
        }
        self.output_states: dict[str, dict] = {
            block.id: {} for block in cfg.blocks.values()
        }

    def __str__(self) -> str:
        """
        Return a string representation describing the current dataflow
        analysis state over the control-flow graph (CFG).

        The output lists each basic block in the order
        produced by self.cfg.blocks.values().
        For every block the printed form is:

            <block.id>: #succs=[<succ_id>, ...]
                <stmt_representation_for_first_stmt>
                <stmt_representation_for_second_stmt>
                ...
        """
        lines: list[str] = []
        blocks = list(self.cfg.blocks.values())

        for block in blocks:
            # get successors of the block
            if block.successors:
                succ_names = [succ.id for succ in block.successors]
                succ_str = f"#succs=[{', '.join(succ_names)}]"
            else:
                succ_str = "#succs=[]"

            lines.append(f"{block.id}: {succ_str}")

            # get the input state for this block
            input_state = self.input_states.get(block.id, {})

            if self.direction() == "forward":
                # deep copy the input_state, otherwise original state
                # in the input_states will be changed,
                # which can lead to errors in future
                state = copy.deepcopy(input_state)

                # print each statement using subclasses stmt_str method
                for stmt in block.statements:
                    lines.append(f"    {self.stmt_str(stmt, state)}")
                    # advance state with current statement
                    state = self.transfer([stmt], state)
            else:
                # TODO: implement "backward" direction printing
                lines.extend(
                    f"    {self.stmt_str(stmt, input_state)}"
                    for stmt in block.statements
                )

            lines.append("")

        return "\n".join(lines)

    @abstractmethod
    def stmt_str(self, stmt, state: dict) -> str:
        """
        Format a single statement given the current lattice state.
        Implement in subclasses or IR-specific abstract bases.
        """
        ...

    @abstractmethod
    def transfer(self, stmts, state: dict) -> dict:
        """
        Transfer function for the data flow analysis.
        This should be implemented by subclasses.
        """
        ...

    @abstractmethod
    def join(self, state_1: dict, state_2: dict) -> dict:
        """
        Join function for the data flow analysis.
        This should be implemented by subclasses.
        """
        ...

    @abstractmethod
    def direction(self) -> str:
        """
        Return the direction of the data flow analysis, either "forward" or "backward".
        This should be implemented by subclasses.
        """
        ...

    def analyze(self) -> None:
        """
        Perform the data flow analysis on the control flow graph.
        This method initializes the work list and processes each block.
        """
        work_list: list[BasicBlock] = list(self.cfg.blocks.values())
        if self.direction() == "forward":
            while work_list:
                block = work_list.pop(0)

                # get current input state based on the predecessors output states
                if not block.predecessors:
                    input_state = {}
                else:
                    pred_outputs = [
                        self.output_states.get(pred.id, {})
                        for pred in block.predecessors
                    ]
                    input_state = pred_outputs[0].copy()
                    for pred_output in pred_outputs[1:]:
                        input_state = self.join(input_state, pred_output)

                self.input_states[block.id] = input_state

                # perform transfer based on the statements in the current basic block
                output_state = self.transfer(block.statements, input_state)

                # check if output_state changed
                if output_state != self.output_states.get(block.id, {}):
                    self.output_states[block.id] = output_state

                    for successor in block.successors:
                        if successor not in work_list:
                            work_list.append(successor)
        else:
            while work_list:
                block = work_list.pop(0)

                # get current input state based on the successors output states
                if not block.successors:
                    input_state = {}
                else:
                    succ_outputs = [
                        self.output_states.get(succ.id, {}) for succ in block.successors
                    ]
                    input_state = succ_outputs[0].copy()
                    for succ_output in succ_outputs[1:]:
                        input_state = self.join(input_state, succ_output)

                self.input_states[block.id] = input_state

                # perform trasnfer based on the statements in the current basic block
                output_state = self.transfer(block.statements, input_state)

                # check if output state changed
                if output_state != self.output_states.get(block.id, {}):
                    self.output_states[block.id] = output_state

                    for predecessor in block.predecessors:
                        if predecessor not in work_list:
                            work_list.append(predecessor)

__str__()

Return a string representation describing the current dataflow analysis state over the control-flow graph (CFG).

The output lists each basic block in the order produced by self.cfg.blocks.values(). For every block the printed form is:

<block.id>: #succs=[<succ_id>, ...]
    <stmt_representation_for_first_stmt>
    <stmt_representation_for_second_stmt>
    ...
Source code in src/finch/symbolic/dataflow.py 
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
def __str__(self) -> str:
    """
    Return a string representation describing the current dataflow
    analysis state over the control-flow graph (CFG).

