Performance and Recursion Limit Issues with Large Filter Chains

[lucemia]

Hi @jidasheng,

Thanks for the detailed report!

The “Argument list too long” issue has been addressed and released in v3.5.2. You can now avoid this by enabling use_filter_complex_script=True when calling run():

import ffmpeg
import sys

sys.setrecursionlimit(1000)

f = ffmpeg.input("in.mp4")
for _ in range(100):
    f = f.drawtext(
        text="123",
        x="100",
        y="100",
        fontsize=100,
        fontcolor="red",
        fontfile="/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf"
    )
f = f.output("out.mp4")
print(f.run(use_filter_complex_script=True))

This will switch to using -filter_complex_script instead of passing everything via the command line, which avoids hitting system argument limits.

As for the other two issues:
• RecursionError: maximum recursion depth exceeded
• Slow execution of compile_line() on very large filter chains

I don’t have a clean fix for those yet. Manually increasing the recursion limit with sys.setrecursionlimit() can help as a temporary workaround. I’ll go ahead and open a separate issue to track these performance limitations.

Thanks again for flagging this!

Originally posted by @lucemia in #706

[jidasheng]

Very slow execution of compile_line() on very large filter chains

Any progress on this problem?

[lucemia]

Progress:

1. Removed the is_dag check — with frozen dataclasses, cyclic references are impossible, making the check redundant.
2. Removed the hex method — now relying on Python’s built-in hash() for frozen dataclasses.

These changes may improve performance, especially with large filter chains.

Issues:

1. Recursion is pervasive across the codebase. Some instances stem from Python’s default behavior and are difficult to eliminate.