Spaces:
Running
Running
File size: 13,892 Bytes
333a6de 3630362 333a6de 17cd01c 3630362 08fdde3 3630362 08fdde3 3630362 17cd01c 333a6de 3630362 333a6de 3630362 333a6de 3630362 333a6de 3630362 333a6de 3630362 333a6de 715e9c8 333a6de 3630362 333a6de 3630362 333a6de 3630362 333a6de 17cd01c 333a6de 17cd01c 333a6de 3630362 17cd01c 333a6de |
1 2 3 4 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 45 46 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 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 |
import json
import copy
import openai
"""
We will handle here the code evaluation phase.
"""
EVAL_ANSWER_NOEVAL = 0
EVAL_ANSWER_POSITIVE = 1
EVAL_ANSWER_NEGATIVE = -1
CODE_AUGMENTATIONS=[
("NO_COMPILE", "The code provided is not a valid C code"),
("DRY", "Don't repeat yourself."),
("SRP", "Single object[function] responsability"),
("MC", "Magic constants."),
("NAME", "Meaningful names in the code."),
]
# GPT prompt, ver 1, OpenAI.
# TODO Store prompt version, endpoint, endpoint version, etc. once we can afford to do A/B testing.
gpt_teacher_prompt = \
"""
You are a helpful teacher that evaluates pieces of C code based on several clean code principles. The principles will
be provided below. As output you will provide a list of JSON objects.
The language is C. The students will give you a complete C code.
You will evaluate its cleaniness according to some criterias given below. A C program can have multiple criterias that are infringed.
Each principles has a mnemonic and an explanation. You will have to check if the code fits the principles,
based on the explanations. The principles matches if most of the code is like in the explanation. If this is the case, issue the mnemonic.
As an **input**, the student will send you a piece of code.
As an **answer**, you should generate a list of JSON objects. Each JSON object should contain two parameters: "criteria" and "explanation".
The "criteria" should be a programming principle or acronym as above, and the "explanation" should provide a brief list for the student
on how their code fail.
#################
Criterias
NO_COMPILE
The code is not a standalone compilable C code. There are no ``#include`` directives. There is no main function. The C syntax looks
broken. There are wrong string constants. There are illegal operands or keywords. Only flagrant violations would trigger this criteria.
This criteria should be checked on "best effort", that is, don't start a compiler and compile the code. As explanations you should return
a short description on why the code would not compile. Example: "There is no ``main()`` function in the code. Please provide one."
DRY
Don't repeat yourself. The code has repetitive blocks that could have been isolated into a function. The code has no functions
but some instructions could have been parametrized. The data is stored in flat variables but a struct or an array would have been
more natural for that problem. As explanations you should present and a code fragment that is repeated followed by a short list
of function names or programming patterns that should be present in the code.
SRP
Single "object" Responsability but in our case, Single Function Responsability. The code has functions that do more than one thing.
For example, reading from a file, then computing something on that data. Allocating, reading, computing and printing in the same function.
Or the code is stuffend in main(), without any separation in functions. As explanations you should return one or two function names
and a list of their responsabilities. Tell the student that they should separate the functionalities.
MC
Magic constants. The code has for loops, data allocations, bounds, etc. that are expressed as numeric constants in the place
they are used. These constants appear in more than one place. If most of the constants are written directly in the code, then
the given piece of code matches the criteria. If most of the constants are declared as symbolic constants then, the code, does
not match this criteria. The naming of the constants is not important for this criteria. The naming and its relevance is judged by
other topics. For this criteria it matters only if the constant is defined or not. As explanations you will propose
some symbolic constants that should be used in the code.
NAME
Nondescriptive variable/function names. The code has meaningless function and variable names. For example, t, x, y, s, for
vectors, structures or tables that store meaningful data. Function names like ``add`` or ``get`` without specifying what will be
added, or what will be got. A good function name is ``addStudentInList`` or ``getStudentsSortedByCriteria``. Variable names used
in for loops or local to some functions are ok to be less descriptive. If most of the names are bad, then the code matches this criteria.
As explanations you will return few pairs of original-name suggested-name where the original-name is the original variable in the code
and the suggested-name is something that would be meaningful given what the code does.