    The output lists each basic block in the order
    produced by self.cfg.blocks.values().
    For every block the printed form is:

        <block.id>: #succs=[<succ_id>, ...]
            <stmt_representation_for_first_stmt>
            <stmt_representation_for_second_stmt>
            ...
    """
    lines: list[str] = []
    blocks = list(self.cfg.blocks.values())

    for block in blocks:
        # get successors of the block
        if block.successors:
            succ_names = [succ.id for succ in block.successors]
            succ_str = f"#succs=[{', '.join(succ_names)}]"
        else:
            succ_str = "#succs=[]"

        lines.append(f"{block.id}: {succ_str}")

        # get the input state for this block
        input_state = self.input_states.get(block.id, {})

        if self.direction() == "forward":
            # deep copy the input_state, otherwise original state
            # in the input_states will be changed,
            # which can lead to errors in future
            state = copy.deepcopy(input_state)

            # print each statement using subclasses stmt_str method
            for stmt in block.statements:
                lines.append(f"    {self.stmt_str(stmt, state)}")
                # advance state with current statement
                state = self.transfer([stmt], state)
        else:
            # TODO: implement "backward" direction printing
            lines.extend(
                f"    {self.stmt_str(stmt, input_state)}"
                for stmt in block.statements
            )

        lines.append("")

    return "\n".join(lines)

analyze()

Perform the data flow analysis on the control flow graph. This method initializes the work list and processes each block.

Source code in src/finch/symbolic/dataflow.py 
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
def analyze(self) -> None:
    """
    Perform the data flow analysis on the control flow graph.
    This method initializes the work list and processes each block.
    """
    work_list: list[BasicBlock] = list(self.cfg.blocks.values())
    if self.direction() == "forward":
        while work_list:
            block = work_list.pop(0)

            # get current input state based on the predecessors output states
            if not block.predecessors:
                input_state = {}
            else:
                pred_outputs = [
                    self.output_states.get(pred.id, {})
                    for pred in block.predecessors
                ]
                input_state = pred_outputs[0].copy()
                for pred_output in pred_outputs[1:]:
                    input_state = self.join(input_state, pred_output)

            self.input_states[block.id] = input_state

            # perform transfer based on the statements in the current basic block
            output_state = self.transfer(block.statements, input_state)

            # check if output_state changed
            if output_state != self.output_states.get(block.id, {}):
                self.output_states[block.id] = output_state

                for successor in block.successors:
                    if successor not in work_list:
                        work_list.append(successor)
    else:
        while work_list:
            block = work_list.pop(0)

            # get current input state based on the successors output states
            if not block.successors:
                input_state = {}
            else:
                succ_outputs = [
                    self.output_states.get(succ.id, {}) for succ in block.successors
                ]
                input_state = succ_outputs[0].copy()
                for succ_output in succ_outputs[1:]:
                    input_state = self.join(input_state, succ_output)

            self.input_states[block.id] = input_state

            # perform trasnfer based on the statements in the current basic block
            output_state = self.transfer(block.statements, input_state)

            # check if output state changed
            if output_state != self.output_states.get(block.id, {}):
                self.output_states[block.id] = output_state

                for predecessor in block.predecessors:
                    if predecessor not in work_list:
                        work_list.append(predecessor)

direction() abstractmethod

Return the direction of the data flow analysis, either "forward" or "backward". This should be implemented by subclasses.

Source code in src/finch/symbolic/dataflow.py 
187
188
189
190
191
192
193
@abstractmethod
def direction(self) -> str:
    """
    Return the direction of the data flow analysis, either "forward" or "backward".
    This should be implemented by subclasses.
    """
    ...

join(state_1, state_2) abstractmethod

Join function for the data flow analysis. This should be implemented by subclasses.

Source code in src/finch/symbolic/dataflow.py 
179
180
181
182
183
184
185
@abstractmethod
def join(self, state_1: dict, state_2: dict) -> dict:
    """
    Join function for the data flow analysis.
    This should be implemented by subclasses.
    """
    ...

stmt_str(stmt, state) abstractmethod

Format a single statement given the current lattice state. Implement in subclasses or IR-specific abstract bases.

Source code in src/finch/symbolic/dataflow.py 
163
164
165
166
167
168
169
@abstractmethod
def stmt_str(self, stmt, state: dict) -> str:
    """
    Format a single statement given the current lattice state.
    Implement in subclasses or IR-specific abstract bases.
    """
    ...

transfer(stmts, state) abstractmethod

Transfer function for the data flow analysis. This should be implemented by subclasses.

