asterrdetection 0.9.6

A package that identifies semantic errors in faulty code by comparing it against a set of expected correct code versions.

AST ERROR DETECTION

Overview

ast-error-detection is a Python library designed for analyzing and annotating algorithmic errors in code. It leverages the Abstract Syntax Tree (AST) module from Python's standard library to identify, categorize, and contextualize errors. The output of the library is a list of errors, where each error is a dictionary describing the issue and its context.

This library is under an ongoing scientific study. If you use it for academic purposes, please cite the forthcoming publication (details will be provided).

---

Features

• Parses and processes Python code using the AST module.
• Identifies and categorizes errors with detailed contextual information.
• Provides actionable insights for debugging.

---

Installation

Install the package via pip:

pip install ast_error_detection

---

Usage

There are 2 ways to use this library.

Primary Error Detection

The Primary Error Detection layer is responsible for identifying low-level structural and content modifications between two trees (typically a reference and a hypothesis tree). It serves as the foundational layer for subsequent high-level error interpretation. This component relies on the Zhang-Shasha Edit Distance Algorithm, a classic algorithm for computing the minimum-cost sequence of operations needed to transform one tree into another. The algorithm outputs three types of edit operations:

• Insertions
• Deletions
• Updates

Each operation corresponds to a specific node-level change and includes metadata context. These operations collectively represent the raw set of changes (or "primary errors") from which higher-level interpretations can be derived.

from ast-error-detection import get_primary_code_errors

# Example erroneous code to analyze
code_1 = """ 
# Code snippet here
"""
# Example expected code
code_2 = """ 
# Code snippet here
"""

# Convert AST to custom node representation
result = get_primary_code_errors(code_1, code_2)

# Print the results
print(result)

Output Format

The output is always a list of errors. Each error is a dictionary structured as follows:

• For delete or insert errors (3 elements):

  • error: Description of the error.
  • value: The value to delete or insert.
  • context: Contextual location of the error (see below).

• For update errors (4 elements):

  • error: Description of the error.
  • old_value: The old value to update.
  • new_value: The new value to update.
  • context: Contextual location of the error.

Context Format

The context describes where the error occurred in the code execution hierarchy. For example:

Module > Function Name > For Loop > If > Condition

This indicates that the error is in the condition of an if statement inside a for loop within a function in the module.

Typology Based Error Detection

While primary error detection provides atomic operations, Typology-Based Error Detection serves as a semantic interpretation layer. It classifies the primary operations into predefined error types (typologies), allowing for more meaningful feedback and error analysis.

This module:

• Consumes the output of the Primary Error Detection layer.
• Applies a mapping schema to group and classify operations into specific typology classes.
• Detects compound or context-dependent errors by evaluating relationships between multiple primary operations. (logic of each error is defined in the Table below)

Unlike simple pairwise comparison, this component supports comparison of one erroneous code instance against multiple possible correct codes.

• The first input is the hypothesis (erroneous AST).
• The second input is a list of reference ASTs (multiple correct solutions).
• The system computes the edit distance to each reference, selects the closest match, and performs typology-based annotation against that reference.

This ensures the most contextually relevant and minimal-error interpretation is selected for annotation.

from ast-error-detection import get_typology_based_code_error

# Example erroneous code to analyze
code_1 = """ 
# Code snippet here
"""
# Example expected code list
expected_codes = [
  """ 
  # Code snippet 1
  """,
  """ 
  # Code snippet 2
  """
]

# Convert AST to custom node representation
result = get_typology_based_code_error(code_1, expected_codes)

# Print the results
print(result)

Output Format

A list of string (Error tags) from the predefined set of Error tags (Ref Table below)

---

Example

Input Code 1

print('Hello')

Input Code 2

print('Hello1')

Output with Primary error Detection

[('CONST_VALUE_MISMATCH', "Const: 'Hello'", "Const: 'Hello1'", "Module > Call: print > Const: 'Hello'")]

Output with Typology error Detection

['FUNCTION_CALL_PARAMETER_ERROR']