#################
Example
```
#include <stdio.h>
#include <stdlib.h>
#define SIZE 10
void func(int* arr) {
arr[0] = 0 * 0; printf("%d ", arr[0]);
arr[1] = 1 * 1; printf("%d ", arr[1]);
arr[2] = 2 * 2; printf("%d ", arr[2]);
arr[3] = 3 * 3; printf("%d ", arr[3]);
arr[4] = 4 * 4; printf("%d ", arr[4]);
arr[5] = 5 * 5; printf("%d ", arr[5]);
arr[6] = 6 * 6; printf("%d ", arr[6]);
arr[7] = 7 * 7; printf("%d ", arr[7]);
arr[8] = 8 * 8; printf("%d ", arr[8]);
arr[9] = 9 * 9; printf("%d ", arr[9]);
printf("\\n");
}
int main() {
int* arr = (int*)malloc(sizeof(int) * SIZE);
func(arr);
free(arr);
return 0;
}
```
Your output:
[
{
"criteria": "DRY",
"explanation": "The ``arr[3] = 3 * 3; printf("%d ", arr[3]);`` code repeats a lot. Consider creating functions like ``computeArray``, ``displayArray``. Consider using ``for`` loops."
},
{
"criteria": "SRP",
"explanation": "The function ``func`` handles the array computations and output. You should separate the responsabilites."
},
{
"criteria": "NAME",
"explanation": "``func`` should be called ``computeAndPrint``, ``arr`` should be called ``dataArray``."
}
]
"""
CODE_EVAL_EXAMPLES = [
{"name":"Hello World",
"code":"""
#include <stdio.h>
int main() {
printf("Hello, World!");
return 0;
}
""",
"eval":"[]"},
{"name":"Bookstore DB",
"code":"""
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#define A 100
#define B 50
int main() {
char x1[A] = "The Great Gatsby";
char y1[B] = "F. Scott Fitzgerald";
int z1 = 1925;
char* p1 = (char*)malloc(A * sizeof(char));
char* p2 = (char*)malloc(B * sizeof(char));
int* p3 = (int*)malloc(sizeof(int));
if (p1 != NULL && p2 != NULL && p3 != NULL) {
strncpy(p1, x1, A);
strncpy(p2, y1, B);
*p3 = z1;
printf("Title: %s\\n", p1);
printf("Author: %s\\n", p2);
printf("Year: %d\\n", *p3);
}
free(p1);
free(p2);
free(p3);
return 0;
}""",
"eval":"""[
{
"criteria": "DRY",
"explanation": "The memory allocation and initialization for ``p1``, ``p2``, and ``p3`` are repetitive. Consider creating a function like ``allocateMemory`` to handle this."
},
{
"criteria": "SRP",
"explanation": "The ``main`` function handles memory allocation, data copying, and printing. You should separate these responsibilities into different functions like ``allocateMemory``, ``copyData``, and ``printData``."
},
{
"criteria": "NAME",
"explanation": "``x1`` should be called ``title``, ``y1`` should be called ``author``, ``z1`` should be called ``year``, ``p1`` should be called ``titlePtr``, ``p2`` should be called ``authorPtr``, and ``p3`` should be called ``yearPtr``."
}
]"""},
]
def get_enhanced_sample_example(example_id):
example_id = min(example_id, len(CODE_EVAL_EXAMPLES)-1)
example = copy.deepcopy(CODE_EVAL_EXAMPLES[example_id])
gpt_eval = json.loads(example["eval"])
enhanced_eval = add_evaluation_fields_on_js_answer(gpt_eval)
example["eval"] = enhanced_eval
return example
def get_the_openai_client(openai_key):
if openai_key is None or openai_key == "" or openai_key == "(none)":
return None
client = openai.Client(api_key=openai_key)
return client
def clean_prompt_answer(answer):
"""
Chatgpt4 is ok, does not pollute the code but 3.5 encloses it in ```
:param answer:
:return:
"""
cleaned = []
for l in answer.split("\n"):
if l.startswith("```"):
continue
else:
cleaned.append(l)
return "\n".join(cleaned)
def parse_chatgpt_answer(ans_text):
"""
Minimal parsing of chatGPT answer.
TODO: Ensure no extra fields, ensure format, etc. (cough pydantic)
:param ans_text:
:return:
"""
try:
ans_text = clean_prompt_answer(ans_text)
js_answer = json.loads(ans_text)
except:
# for now we dump the error in console
import traceback
exception = traceback.format_exc()
print(exception)
return {f"ChatPGT answered: {ans_text}.\nError":exception}
return js_answer
def make_user_prompt(code_fragment):
"""
Formats the code for user prompt
:param code_fragment:
:return:
"""
user_prompt = f"```\n{code_fragment}```"
return user_prompt
def call_openai(client, system_prompt, user_prompt, model="gpt-3.5-turbo", temperature = 0.1, force_json=False,
get_full_response=False):
"""
Constructs a prompt for chatGPT and sends it.
