I found that multiple calls to get_lexer_for_filename were somewhat slow in my environment, due to find_plugin_lexers discovering entry points every time it ran.

Its interesting that pkg_resources (used in the past) seems to cache entry points: https://github.com/pypa/pkg_resources/blob/main/__init__.py#L3212 But importlib.metadata does not.

I did some simple testing like this:

start = time.perf_counter_ns()
for i in range(100):
    get_lexer_for_filename("some.py", encoding="chardet")
    get_lexer_for_filename("some.c", encoding="chardet")
    get_lexer_for_filename("some.cpp", encoding="chardet")
    get_lexer_for_filename("some.java", encoding="chardet")
    try:
        get_lexer_for_filename("some.fasdf", encoding="chardet")
    except pygments.util.ClassNotFound:
        pass
end = time.perf_counter_ns()
print(f"Time: {(end - start) / 1_000_000} ms")

On my system I got these timings:

• Pygments 2.12 (using pkg_resources) took about 1,400ms
• Pygments 2.13-2.19.2 (importlib) took about 14,000ms
• 2.19.2 with a cache added took about 800-1100ms

It seems to me that adding this caching would not be a problem if pkg_resources cached entry points in the past. I'm not sure how someone would modify entry points at runtime anyway.