List of error tags

ERROR	DETAILS / EXAMPLES	TASK TYPE(S)	ERROR TAG	IN LIBRARY?
	VARIABLES
(VA-Err1) Error in variable declaration and initialization	Wrong type chosen for the variable or wrong value during initialization	VA – Produce a variable (and children)	VARIABLE_DECLARATION_INITIALIZATION_ERROR	✗
(VA-Err2) Missing variable increment	Typical counter problem: variable never modified or reset each time	—	VARIABLE_MISSING_INCREMENT	✗
(VA-Err3) Invalid variable name	Forbidden character or reserved keyword used	VA – Choose a valid name	VARIABLE_INVALID_NAME	✗
	CONDITIONAL STATEMENTS
(CS-Err1) Incorrect number of branches	A case is missing or branches overlap	IC – Determine the necessary branches; partition the cases	CONDITIONAL_MISSING_BRANCH	✗
(CS-Err2) Misplaced instructions in a conditional	An instruction is in the wrong branch or outside the conditional	IC – Assign the right action to each branch	CONDITIONAL_MISPLACED_INSTRUCTIONS	✗
	FUNCTIONS
(F-Err1) Function definition error	Missing/incorrect parameters, wrong preconditions, incorrect return	F – Define parameters, preconditions, returns	FUNCTION_DEFINITION_ERROR	✗
(F-Err2) Function call parameter error	Called with wrong parameters or return value not captured	—	FUNCTION_CALL_PARAMETER_ERROR	✗
(F-Err3) Invalid function or parameter name	Forbidden character or reserved keyword used	F – Name function and parameters	FUNCTION_INVALID_NAME	✗
	LOOPS
(LO-Err1) Loop iterator usage error	Wrong start/end values or iterator ignored	B – Correct use of loop iterator	LOOP_ITERATOR_USAGE_ERROR	✗
(LO-Err2) Off-by-one loop error	Loop runs one time too many or too few	B – Determine iteration count	LOOP_OFF_BY_ONE_ERROR	✗
(LO-Err3) Wrong loop iteration count (> 2)	Loop executes an entirely wrong number of times	B – Determine iteration count	LOOP_WRONG_ITERATION_COUNT	✗
(LO-Err4) Start condition error	Incorrect initial loop condition	B – Stop condition (unbounded loop)	LOOP_START_CONDITION_ERROR	✗
(LO-Err5) Update condition error	Stop condition never updated or updated incorrectly	B – Modify stop condition	LOOP_UPDATE_CONDITION_ERROR	✗
(LO-Err6) Missing instruction in loop body (not present anywhere)	Expected instruction absent from both loop and program	B – Instructions each iteration	LOOP_BODY_MISSING_INSTRUCTIONS	✗
(LO-Err7) Missing instruction in loop body (moved elsewhere)	Instruction exists but outside the loop	B – Instructions each iteration	LOOP_BODY_INSTRUCTIONS_MOVED_OUT	✗
(LO-Err8) Incorrect instruction close to expected	Instruction present but incorrect / nearly correct	B – Instructions each iteration	LOOP_BODY_INCORRECT_INSTRUCTIONS_NEAR	✗
(LO-Err9) Extra unwanted instruction in loop body	Superfluous instruction unrelated to task	B – Instructions each iteration	LOOP_BODY_EXTRA_INSTRUCTIONS	✗
	EXPRESSIONS
(EXP-Err1) Boolean condition error	Uses < instead of ≤, etc.	IC – Correct boolean expression	EXPRESSION_BOOLEAN_CONDITION_ERROR	✗
(EXP-Err2) Assignment expression error	Part of expression missing or wrong operator	VA – Assign correct expression	EXPRESSION_ASSIGNMENT_ERROR	✗
	PROGRAM & ALGORITHM
(PA-Err1) Problem decomposition / strategy error	Program omits expected control structures	P/A – Design & decompose algorithm	PROGRAM_DECOMPOSITION_STRATEGY_ERROR	✗
(PA-Err2) Requirements misunderstood	Program correct but solves a different task	—	PROGRAM_REQUIREMENTS_MISUNDERSTOOD	✗
(PA-Err3) Incomplete program	Correct fragments present but sequence missing	A – Complete instructions	PROGRAM_INCOMPLETE	✗
(PA-Err4) Program not optimized	Task done but redundantly (no loops/functions)	P/A – Optimize or adapt algorithm	PROGRAM_NOT_OPTIMIZED	✗

These error tags were created as part of an ongoing research project that systematically analyzes novice programming mistakes in a controlled study. The formal paper detailing the methodology and validation of the taxonomy is still in preparation and has not yet been published.

