SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2403.16218 · 2024

CoverUp: Effective High Coverage Test Generation for Python

arXiv · PDF · Open in the Atlas

Code that ran

12 of 17 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 1 honoured the contract we drafted, 1 violated it and 1 has no contract verdict; 9 failures of our instrument; 5 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2403.16218.json · badge: /api/ran/2403.16218.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
plasma-umass/coverup canonical (classifier label)
classifier basis: census
9 ran · 5 no recorded run
plasma-umass/coverup-eval canonical (classifier label)
classifier basis: census
3 ran · 0 no recorded run
FunctionStatusWhere it lives
basedir Ran
ran; our contract draft was wrong, not the code
plasma-umass/coverup-eval/output/cm/ansible/coverup-tests/disabled_test_coverup_1137.py
code served (permissive licence) · get_code("e780f2ac9b06a8c2")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

def basedir(source):
    return '/fake/dir'

ARGS = {
    'source': '/some/path'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
count_tokens Ran
ran, honoured the contract we drafted
plasma-umass/coverup/src/coverup/llm.py
code served (permissive licence) · get_code("145acc26153dc1f4")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned int, ndim 0, dtype family int, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# The function uses:
# - model_name: str - passed to token_counter
# - completion: dict with 'messages' key - messages are passed to token_counter
# The token_counter function expects messages in a chat format

ARGS = {
    "model_name": "gpt-3.5-turbo",
    "completion": {
        "messages": [
            {"role": "user", "content": "Hello, how are you?"},
            {"role": "assistant", "content": "I'm doing well, thank you for asking!"}
        ]
    }
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
litellm, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type int, ndim 0, dtype family int, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
custom_object_representer Ran
ran; our contract draft was wrong, not the code
plasma-umass/coverup-eval/output/cm/ansible/coverup-tests/disabled_test_coverup_1108.py
code served (permissive licence) · get_code("ae89c975e4a8f393")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ScalarNode. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import yaml

torch.manual_seed(0)

class CustomDumper(yaml.Dumper):
    pass

class CustomObject:
    pass

dumper = CustomDumper(stream=None)
data = CustomObject()

ARGS = {
    'dumper': dumper,
    'data': data
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, yaml — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ScalarNode — a ScalarNode output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a ScalarNode output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
format_ranges Ran
ran; our contract draft was wrong, not the code
plasma-umass/coverup/src/coverup/utils.py
code served (permissive licence) · get_code("ecca514f32bdaa28")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import typing as T

# The function format_ranges takes two parameters:
# - lines: T.Set[int] - a set of line numbers to format
# - negative: T.Set[int] - a set of line numbers to exclude from ranges

# Create test sets of integers
lines = {1, 2, 3, 5, 6, 10, 15, 16, 17}
negative = {4, 7, 8, 9}

ARGS = {
    "lines": lines,
    "negative": negative,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_global_imports Ran
ran; our contract draft was wrong, not the code
plasma-umass/coverup/src/coverup/codeinfo.py
code served (permissive licence) · get_code("6f9ea5e7c5e3e194")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import ast
from pathlib import Path
import tempfile
import sys

# Create a temporary directory to simulate a package structure
temp_dir = tempfile.mkdtemp()
temp_path = Path(temp_dir)

# Add temp_dir to sys.path so _package_path can find it
sys.path.insert(0, temp_dir)

# Create a simple package structure
package_dir = temp_path / "mypackage"
package_dir.mkdir()
(package_dir / "__init__.py").write_text("")

# Create a test module file within the package
test_module_file = package_dir / "test_module.py"
test_module_code = """
import os
from pathlib import Path
from typing import List

def my_function(x: int) -> str:
    return str(x)

class MyClass:
    def __init__(self):
        pass
"""
test_module_file.write_text(test_module_code)

