Trigger GitHub CI with a comment on PR (#35211)

* fix

* fix

* comment

* final

* final

* final

---------

Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>

utils/pr_slow_ci_models.py

@@ -15,19 +15,20 @@
 """
 This script is used to get the models for which to run slow CI.
 
-A new model added in a pull request will be included, as well as models specified in a commit message with a prefix
-`[run-slow]`, `[run_slow]` or `[run slow]`. For example, the commit message `[run_slow]bert, gpt2` will give `bert` and
-`gpt2`.
+A new model added in a pull request will be included, as well as models specified in a GitHub pull request's comment
+with a prefix `run-slow`, `run_slow` or `run slow`. For example, the commit message `run_slow: bert, gpt2` will give
+`bert` and `gpt2`.
 
 Usage:
 
 ```bash
-python utils/pr_slow_ci_models.py.py
+python utils/pr_slow_ci_models.py
 ```
 """
 
 import argparse
 import re
+import string
 from pathlib import Path
 from typing import List
 
@@ -89,7 +90,7 @@ def get_new_python_files() -> List[str]:
 
 def get_new_model():
     new_files = get_new_python_files()
-    reg = re.compile(r"src/transformers/(models/.*)/modeling_.*\.py")
+    reg = re.compile(r"src/transformers/models/(.*)/modeling_.*\.py")
 
     new_model = ""
     for x in new_files:
@@ -101,45 +102,53 @@ def get_new_model():
     return new_model
 
 
-def parse_commit_message(commit_message: str) -> str:
+def parse_message(message: str) -> str:
     """
-    Parses the commit message to find the models specified in it to run slow CI.
+    Parses a GitHub pull request's comment to find the models specified in it to run slow CI.
 
     Args:
-        commit_message (`str`): The commit message of the current commit.
+        message (`str`): The body of a GitHub pull request's comment.
 
     Returns:
-        `str`: The substring in `commit_message` after `[run-slow]`, [run_slow]` or [run slow]`. If no such prefix is
-         found, the empty string is returned.
+        `str`: The substring in `message` after `run-slow`, run_slow` or run slow`. If no such prefix is found, the
+        empty string is returned.
     """
-    if commit_message is None:
+    if message is None:
         return ""
 
-    command_search = re.search(r"\[([^\]]*)\](.*)", commit_message)
-    if command_search is None:
+    message = message.strip().lower()
+
+    # run-slow: model_1, model_2
+    if not message.startswith(("run-slow", "run_slow", "run slow")):
         return ""
+    message = message[len("run slow") :]
+    # remove leading `:`
+    while message.strip().startswith(":"):
+        message = message.strip()[1:]
 
-    command = command_search.groups()[0]
-    command = command.lower().replace("-", " ").replace("_", " ")
-    run_slow = command == "run slow"
-    if run_slow:
-        models = command_search.groups()[1].strip()
-        return models
-    else:
-        return ""
+    return message
 
 
-def get_models(commit_message: str):
-    models = parse_commit_message(commit_message)
-    return [f"models/{x}" for x in models.replace(",", " ").split()]
+def get_models(message: str):
+    models = parse_message(message)
+    return models.replace(",", " ").split()
+
+
+def check_model_names(model_name: str):
+    allowed = string.ascii_letters + string.digits + "_"
+    return not (model_name.startswith("_") or model_name.endswith("_")) and all(c in allowed for c in model_name)
 
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()
-    parser.add_argument("--commit_message", type=str, default="", help="The commit message.")
+    parser.add_argument("--message", type=str, default="", help="The content of a comment.")
     args = parser.parse_args()
 
     new_model = get_new_model()
-    specified_models = get_models(args.commit_message)
+    specified_models = get_models(args.message)
     models = ([] if new_model == "" else [new_model]) + specified_models
+    # a guard for strange model names
+    models = [model for model in models if check_model_names(model)]
+    # Add "models/"
+    models = [f"models/{model}" for model in models]
     print(sorted(set(models)))