I noticed a very big difference in execution time between using chardet.detect with a complete bytes string and using UniversalDetector.feed with chunks of 1 MB.

• With a 100 MB file, composed only of "tests tests tests tests [....]":

  • chardet.detect takes ~64 seconds.
  • UniversalDetector.feed takes ~3 seconds.

• With the previous file on which I appended a file in MacRoman of ~10 KB (containing the character ’ in MacRoman):

  • chardet.detect: I interrupted the execution after 20 minutes...
  • UniversalDetector.feed takes ~3 seconds.

In case you wonder what code I used, I compared the execution time of the following:

• chardet.detect:

print(detect(original_txt))

• UniversalDetector.feed:

num_chunks_processed = 0
for start, end in _get_chunk_slice_intervals(len(original_txt), CHUNK_SIZE):
    chunk = original_txt[start:end]
    detector.feed(chunk)
    num_chunks_processed += 1
detector.close()