---

License

This project is licensed under the GNU Affero General Public License v3 (AGPL-3.0). If you wish to use this library for proprietary or commercial purposes, you must obtain a separate license.

Please contact Badmavasan at [[e-mail]] for commercial licensing inquiries.

---

Scientific Publication

This library is part of an ongoing scientific study. If you use it for academic purposes, please cite the forthcoming publication:

[Publication details will be added here once available.]

Stay tuned for updates!

---

Internal Architecture & Implementation Reference

This section documents the internal mechanics of both annotation layers in detail. It is intended as a persistent reference for future development and context recovery.

File Map

File	Role
node.py	Custom tree Node class: label, children, parent, index, get_path()
annotated_tree.py	Zhang-Shasha annotated tree: post-order, LMD, keyroots, nodes_path
convert_ast_to_custom_node.py	Converts Python ast nodes → custom Node trees
node_functions.py	Tree utilities: anonymize_variable_names, print_ast_nodes
zang_shasha_distance.py	Core Zhang-Shasha algorithm; returns (dist, ops) where ops is a list of {type, path, current, new} dicts
error_annotation.py	Layer 1 — produces primary errors from edit ops
error_checks.py	Layer 2 — maps primary errors to typed error code strings
error_diagnosis.py	Entry points: get_primary_code_errors, get_typology_based_code_error
constants.py	All tag strings and regex context constants

Variable Anonymization (node_functions.anonymize_variable_names)

Applied in get_primary_code_errors() after tree construction, before distance computation.

Why it is needed: Two programs that solve the same problem but use different variable names (e.g. student uses n, correct solution uses x) would otherwise produce spurious UNNECESSARY_VARIABLE update operations for every single occurrence of those variables, drowning out the real errors.

Algorithm: Pre-order traversal of each tree. The first distinct variable name encountered is mapped to VAR_0, the second to VAR_1, and so on. The mapping is applied independently to each tree.

student:  n = 9; for k in range(4): avancer(n) ...
          ↓ anonymize (n→VAR_0, k→VAR_1)
          VAR_0 = 9; for VAR_1 in range(4): avancer(VAR_0) ...

correct:  x = 9; for k in range(5): avancer(x) ...
          ↓ anonymize (x→VAR_0, k→VAR_1)
          VAR_0 = 9; for VAR_1 in range(5): avancer(VAR_0) ...

After anonymization the only remaining difference is Const: 4 vs Const: 5 → LO_FOR_NUMBER_ITERATION_ERROR_UNDER2. No variable errors are produced.

Handles: Var: name and Var: -name (negated variable from UnaryOp). Does not anonymize Arg: (function parameter) or Const: nodes — only Var: nodes.

Limitation — ordering sensitivity: If two programs declare variables in a different order, the anonymous indices differ and a mismatch is still reported. This is intentional: declaration order is itself a structural property of the program.

AST → Custom Node Mapping

Python AST node	Custom Node label	Notes
ast.Module	"Module"
ast.For	"For"	Children: "Condition:" (with loop var + iter), "Body:"
ast.While / ast.If	"While" / "If"	Children: "Condition:", "Body:", optional "Else:"
ast.Assign	"Assign"	Children: Var: <name> + value node(s)
ast.BinOp / ast.AugAssign	"Operation: <op>"	op in {+, -, *, /, //, **, %}
ast.Compare	"Compare: <op>"	e.g. "Compare: <"
ast.Call	"Call: <func_name>"	Arguments are children
ast.Constant	"Const: <value>"	Strings quoted: "Const: 'hello'"
ast.Name	"Var: <name>"
ast.FunctionDef	"Function: <name>"
ast.Return	"Return"	Value nodes as children
ast.arg	"Arg: <name>"
ast.UnaryOp (USub)	"Const: -<val>" or "Var: -<name>"

node.get_path() returns the full root-to-self path list, e.g. ['Module', 'For[0]', 'Body:[1]', 'Call: print[0]', 'Operation: +[0]']. The [index] suffix is added for every non-root node.

