profiling.sampling is not working for `replay` mode

[seanleeee13]

Bug report

Bug description:

I ran next command with a.py code, and it made the error.

python3.15 -m profiling.sampling run -o output.prof a.py
python3.15 -m profiling.sampling replay output.prof

# a.py
for i in range(10000):
    a = [*p for p in [[1, 2], [2], [3, 4, 4]]]

This is the error message:

Traceback (most recent call last):
  File "<frozen runpy>", line 196, in _run_module_as_main
  File "<frozen runpy>", line 87, in _run_code
  File "C:\Users\XXX\AppData\Local\Python\pythoncore-3.15-64\Lib\profiling\sampling\__main__.py", line 65, in <module>
    main()
    ~~~~^^
  File "C:\Users\XXX\AppData\Local\Python\pythoncore-3.15-64\Lib\profiling\sampling\cli.py", line 788, in main
    _main()
    ~~~~~^^
  File "C:\Users\XXX\AppData\Local\Python\pythoncore-3.15-64\Lib\profiling\sampling\cli.py", line 922, in _main
    handler(args)
    ~~~~~~~^^^^^^
  File "C:\Users\XXX\AppData\Local\Python\pythoncore-3.15-64\Lib\profiling\sampling\cli.py", line 1175, in _handle_replay
    with BinaryReader(args.input_file) as reader:
         ~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
  File "C:\Users\XXX\AppData\Local\Python\pythoncore-3.15-64\Lib\profiling\sampling\binary_reader.py", line 32, in __enter__
    self._reader = _remote_debugging.BinaryReader(self.filename)
                   ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^
ValueError: Invalid magic number: 0xfa03a97b

The reported magic number 0xfa03a97b corresponds to Python 3.13/3.14. It seems that profiling.sampling might be inadvertently picking up components or generating data using the 3.14 environment available on the PATH, even when explicitly invoked with python3.15.
I used python 3.15.0a7, and I also downloaded python 3.14.3, and only 3.14.3 is on PATH.
Environments - Window 11 64bit, Using VSCode powershell terminal, pwsh 7.5.5
Also, I tried to do this in venv but python -m profiling.sampling run -o output.prof a.py made another error.
Failed to find the PyRuntime section in process 15236 on Windows platform

CPython versions tested on:

3.15

Operating systems tested on:

Windows

Linked PRs

• gh-148225: Validate profiling.sampling replay input #148243

[Ignyra]

I looked into this a bit, and it seems that the initial error comes from reading the data as the binary format while the original command produced a default pstats output. The replay command only works with binary format files that can be generated using --binary, and it doesn’t validate the input format, so passing a pstats file leads to the confusing error in which the BinaryReader expected a magic number in the header where it is only generated for binary files. I think this should work as expected:

python -m profiling.sampling run --binary -o output.prof test.py
python -m profiling.sampling replay output.prof

[seanleeee13]

I tried to do with --binary option, and it made Failed to find the PyRuntime section in process 26260 on Windows platform error, a simmilar error with when using venv.

[seanleeee13]

Tried with another a.py - and it worked.

import time

for i in range(1000000):
    a = [p for p in [[1, 2], [2], [3, 4, 4]]]
    if i % 100000 == 0:
        time.sleep(0.1)

I think it could be an error because a.py is to short for sampling.
Also, I updated to python 3.15.0a8, but it doesn't work on venv yet.

[Ignyra]

I tried to reproduce this on Windows with Python 3.15.0a8, but I wasn’t able to hit the same error. It might be related to something environment-specific.

[pablogsal]

Yeah I think the only bug here that we can validate is that the binary files are not validated before checking and the error is confusing. I will work on a fix