# Parse the module code into an AST
parsed_ast = ast.parse(test_module_code)

# Create a Module instance with the parsed AST and the file path
module = ast.Module(parsed_ast.body, parsed_ast.type_ignores)
custom_module = type('Module', (ast.Module,), {
    'path': test_module_file,
    '__init__': lambda self, body, type_ignores: ast.Module.__init__(self, body, type_ignores)
})(parsed_ast.body, parsed_ast.type_ignores)
custom_module.path = test_module_file

# Create a node to search for - we'll use a simple Name node
# that references one of the imported names
node = ast.parse("os.getcwd()").body[0]

ARGS = {
    'module': custom_module,
    'node': node,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
is_same_as_P Ran
ran, violated the contract we drafted
plasma-umass/coverup/src/coverup/logreader.py
code served (permissive licence) · get_code("218a7174bbbafd7c")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned bool, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import re

def is_same_as_P(content, begin, end):
    """Attempts to detect 'C' (context) prompts that were only issued as context prompts
       because their 'def' executed when their module was loaded to 'P' (initial) prompts,
       which is what they should have been."""
    begin, end = (int(begin), int(end))
    if (rng := re.search('^when tested, lines (\\d+)-(\\d+) do not execute', content, re.M)) and (py := re.search('```python\\n(.*)```', content, re.S)):
        first, last = (int(rng.group(1)), int(rng.group(2)))

        def del_line_markup(s):
            line_markup_len = 12
            s = '\n'.join((l[line_markup_len:] for l in s.splitlines()))
            import textwrap
            return textwrap.dedent(s)
        try:
            import ast
            tree = ast.parse(del_line_markup(py.group(1)))
            block = next(iter((node for node in ast.walk(tree) if isinstance(node, (ast.FunctionDef, ast.AsyncFunctionDef)))), None)
            if block is None:
                block = next(iter((node for node in ast.walk(tree) if isinstance(node, ast.ClassDef))))
        except Exception:
            print('----')
            print(py.group(1))
            print('----')
            print(del_line_markup(py.group(1)))
            print('----')
            raise
        if ast.get_docstring(block, clean=False) is not None:
            block.body.pop(0)
        for first_stmt in block.body:
            if not isinstance(first_stmt, (ast.Global, ast.Nonlocal)):
                break
        last_stmt = block.body[-1]
        block.lineno = min([block.lineno, *(d.lineno for d in block.decorator_list)])
        if begin - 1 + first_stmt.lineno == first and (begin - 1 + last_stmt.lineno <= last and begin - 1 + last_stmt.end_lineno >= last):
            return True
        elif begin - block.lineno + first_stmt.lineno == first and (begin - block.lineno + last_stmt.lineno <= last and begin - block.lineno + last_stmt.end_lineno >= last):
            return True
    return False


content = """when tested, lines 2-5 do not execute


            def example_func():
                x = 1
                y = 2
                z = 3
                return x + y + z

"""

begin = 1
end = 10

ARGS = {
    "content": content,
    "begin": begin,
    "end": end,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type bool, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
lines_branches_do Ran
ran; our contract draft was wrong, not the code
plasma-umass/coverup/src/coverup/utils.py
code served (permissive licence) · get_code("ec9b4903be33464d")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

# Setup code for format_ranges and lines_branches_do

ARGS = {
    "lines": {1, 3, 5, 7},
    "neg_lines": {4, 6},
    "branches": {(10, 11), (12, 0), (15, 16)}
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
parse_file Ran
ran; our contract draft was wrong, not the code
plasma-umass/coverup/src/coverup/codeinfo.py
code served (permissive licence) · get_code("7bb4f87378c340a8")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Module. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import ast
from pathlib import Path
import tempfile

# Create a temporary Python file with valid Python code
with tempfile.NamedTemporaryFile(mode='w', suffix='.py', delete=False) as f:
    f.write("""
def hello_world():
    x = 42
    return x

class MyClass:
    def __init__(self):
        self.value = 10
""")
    temp_file_path = Path(f.name)