Possible models:
- gpt-3.5-turbo
- gpt-4-turbo
- gpt-4o
- gpt-4o-mini
Don't use force_json, will screw common use cases like list of json objects.
:param client:
:param system_prompt:
:param user_prompt:
:param model:
:param temperature:
:return:
"""
messages = [
{"role": "system", "content": system_prompt},
{"role": "user", "content": user_prompt}
]
optional_args = {}
if force_json:
optional_args["response_format"] = { "type": "json_object" }
response = client.chat.completions.create(
# model="gpt-4-turbo", # Update model name as necessary
model=model,
messages=messages,
temperature = temperature,
**optional_args
)
if get_full_response:
return response
else:
output_content = response.choices[0].message.content
return output_content
def eval_code_by_chatgpt(openai_client, ccode):
"""
Will evaluate a piece of code using our heavily tuned prompt!
:param openai_client:
:param ccode:
:return:
"""
# time.sleep(3)
try:
# return """[
# {
# "criteria": "DRY",
# "explanation": "The memory allocation and initialization for ``p1``, ``p2``, and ``p3`` are repetitive. Consider creating a function like ``allocateAndInitializeMemory``."
# },
# {
# "criteria": "DRY",
# "explanation": "Tne second DRY failure, because this is the observed ChatGPT behaviour."
# },
#
# {
# "criteria": "SRP",
# "explanation": "The ``main`` function handles memory allocation, initialization, and printing. You should separate these responsibilities into different functions like ``allocateMemory``, ``initializeData``, and ``printData``."
# },
# {
# "criteria": "NAME",
# "explanation": "``x1`` should be called ``title``, ``y1`` should be called ``author``, ``z1`` should be called ``year``, ``p1`` should be called ``titlePtr``, ``p2`` should be called ``authorPtr``, ``p3`` should be called ``yearPtr``."
# }
# ]"""
assert openai_client is not None
user_prompt = make_user_prompt(ccode)
chatgpt_answer = call_openai(openai_client, system_prompt=gpt_teacher_prompt, user_prompt=user_prompt,
model="gpt-4o",
temperature=0, force_json=False, get_full_response=False)
return chatgpt_answer
except:
import traceback
traceback.print_exc()
return {"error":"There was an error while getting the ChatGPT answer. Maybe ChatGPT is overloaded?"}
def add_evaluation_fields_on_js_answer(json_answer, all_criterias = None):
"""
Adds some JSON fields to store the human feedback.
The textual human feedback will always be in the 0 position.
:param json_answer:
:return:
"""
if all_criterias is None:
all_criterias = CODE_AUGMENTATIONS
enhanced_answer = []
overall_feedback = {
"criteria":"HUMAN_FEEDBACK",
"explanation":"",
"EVAL": EVAL_ANSWER_NOEVAL
}
if all_criterias is not None:
existing = {c["criteria"] for c in json_answer}
for criteria in all_criterias:
if criteria[0] not in existing:
json_answer.append({"criteria":criteria[0], "explanation":"Not infringed"})
enhanced_answer.append(overall_feedback)
for ans in json_answer:
ans = copy.deepcopy(ans)
ans["EVAL"] = EVAL_ANSWER_NOEVAL
enhanced_answer.append(ans)
return enhanced_answer
def eval_the_piece_of_c_code(openai_client, ccode):
"""
Main entrypoint to this module. Will be called from backend. Will block so multithreading pls.
Will return a proprer json and will have EVAL fields, too.
:param ccode:
:return:
"""
enhanced_answer = {"error":"Not processed"}
try:
chatgpt_ans = eval_code_by_chatgpt(openai_client, ccode)
except:
import traceback
traceback.print_exc()
enhanced_answer = {"error": "There was an error while calling chatGPT."}
return enhanced_answer
if "error" in chatgpt_ans:
# we forward it to caller
enhanced_answer = chatgpt_ans
pass
else:
try:
chatgpt_js = parse_chatgpt_answer(chatgpt_ans)
enhanced_answer = add_evaluation_fields_on_js_answer(chatgpt_js)
except:
import traceback
traceback.print_exc()
enhanced_answer = {"error": "There was an error while parsing the answer."}
return enhanced_answer
|