Source code in src/finch/symbolic/dataflow.py 
171
172
173
174
175
176
177
@abstractmethod
def transfer(self, stmts, state: dict) -> dict:
    """
    Transfer function for the data flow analysis.
    This should be implemented by subclasses.
    """
    ...

Fixpoint

Bases: Generic[T]

A rewriter which repeatedly applies rw to x until no changes are made. If the rewriter first returns nothing, returns nothing.

Attributes:

Name Type Description
rw RwCallable

The rewriter function to apply.
Source code in src/finch/symbolic/rewriters.py 
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
class Fixpoint(Generic[T]):
    """
    A rewriter which repeatedly applies `rw` to `x` until no changes are made. If
    the rewriter first returns `nothing`, returns `nothing`.

    Attributes:
        rw (RwCallable): The rewriter function to apply.
    """

    def __init__(self, rw: RwCallable):
        self.rw = rw

    def __call__(self, x: T) -> T | None:
        y = self.rw(x)
        if y is not None:
            while y is not None and x != y:
                x = y
                y = self.rw(x)
            return x
        return None

Form

Bases: ABC

A Form represents the invariants that must hold for the inputs to be valid for a given stage. It will typically be a combination of a grammar and a set of constraints on the inputs. Stages should inherit from a Form to validate their inputs.

Source code in src/finch/symbolic/stage.py 
25
26
27
28
29
30
31
32
33
34
class Form(ABC):
    """
    A Form represents the invariants that must hold for the inputs to be valid
    for a given stage. It will typically be a combination of a grammar and a
    set of constraints on the inputs. Stages should inherit from a Form to
    validate their inputs.
    """

    @abstractmethod
    def validate_inputs(cls, *inputs): ...

LiteralTerm dataclass

Bases: Term, Generic[T]

A leaf term which wraps the constant val.

Source code in src/finch/symbolic/term.py 
89
90
91
92
93
94
95
@dataclass(eq=True, frozen=True)
class LiteralTerm(Term, Generic[T]):
    """
    A leaf term which wraps the constant `val`.
    """

    val: T

Memo

Bases: Generic[T]

A rewriter which caches the results of rw in cache and returns the result.

Attributes:

Name Type Description
rw RwCallable

The rewriter function to apply.
cache dict

A dictionary to store cached results.
Source code in src/finch/symbolic/rewriters.py 
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
class Memo(Generic[T]):
    """
    A rewriter which caches the results of `rw` in `cache` and returns the
    result.

    Attributes:
        rw (RwCallable): The rewriter function to apply.
        cache (dict): A dictionary to store cached results.
    """

    def __init__(self, rw: RwCallable, cache: dict | None = None):
        self.rw = rw
        self.cache = cache if cache is not None else {}

    def __call__(self, x: T) -> T | None:
        if x not in self.cache:
            self.cache[x] = self.rw(x)
        return self.cache[x]

NamedTerm

Bases: Term, ABC

A named program node, e.g. a variable, function, or index.

Source code in src/finch/symbolic/environment.py 
11
12
13
14
15
16
17
18
class NamedTerm(Term, ABC):
    """
    A named program node, e.g. a variable, function, or index.
    """

    @property
    @abstractmethod
    def symbol(self) -> str: ...

Namespace 

Namespace

A namespace for managing variable names and aesthetic fresh variable generation. You can construct a namespace from an existing tree of Terms to avoid name collisions.

Source code in src/finch/symbolic/environment.py 
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
class Namespace:
    """
        Namespace

    A namespace for managing variable names and aesthetic fresh variable
    generation.  You can construct a namespace from an existing tree of `Term`s
    to avoid name collisions.
    """

    def __init__(self, root=None):
        self.counts = defaultdict(int)
        self.resolutions = {}
        if root is not None:
            for node in PostOrderDFS(root):
                if isinstance(node, NamedTerm):
                    self.freshen(node.symbol)

    def freshen(self, *tags) -> str:
        """
        Generate a fresh variable name based on the provided tags.
        """
        name = "_".join(str(tag) for tag in tags)
        m = re.match(r"^(.*)_(\d*)$", name)
        if m is None:
            tag = name
            n = 1
        else:
            tag = m.group(1)
            n = int(m.group(2))
        n = max(self.counts[tag] + 1, n)
        self.counts[tag] = n
        if n == 1:
            return tag
        return f"{tag}_{n}"

    def resolve(self, *names: str):
        """
        Resolve a list of namespaced variable names to a unique name.
        e.g. `resolve("a", "b")` might return `a_b_1` if `a_b` has already been
        used in scope.
        """
        self.resolutions.setdefault(names, lambda: self.freshen(*names))

freshen(*tags)

Generate a fresh variable name based on the provided tags.