Zhang-Shasha Edit Operations Format

Each op is a dict:

{'type': 'insert',  'path': [...],  'current': None,       'new': 'Call: print'}
{'type': 'delete',  'path': [...],  'current': 'Const: 1', 'new': None}
{'type': 'update',  'path': [...],  'current': 'Const: 9', 'new': 'Const: 10'}
{'type': 'match',   'path': [...],  'current': 'For',      'new': 'For'}

path = node.get_path() (includes the node itself as the last element).

---

Layer 1 — Primary Error Detection (error_annotation.py)

Called via ErrorAnnotation().concatenate_all_errors(ops). Runs 5 detectors then applies high-level filtering.

Detector 1 — detect_specific_missing_constructs → MISSING_* tags

Trigger: insert operation whose node is NOT also being deleted or updated elsewhere (truly absent, not moved).

Node type in insert['new']	Tag
FOR	MISSING_FOR_LOOP
WHILE	MISSING_WHILE_LOOP
CALL	MISSING_CALL_STATEMENT
IF	MISSING_IF_STATEMENT
ASSIGN	MISSING_ASSIGN_STATEMENT
FUNCTION	MISSING_FUNCTION_DEFINITION
RETURN	MISSING_RETURN
CONST	MISSING_CONST_VALUE
OPERATION	MISSING_OPERATION
ARG	MISSING_ARGUMENT
VAR	MISSING_VARIABLE

Output: (tag, insert['new'], context_path) where context_path = " > ".join(insert['path']).

Detector 2 — detect_unnecessary_deletions → UNNECESSARY_* tags

Trigger: delete operation whose node is NOT also being inserted elsewhere (truly redundant, not moved).

Node type in delete['current']	Tag
FOR	UNNECESSARY_FOR_LOOP
WHILE	UNNECESSARY_WHILE_LOOP
FUNCTION	UNNECESSARY_FUNCTION
RETURN	UNNECESSARY_RETURN_IN_FUNCTION
IF	UNNECESSARY_CONDITIONAL
CALL	UNNECESSARY_CALL_STATEMENT
ASSIGN	UNNECESSARY_ASSIGN_STATEMENT
CONST	UNNECESSARY_CONST_VALUE
OPERATION	UNNECESSARY_OPERATION
ARG	UNNECESSARY_ARGUMENT
VAR	UNNECESSARY_VAR

Output: (tag, value_after_colon_or_None, context_path) where context_path = " > ".join(delete['path']).

Important: delete['path'] INCLUDES the deleted node itself as the last path element. So context_path for an Operation: + node will end with "... > Operation: +[0]".

Detector 3 — detect_incorrect_statement_positions → INCORRECT_STATEMENT_POSITION_* tags

Trigger: A node is deleted from location A AND a node of the same kind+label is inserted at location B. The node was moved/misplaced. Context points to the TARGET (insert) location.

Kind	Tag
FOR	INCORRECT_STATEMENT_POSITION_FOR
WHILE	INCORRECT_STATEMENT_POSITION_WHILE
IF	INCORRECT_STATEMENT_POSITION_IF
CALL	INCORRECT_STATEMENT_POSITION_CALL
ASSIGN	INCORRECT_STATEMENT_POSITION_ASSIGN
FUNCTION	INCORRECT_STATEMENT_POSITION_FUNCTION
RETURN	INCORRECT_STATEMENT_POSITION_RETURN

Output: (tag, value_or_None, context_path).