ARGS = {
    "file": temp_file_path
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Module — a Module output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Module output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
remove_omit Ran
ran; our contract draft was wrong, not the code
plasma-umass/coverup-eval/output/cm/ansible/coverup-tests/disabled_test_coverup_1123.py
code served (permissive licence) · get_code("e134f05ecc5345c7")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# The function remove_omit takes a dictionary and an omit_token,
# and returns a new dictionary with all entries where v != omit_token removed.
# It uses:
# - d.items() to iterate over key-value pairs
# - comparison operator != with omit_token

# Create a simple dictionary with various values
d = {
    "key1": "value1",
    "key2": None,
    "key3": 42,
    "key4": None,
    "key5": "another_value"
}

# The omit_token is what we want to filter out
omit_token = None

ARGS = {
    "d": d,
    "omit_token": omit_token
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
subprocess_run Ran
ran; our contract draft was wrong, not the code
plasma-umass/coverup/src/coverup/utils.py
code served (permissive licence) · get_code("a6bcfa534306ab26")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned coroutine. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import typing as T
import subprocess
import asyncio

# This function runs async subprocess commands
# We need to provide valid arguments

ARGS = {
    "args": ["echo", "hello world"],
    "check": False,
    "timeout": None,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type coroutine — a coroutine output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a coroutine output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
test_file_path Ran
ran; our contract draft was wrong, not the code
plasma-umass/coverup/src/coverup/coverup.py
code served (permissive licence) · get_code("28aa58f0339c64af")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned PosixPath. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

from pathlib import Path

class Args:
    def __init__(self):
        self.tests_dir = Path("/tmp/tests")
        self.prefix = "example"

ARGS = {
    "args": Args(),
    "test_seq": 42
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type PosixPath — a PosixPath output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a PosixPath output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
token_rate_limit_for_model Ran
ran on a synthesized input
plasma-umass/coverup/src/coverup/llm.py
code served (permissive licence) · get_code("f840bd6712a9a03f")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple, ndim 1, dtype family int; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import typing as T
import warnings

MODEL_RATE_LIMITS = {'gpt-3.5-turbo': {'token': (50000000, 60), 'request': (10000, 60)}, 'gpt-3.5-turbo-0125': {'token': (50000000, 60), 'request': (10000, 60)}, 'gpt-3.5-turbo-1106': {'token': (50000000, 60), 'request': (10000, 60)}, 'gpt-3.5-turbo-16k': {'token': (50000000, 60), 'request': (10000, 60)}, 'gpt-3.5-turbo-instruct': {'token': (90000, 60), 'request': (3500, 60)}, 'gpt-3.5-turbo-instruct-0914': {'token': (90000, 60), 'request': (3500, 60)}, 'gpt-4': {'token': (1000000, 60), 'request': (10000, 60)}, 'gpt-4-0314': {'token': (1000000, 60), 'request': (10000, 60)}, 'gpt-4-0613': {'token': (1000000, 60), 'request': (10000, 60)}, 'gpt-4-32k-0314': {'token': (150000, 60), 'request': (1000, 60)}, 'gpt-4-turbo': {'token': (2000000, 60), 'request': (10000, 60)}, 'gpt-4-turbo-2024-04-09': {'token': (2000000, 60), 'request': (10000, 60)}, 'gpt-4-turbo-preview': {'token': (2000000, 60), 'request': (10000, 60)}, 'gpt-4-0125-preview': {'token': (2000000, 60), 'request': (10000, 60)}, 'gpt-4-1106-preview': {'token': (2000000, 60), 'request': (10000, 60)}, 'gpt-4o': {'token': (30000000, 60), 'request': (10000, 60)}, 'gpt-4o-2024-05-13': {'token': (30000000, 60), 'request': (10000, 60)}, 'gpt-4o-2024-08-06': {'token': (30000000, 60), 'request': (10000, 60)}, 'gpt-4o-mini': {'token': (150000000, 60), 'request': (30000, 60)}, 'gpt-4o-mini-2024-07-18': {'token': (150000000, 60), 'request': (30000, 60)}, 'bedrock/anthropic.claude-3-haiku-20240307-v1:0': {'token': (1000000, 60), 'request': (10000, 60)}, 'bedrock/anthropic.claude-3-sonnet-20240229-v1:0': {'token': (1000000, 60), 'request': (10000, 60)}, 'bedrock/anthropic.claude-3-opus-20240229-v1:0': {'token': (1000000, 60), 'request': (10000, 60)}, 'anthropic.claude-3-5-sonnet-20241022-v2:0': {'token': (400000, 50), 'request': (10000, 50)}, 'us.anthropic.claude-3-5-sonnet-20241022-v2:0': {'token': (400000, 50), 'request': (10000, 50)}, 'us.anthropic.claude-3-7-sonnet-20250219-v1:0': {'token': (400000, 50), 'request': (10000, 50)}}