Source code in src/finch/symbolic/environment.py 
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
def freshen(self, *tags) -> str:
    """
    Generate a fresh variable name based on the provided tags.
    """
    name = "_".join(str(tag) for tag in tags)
    m = re.match(r"^(.*)_(\d*)$", name)
    if m is None:
        tag = name
        n = 1
    else:
        tag = m.group(1)
        n = int(m.group(2))
    n = max(self.counts[tag] + 1, n)
    self.counts[tag] = n
    if n == 1:
        return tag
    return f"{tag}_{n}"

resolve(*names)

Resolve a list of namespaced variable names to a unique name. e.g. resolve("a", "b") might return a_b_1 if a_b has already been used in scope.

Source code in src/finch/symbolic/environment.py 
56
57
58
59
60
61
62
def resolve(self, *names: str):
    """
    Resolve a list of namespaced variable names to a unique name.
    e.g. `resolve("a", "b")` might return `a_b_1` if `a_b` has already been
    used in scope.
    """
    self.resolutions.setdefault(names, lambda: self.freshen(*names))

PostWalk

A rewriter which recursively rewrites the arguments of each node using rw, then rewrites the resulting node. If all rewriters return nothing, returns nothing.

Attributes:

Name Type Description
rw RwCallable

The rewriter function to apply.
Source code in src/finch/symbolic/rewriters.py 
 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
class PostWalk:
    """
    A rewriter which recursively rewrites the arguments of each node using
    `rw`, then rewrites the resulting node. If all rewriters return `nothing`,
    returns `nothing`.

    Attributes:
        rw (RwCallable): The rewriter function to apply.
    """

    def __init__(self, rw: RwCallable):
        self.rw = rw

    def __call__(self, x: T) -> T | None:
        if isinstance(x, TermTree):
            args = x.children
            new_args = list(map(self, args))
            if all(arg is None for arg in new_args):
                return self.rw(x)
            y = x.make_term(
                x.head(), *map(lambda x1, x2: default_rewrite(x1, x2), new_args, args)
            )
            assert isinstance(y, type(x))
            return default_rewrite(self.rw(y), y)
        return self.rw(x)

PreWalk

A rewriter which recursively rewrites each node using rw, then rewrites the arguments of the resulting node. If all rewriters return nothing, returns nothing.

Attributes:

Name Type Description
rw RwCallable

The rewriter function to apply.
Source code in src/finch/symbolic/rewriters.py 
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
class PreWalk:
    """
    A rewriter which recursively rewrites each node using `rw`, then rewrites
    the arguments of the resulting node. If all rewriters return `nothing`,
    returns `nothing`.

    Attributes:
        rw (RwCallable): The rewriter function to apply.
    """

    def __init__(self, rw: RwCallable):
        self.rw = rw

    def __call__(self, x: T) -> T | None:
        y = self.rw(x)
        if y is not None:
            if isinstance(y, TermTree):
                args = y.children
                return y.make_term(
                    y.head(), *[default_rewrite(self(arg), arg) for arg in args]
                )
            return y
        if isinstance(x, TermTree):
            args = x.children
            new_args = list(map(self, args))
            if not all(arg is None for arg in new_args):
                ret = x.make_term(
                    x.head(),
                    *map(lambda x1, x2: default_rewrite(x1, x2), new_args, args),
                )
                assert isinstance(ret, type(x))
                return ret
        return None

Rewrite

Bases: Generic[T]

A rewriter which returns the original argument even if rw returns nothing.

Attributes:

Name Type Description
rw RwCallable

The rewriter function to apply.
Source code in src/finch/symbolic/rewriters.py 
36
37
38
39
40
41
42
43
44
45
46
47
48
class Rewrite(Generic[T]):
    """
    A rewriter which returns the original argument even if `rw` returns nothing.

    Attributes:
        rw (RwCallable): The rewriter function to apply.
    """

    def __init__(self, rw: RwCallable):
        self.rw = rw

    def __call__(self, x: T) -> T:
        return default_rewrite(self.rw(x), x)

ScopedDict

Bases: Generic[T]

A dictionary that allows for scoped variable resolution.