Detector 4 — track_all_updates → UPDATE tags

Trigger: update operations (node label changed). Categorized by the updated node's type (last element of path):

Updated node type	Tag	Format
CONST	CONST_VALUE_MISMATCH	4-tuple: (tag, current, new, context)
COMPARE	INCORRECT_OPERATION_IN_CONDITION	4-tuple
OPERATION	INCORRECT_OPERATION_IN_ASSIGN	4-tuple
ASSIGN	NODE_TYPE_MISMATCH	4-tuple
VAR	skipped	—

Also handles node-TYPE replacements (e.g. FOR replaced by WHILE in update): emits UNNECESSARY_<old> + MISSING_<new>.

Detector 5 — detect_variable_mismatches → VARIABLE_MISMATCH

Trigger: A Var node is updated to different target names at different locations.

Output: ("VARIABLE_MISMATCH", var_name, context_path).

High-Level Filtering Rules

Applied after all detectors. Each rule suppresses child-context noise when a higher-level structural error is present:

Trigger tag	Suppresses
UNNECESSARY_CALL_STATEMENT at context C	All errors with context starting C >
UNNECESSARY_FOR_LOOP or UNNECESSARY_WHILE_LOOP at context C	All errors with context starting C >
MISSING_CALL_STATEMENT at context C	All errors with context starting C >
MISSING_FOR_LOOP at context C	All errors with context starting C >

---

Layer 2 — Typology Error Detection (error_checks.py)

get_customized_error_tags(input_list) → set of error tag strings.

Input: list of 3-tuples (tag, value, context) or 4-tuples (tag, current, new, context).

Pre-processing: Assignment presence/absence

process_tag_triplets() checks if all tags in a required set are simultaneously present with the same context segment. Used for:

• EXP_ERROR_ASSIGNMENT_MISSING: requires {MISSING_CONST_VALUE, MISSING_ASSIGN_STATEMENT, MISSING_VARIABLE} same context
• EXP_ERROR_ASSIGNMENT_UNNECESSARY: requires {UNNECESSARY_CONST_VALUE, UNNECESSARY_ASSIGN_STATEMENT, UNNECESSARY_VAR} same context

Rule: EXP_ERROR_OPERANDS

The operation expression exists in the student code, but there is a problem with its operands (the values or sub-expressions being operated on). Also covers cases where an entire operation is extra or missing, since the net effect is a wrong operand being evaluated. The operator type is NOT the issue here.

• Case 1 (wrong constant operand): tag == CONST_VALUE_MISMATCH AND context matches r"Operation:.*" → operation present in both trees but a constant operand has the wrong value (e.g. k+2 vs k+1)
• Case 2 (extra operation): tag == UNNECESSARY_OPERATION AND "Assign" NOT in context → student wrote an entire operation that should not be there (e.g. print(k+1) vs print(k))
• Case 3 (missing operation): tag == MISSING_OPERATION AND "Assign" NOT in context → student is missing an entire operation (e.g. print(k) vs print(k+1))
• Guard: Cases 2 and 3 inside "Assign" context are handled by VA_EXPRESSION_ASSIGNMENT_TO_VARIABLE_ERROR instead.

Rule: EXP_ERROR_OPERATOR

The operation expression exists in both student and correct code, but the operator type is wrong (e.g. student wrote k-1 but should have written k+1):

• tag == INCORRECT_OPERATION_IN_ASSIGN AND "Assign" NOT in context → the Operation node label changed between trees (operator type replaced), outside an assignment scope
• Guard: inside "Assign" context, VA_EXPRESSION_ASSIGNMENT_TO_VARIABLE_ERROR covers operator-type errors at the assignment level.

Rule: EXP_ERROR_OPERATION (umbrella)

Fires alongside both EXP_ERROR_OPERANDS and EXP_ERROR_OPERATOR — emitted on every case where an existing operation expression does not match what is expected, regardless of whether the problem is with the operands or the operator. Useful when only a coarse-grained signal is needed.

Distinction from assignment errors: EXP_ERROR_OPERATION, EXP_ERROR_OPERANDS, and EXP_ERROR_OPERATOR all imply the operation expression is structurally present. The assignment-level errors (EXP_ERROR_ASSIGNMENT_MISSING, EXP_ERROR_ASSIGNMENT_UNNECESSARY, EXP_ERROR_ASSIGNMENT_MISPLACED) are about the entire assignment expression being absent, extra, or in the wrong place.

