Re: [PATCH v2] kunit: tool: add ability to parse multiple files

[Daniel Latypov]

On Wed, Mar 6, 2024 at 2:30 PM Rae Moar <rmoar@xxxxxxxxxx> wrote:
>

Note: looks like there's two small bugs w/ the stdin codepath.
If both are addressed, it looks like stdin works again for me.

<snip>

> Changes since v1:
> - Annotate type of parsed_files
> - Add ability to input file name from stdin again
> - Make for loops a bit terser
> - Add no output warning
> - Change feature to take in multiple fields rather than a directory.
>   Currently nonrecursive. Let me know if people would prefer this as
>   recursive.

Just noting that I'd like to hear other's opinions on this.

I personally prefer the current approach.
I don't imagine there are going to be many nested directories of just
KTAP output files.

I.e. I'm assuming users would either be fine with
# just one dir w/ all KTAP outputs
$ kunit.py parse some_dir/*
# KTAP mixed in w/ other files, like we see in debugfs
$ find some_dir/ -name 'ktap_output' | xargs kunit.py parse

>
>  tools/testing/kunit/kunit.py | 45 +++++++++++++++++++++++++-----------
>  1 file changed, 32 insertions(+), 13 deletions(-)
>
> diff --git a/tools/testing/kunit/kunit.py b/tools/testing/kunit/kunit.py
> index bc74088c458a..df804a118aa5 100755
> --- a/tools/testing/kunit/kunit.py
> +++ b/tools/testing/kunit/kunit.py
> @@ -511,19 +511,37 @@ def exec_handler(cli_args: argparse.Namespace) -> None:
>
>
>  def parse_handler(cli_args: argparse.Namespace) -> None:
> -       if cli_args.file is None:
> +       parsed_files = [] # type: List[str]
> +       total_test = kunit_parser.Test()
> +       total_test.status = kunit_parser.TestStatus.SUCCESS
> +       if cli_args.files is None:
>                 sys.stdin.reconfigure(errors='backslashreplace')  # type: ignore
> -               kunit_output = sys.stdin  # type: Iterable[str]
> +               parsed_files.append(sys.stdin)
> +       elif cli_args.files[0] == "debugfs" and len(cli_args.files) == 1:

For me, the stdin branch doesn't get taken, i.e.

$ ./tools/testing/kunit/kunit.py parse
..
  File "./tools/testing/kunit/kunit.py", line 520, in parse_handler
    elif cli_args.files[0] == "debugfs" and len(cli_args.files) == 1:
         ~~~~~~~~~~~~~~^^^
IndexError: list index out of range

If unspecified, apparently `cli_args.files == []`, so we'd want to change it to
  if not cli_args.files:
     # stdin codepath

> +               for (root, _, files) in os.walk("/sys/kernel/debug/kunit"):
> +                       parsed_files.extend(os.path.join(root, f) for f in files if f == "results")
>         else:
> -               with open(cli_args.file, 'r', errors='backslashreplace') as f:
> +               parsed_files.extend(f for f in cli_args.files if os.path.isfile(f))
> +
> +       if len(parsed_files) == 0:
> +               print("No output found.")

This is what a user sees if they pass a dir in now
$ ./tools/testing/kunit/kunit.py parse tools/testing/kunit/test_data/
No output found.

I'm wondering if we should try to make the user's error more obvious.
E.g. we could add a list where `not os.path.isfile(f)` and print it like:

$ ./tools/testing/kunit/kunit.py parse tools/testing/kunit/test_data/
Ignoring 1 non-regular files: tools/testing/kunit/test_data/
No output found.


> +
> +       for file in parsed_files:
> +               print(file)
> +               with open(file, 'r', errors='backslashreplace') as f:

In the stdin case, `file` here is already a File object and not a filename.

Note: mypy/pytype will complain since the type annotation says List[str]
 kunit.py:520: error: Argument 1 to "append" of "list" has
incompatible type "TextIO"; expected "str"

Could do something like
       parsed_files = [] # type: List[Union[str, TextIO]]
..
               if isinstance(file, str):
                       print(file)
                       with open(file, 'r', errors='backslashreplace') as f:
                               kunit_output = f.read().splitlines()
               else:  # file is sys.stdin
                       kunit_output = file.read().splitlines()

With ^ and the change above to the `if`, seems like stdin works for me

$ echo "invalid" | ./tools/testing/kunit/kunit.py parse
$ ./tools/testing/kunit/kunit.py parse <
tools/testing/kunit/test_data/test_skip_tests.log

Thanks,
Daniel