Source code in src/finch/symbolic/environment.py 
 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
class ScopedDict(Generic[T]):
    """
    A dictionary that allows for scoped variable resolution.
    """

    def __init__(
        self,
        bindings: dict[str, T] | None = None,
        parent: Optional["ScopedDict[T]"] = None,
    ):
        if bindings is None:
            bindings = {}
        self.bindings: dict[str, T] = bindings
        self.parent: ScopedDict[T] | None = parent

    def __getitem__(self, key: str) -> T:
        if key in self.bindings:
            return self.bindings[key]
        if self.parent is not None:
            return self.parent[key]
        raise KeyError(f"Key '{key}' not found in scoped dictionary.")

    def set_in_ancestor(self, leaf, key: str, value: T) -> None:
        if key in self.bindings:
            self.bindings[key] = value
        elif self.parent is not None:
            self.parent.set_in_ancestor(leaf, key, value)
        else:
            leaf.bindings[key] = value

    def del_in_ancestor(self, leaf, key: str) -> None:
        if key in self.bindings:
            del self.bindings[key]
        elif self.parent is not None:
            self.parent.del_in_ancestor(leaf, key)
        else:
            del leaf.bindings[key]

    def __setitem__(self, key: str, value: T) -> None:
        self.set_in_ancestor(self, key, value)

    def __contains__(self, key: str) -> bool:
        return key in self.bindings or (self.parent is not None and key in self.parent)

    def __delitem__(self, key: str) -> None:
        self.del_in_ancestor(self, key)

    def scope(self) -> "ScopedDict[T]":
        """
        Create a new scoped dictionary that inherits from this one.
        """
        return ScopedDict(parent=self)

scope()

Create a new scoped dictionary that inherits from this one.

Source code in src/finch/symbolic/environment.py 
115
116
117
118
119
def scope(self) -> "ScopedDict[T]":
    """
    Create a new scoped dictionary that inherits from this one.
    """
    return ScopedDict(parent=self)

Term

Source code in src/finch/symbolic/term.py 
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
class Term:
    @abstractmethod
    def head(self) -> Callable[..., Self]:
        """Return the head type of the S-expression."""
        ...

    @classmethod
    @abstractmethod
    def make_term(cls, head: Callable[..., Self], *children: Term) -> Self:
        """
        Construct a new term in the same family of terms with the given head type and
        children. This function should satisfy
        `x == x.make_term(x.head(), *x.children)`
        """
        ...

head() abstractmethod

Return the head type of the S-expression.

Source code in src/finch/symbolic/term.py 
54
55
56
57
@abstractmethod
def head(self) -> Callable[..., Self]:
    """Return the head type of the S-expression."""
    ...

make_term(head, *children) abstractmethod classmethod

Construct a new term in the same family of terms with the given head type and children. This function should satisfy x == x.make_term(x.head(), *x.children)

Source code in src/finch/symbolic/term.py 
59
60
61
62
63
64
65
66
67
@classmethod
@abstractmethod
def make_term(cls, head: Callable[..., Self], *children: Term) -> Self:
    """
    Construct a new term in the same family of terms with the given head type and
    children. This function should satisfy
    `x == x.make_term(x.head(), *x.children)`
    """
    ...

TermTree dataclass

Bases: Term, ABC

Source code in src/finch/symbolic/term.py 
80
81
82
83
84
85
86
@dataclass(frozen=True, eq=True)
class TermTree(Term, ABC):
    @property
    @abstractmethod
    def children(self) -> list[Term]:
        """Return the children (AKA tail) of the S-expression."""
        ...

children abstractmethod property

Return the children (AKA tail) of the S-expression.

UnvalidatedForm

Bases: Form

UnvalidatedForm does not perform any validation on the input logic. This is essentially a to-do for validation and should be replaced.

Source code in src/finch/symbolic/stage.py 
37
38
39
40
41
42
43
44
class UnvalidatedForm(Form):
    """
    UnvalidatedForm does not perform any validation on the input logic.
    This is essentially a to-do for validation and should be replaced.
    """

    def validate_inputs(self, *inputs):
        return

intree(n1, n2)

Return True iff n1 occurs in the subtree rooted at n2.

Source code in src/finch/symbolic/traversal.py 
20
21
22
23
24
def intree(n1, n2):
    """
    Return True iff `n1` occurs in the subtree rooted at `n2`.
    """
    return any(node == n1 for node in PostOrderDFS(n2))

isdescendant(n1, n2)

True iff n1 is a strict descendant of n2.

Source code in src/finch/symbolic/traversal.py 
27
28
29
30
31
32
33
def isdescendant(n1, n2):
    """
    True iff `n1` is a strict descendant of `n2`.
    """
    if n1 == n2:
        return False
    return intree(n1, n2)