def token_rate_limit_for_model(model_name: str) -> T.Tuple[int, int] | None:
    if model_name.startswith('openai/'):
        model_name = model_name[7:]
    if (model_limits := MODEL_RATE_LIMITS.get(model_name)):
        limit = model_limits.get('token')
        if not 'anthropic' in model_name:
            try:
                import tiktoken
                tiktoken.encoding_for_model(model_name)
            except KeyError:
                warnings.warn(f'Unable to get encoding for {model_name}; will ignore rate limit')
                return None
        return limit
    return None

ARGS = {
    'model_name': 'gpt-4o-mini'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
tiktoken — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 1, dtype family int — a tuple output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
compute_cost Not yet run
unverified (harvested, no recorded run)
plasma-umass/coverup/src/coverup/llm.py
code served (permissive licence) · get_code("def8599068e3dbf6")
How it ran

no invocation recorded.

get_info Not yet run
unverified (harvested, no recorded run)
plasma-umass/coverup/src/coverup/codeinfo.py
code served (permissive licence) · get_code("9b6d481e57689ee5")
How it ran

no invocation recorded.

get_missing_coverage Not yet run
unverified (harvested, no recorded run)
plasma-umass/coverup/src/coverup/segment.py
code served (permissive licence) · get_code("cc43dbef7115b69c")
How it ran

no invocation recorded.

new_test_file Not yet run
unverified (harvested, no recorded run)
plasma-umass/coverup/src/coverup/coverup.py
code served (permissive licence) · get_code("d3e7c47f194e80d7")
How it ran

no invocation recorded.

parse_log Not yet run
unverified (harvested, no recorded run)
plasma-umass/coverup/src/coverup/logreader.py
code served (permissive licence) · get_code("f5192e3a7fcb2025")
How it ran

no invocation recorded.

Also present in other papers' code

Of 17 samples lifted from this paper's linked repositories, 17 carry a function identity and 0 do not. Of the 17 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 17 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Testing is an essential part of software development. Test generation tools attempt to automate the otherwise labor-intensive task of test creation, but generating high-coverage tests remains challenging. This paper proposes CoverUp, a novel approach to driving the generation of high-coverage Python regression tests. CoverUp combines coverage analysis, code context, and feedback in prompts that iteratively guide the LLM to generate tests that improve line and branch coverage. We evaluate our prototype CoverUp implementation across a benchmark of challenging code derived from open-source Python projects and show that CoverUp substantially improves on the state of the art. Compared to CodaMosa, a hybrid search/LLM-based test generator, CoverUp achieves a per-module median line+branch coverage of 80% (vs. 47%). Compared to MuTAP, a mutation- and LLM-based test generator, CoverUp achieves an overall line+branch coverage of 89% (vs. 77%). We also demonstrate that CoverUp's performance stems not only from the LLM used but from the combined effectiveness of its components.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2403.16218")
get_code_for_paper("2403.16218")
have("2403.16218")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2403.16218.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2403.16218.svg)](https://syntology.ai/paper/2403.16218)

Connect an agent — have() is free.