Rule: EXP_ERROR_ASSIGNMENT_MISPLACED

• tag == INCORRECT_STATEMENT_POSITION_ASSIGN

Rule: EXP_ERROR_CONDITIONAL_BRANCH

• tag == INCORRECT_OPERATION_IN_CONDITION AND "If > Condition" in context

Rule: VA_DECLARATION_INITIALIZATION_ERROR

• tag == CONST_VALUE_MISMATCH AND context matches r".*Assign\s>\sConst:\s\d+$" (variable assigned wrong numeric literal)

Rule: VA_EXPRESSION_ASSIGNMENT_TO_VARIABLE_ERROR

Right-hand side expression of an assignment is wrong:

• tag == INCORRECT_OPERATION_IN_ASSIGN AND "Assign" in context
• tag == MISSING_OPERATION AND "Assign" in context AND MISSING_ASSIGN_STATEMENT not in all primary tags
• tag == UNNECESSARY_OPERATION AND "Assign" in context AND UNNECESSARY_ASSIGN_STATEMENT not in all primary tags

Rule: LO_FOR_NUMBER_ITERATION_ERROR / LO_FOR_NUMBER_ITERATION_ERROR_UNDER2

• tag == CONST_VALUE_MISMATCH AND "For > Condition: > Call: range > Const" in context
• Extract integers from context (student) and context2 (correct); |diff| > 1 → LO_FOR_NUMBER_ITERATION_ERROR, |diff| == 1 → UNDER2

Rule: LO_WHILE_NUMBER_ITERATION_ERROR / LO_WHILE_NUMBER_ITERATION_ERROR_UNDER2

• tag == CONST_VALUE_MISMATCH AND "While > Condition: > Compare" in context; same diff rule

Rule: LO_FOR_MISSING / LO_WHILE_MISSING

• tag == MISSING_FOR_LOOP / tag == MISSING_WHILE_LOOP

Rule: LO_FOR_UNNECESSARY / LO_WHILE_UNNECESSARY

• tag == UNNECESSARY_FOR_LOOP AND (context matches r".*For$" OR context2 ends with "For")
• tag == UNNECESSARY_WHILE_LOOP AND (context matches r".*While$" OR context2 ends with "While")

Rule: LO_FOR_MISPLACED

• tag == INCORRECT_STATEMENT_POSITION_FOR

Rule: LO_BODY_MISSING_NOT_PRESENT_ANYWHERE

• Any MISSING_* tag AND ("For > Body" OR "While > Body") in context

Rule: LO_BODY_MISPLACED

• Any INCORRECT_STATEMENT_POSITION tag AND "For > Body" in context

Rule: LO_BODY_ERROR

Any of these (tag + context) combinations:

• MISSING_CONST_VALUE + "For > Body"
• MISSING_CALL_STATEMENT + "For > Body"
• UNNECESSARY_CALL_STATEMENT + "For > Body"
• CONST_VALUE_MISMATCH + "While > Body"
• INCORRECT_STATEMENT_POSITION_ASSIGN + "For > Body"

Rule: LO_CONDITION_ERROR

• tag == INCORRECT_OPERATION_IN_CONDITION AND "While > Condition" in context

Rule: F_CALL_MISSING / F_CALL_MISSING_<NAME>

• tag == MISSING_CALL_STATEMENT
• Identify function name from context2's last word
• Known names → specific tags: F_CALL_MISSING_PRINT, F_CALL_MISSING_AVANCER, F_CALL_MISSING_TOURNER, F_CALL_MISSING_COULEUR, F_CALL_MISSING_ARC, F_CALL_MISSING_GAUCHE, F_CALL_MISSING_HAUT, F_CALL_MISSING_BAS, F_CALL_MISSING_DROITE, F_CALL_MISSING_POSER, F_CALL_MISSING_LEVER
• Unknown → F_CALL_MISSING

Rule: F_CALL_UNNECESSARY / F_CALL_UNNECESSARY_<NAME>

• tag == UNNECESSARY_CALL_STATEMENT
• Identified by context2 last word OR context regex; same known-name list

