Performance issue: mailbox.mbox slow because it does a tell/lseek for every line

Bug report

Bug description:

Parsing a large mbox file with mailbox.mbox is surprisingly slow and I did some profiling and debugging. For a large file (my testcase, 3,000 mails, 26,000,000 lines), somewhere around half the time is spent calling tell(), once per line in PartialFile. As documented in issue #107158, tell() here uses lseek which is a system call rather than just returning a constant.

# Slow code, assuming a large my_mail.mbox
import mailbox
box = mailbox.mbox("my_mail.mbox", create=False)
for message in box:
    pass

Replacing mailbox.mbox with a simple mmapped file that scans for b"\nFrom " and calls BytesParser(policy=policy.default).parsebytes(...) directly gives the same results in my testing (I'm sure it's not as powerful). In my tests that cuts the parsing overhead from 11 seconds to 1, and total parse time from >20 to 9.

CPython versions tested on:

3.12

Operating systems tested on:

Windows

Linked PRs

• gh-157543

Start with the mailbox.mbox iteration path and its PartialFile implementation, focusing on the per-line tell() and lseek() behavior described here. Review linked PR gh-157543 first, then benchmark parsing a large mbox file and confirm the performance issue is addressed without changing parsing results.