Rule: F_CALL_PRINT_ERROR_ARG

Error inside print's argument list:

• tag != UNNECESSARY_CALL_STATEMENT AND context matches r"Call:\s*print > .*" (wrong/extra element inside print args)
• tag in {MISSING_CONST_VALUE, MISSING_VARIABLE, MISSING_ARGUMENT, MISSING_OPERATION} AND context matches r"Call:\s*print" (missing element inside print)

Rule: F_CALL_<NAME>_ERROR (design + robot functions)

• Design (avancer, tourner, couleur, arc): tag not in [VARIABLE_MISMATCH] AND context matches r"Call:\s*<name> > .*"
• Robot (gauche, haut, bas, droite, lever, poser): same rule with respective names

Rule: F_CALL_INCORRECT_POSITION_<NAME>

• tag == INCORRECT_STATEMENT_POSITION_CALL AND context matches r"Call:\s*<name>" for each known function
• Known names: print, haut, bas, gauche, droite, tourner, avancer, lever, poser, arc, couleur

Rule: F_DEFINITION_MISSING / F_DEFINITION_UNNECESSARY

• tag == MISSING_FUNCTION_DEFINITION → F_DEFINITION_MISSING
• tag == UNNECESSARY_FUNCTION → F_DEFINITION_UNNECESSARY

Rule: F_DEFINITION_ERROR_RETURN

• tag == MISSING_RETURN
• tag == UNNECESSARY_RETURN_IN_FUNCTION
• tag == MISSING_VARIABLE AND "Function" in context AND "Return > Tuple" in context

Rule: CS_MISSING

• tag == MISSING_IF_STATEMENT

Rule: CS_BODY_ERROR

• Any MISSING_* tag AND "If > Body" in context

Rule: CS_BODY_MISPLACED

• tag == INCORRECT_STATEMENT_POSITION_IF

---

Context Path Format

Context paths are " > "-separated strings. The node itself is included as the last element.

"Module > For > Body: > Call: print"
"Module > For > Condition: > Call: range > Const: 10"
"Module > For > Body: > Call: print > Operation: +"
"Module > Assign > Const: 5"

When the path comes from detect_unnecessary_deletions / detect_specific_missing_constructs, index suffixes like [0] are present. structural_path_element() strips them for display.

---

Exception Tag Lists (constants.py)

EXP_ERROR_OPERATION_EXCEPTION_ANNOTATION_TAGS     = [VARIABLE_MISMATCH]
F_CALL_PRINT_ERROR_ARG_EXCEPTION_ANNOTATION_TAGS  = [VARIABLE_MISMATCH, UNNECESSARY_CALL_STATEMENT]
F_CALL_DESIGN_ERROR_ARG_EXCEPTION_ANNOTATION_TAGS = [VARIABLE_MISMATCH]
F_CALL_ROBOT_ERROR_ARG_EXCEPTION_ANNOTATION_TAGS  = [VARIABLE_MISMATCH]

---

Changelog

Fix — EXP_ERROR_OPERATION not detected for print(k+1) vs print(k) (2026-03)

Root cause: The original rule only fired for CONST_VALUE_MISMATCH + "Operation:" in context, which handles the case where the operator is wrong inside an existing operation (e.g. k+2 vs k+1 → Const update). When the operation is entirely extra (UNNECESSARY_OPERATION) or entirely missing (MISSING_OPERATION), the tag never matched.

Fix (then refactored into split below): Added handling for UNNECESSARY_OPERATION and MISSING_OPERATION tags. The Assign guard prevents double-reporting with VA_EXPRESSION_ASSIGNMENT_TO_VARIABLE_ERROR.

Refactor — Split EXP_ERROR_OPERATION into EXP_ERROR_OPERANDS + EXP_ERROR_OPERATOR (2026-03)

Motivation: EXP_ERROR_OPERATION conflated two distinct error kinds — wrong operand values vs wrong operator type. The split makes the distinction precise and actionable.

Old tag	New tag	When
EXP_ERROR_OPERATION (Case 1)	EXP_ERROR_OPERANDS	CONST_VALUE_MISMATCH inside an Operation context
EXP_ERROR_OPERATION (Case 2)	EXP_ERROR_OPERANDS	UNNECESSARY_OPERATION outside Assign
EXP_ERROR_OPERATION (Case 3)	EXP_ERROR_OPERANDS	MISSING_OPERATION outside Assign
(new)	EXP_ERROR_OPERATOR	INCORRECT_OPERATION_IN_ASSIGN outside Assign

EXP_ERROR_OPERATION constant is kept in constants.py as deprecated. EXP_ERROR_OPERANDS and EXP_ERROR_OPERATOR are the canonical replacements.

Semantic contract:

• All three errors imply the operation expression structurally exists in the student code.
• EXP_ERROR_OPERATION = umbrella — fires on every operation error (operands OR operator wrong).
• EXP_ERROR_OPERANDS = right operator, wrong values/variables being combined.
• EXP_ERROR_OPERATOR = right operands, wrong operator type (+ vs -, etc.).
• Assignment-level errors (EXP_ERROR_ASSIGNMENT_*) = expression wholly absent / extra / misplaced.
• EXP_ERROR_OPERATION always co-fires with whichever specific tag applies.

Fix — F_CALL_INCORRECT_POSITION_COULEUR not emitted for misplaced couleur call (2026-03)

Root cause: couleur was missing from two places:

1. No F_CALL_INCORRECT_POSITION_COULEUR constant in constants.py.
2. No entry for couleur in the incorrect_position_tags dispatch list in error_checks.py.

The primary layer (detect_incorrect_statement_positions) correctly produced INCORRECT_STATEMENT_POSITION_CALL when Z-S emitted a delete+insert pair for Call: couleur (distance 8). The typology layer simply had no rule to convert it into a specific tag, so the result was empty.

Fix: Added F_CALL_INCORRECT_POSITION_COULEUR = "F_CALL_INCORRECT_POSITION_COULEUR" to constants.py and added (ANNOTATION_CONTEXT_CALL_NATIVE_FUNCTION_COULEUR, F_CALL_INCORRECT_POSITION_COULEUR) to the incorrect_position_tags list in error_checks.py.

Example:

incorrect: couleur(255,0,0)\nfor k in range(6): arc(k+1,90)\navancer(4)\nfor k in range(3): arc(3-k,180)
correct:   for k in range(6): arc(k+1,90)\ncouleur(255,0,0)\navancer(4)\nfor k in range(3): arc(3-k,180)
output:    {'F_CALL_INCORRECT_POSITION_COULEUR'}

---

Fix — Spurious UNNECESSARY_VARIABLE errors from variable-name differences (2026-03)

Root cause: When comparing two programs that use different variable names for the same structural role (e.g. n in student code vs x in correct code), Zhang-Shasha aligned Var: n with Var: x and produced one UNNECESSARY_VARIABLE update per occurrence of the variable — masking the real errors.

Fix — Variable anonymization (node_functions.anonymize_variable_names): Before the Zhang-Shasha distance call, each tree is independently traversed in pre-order and every Var: <name> node is renamed to Var: VAR_k where k is the order of first encounter. Both Var: name and Var: -name (negated variable) are handled.

code1  n=9; for k in range(4): avancer(n) ...
       → after anonymization → VAR_0=9; for VAR_1 in range(4): avancer(VAR_0) ...

code2  x=9; for k in range(5): avancer(x) ...
       → after anonymization → VAR_0=9; for VAR_1 in range(5): avancer(VAR_0) ...

The only remaining difference is Const: 4 vs Const: 5 → LO_FOR_NUMBER_ITERATION_ERROR_UNDER2.

Why independent anonymization works: Variables that play the same structural role appear in the same position in both programs, so they receive the same anonymous index. Genuine structural variable mismatches (e.g. a program that uses two variables where only one is expected) still produce differences after anonymization.

Location: node_functions.anonymize_variable_names(root) called in error_diagnosis.get_primary_code_errors() after tree construction, before distance().