From: David Bremner <david@tethera.net>
To: Anton Khirnov <anton@khirnov.net>, notmuch@notmuchmail.org
Subject: Re: [PATCH v4 3/6] test: add tests for attachment filtering
In-Reply-To: <177649733250.3642.1582452438689018762@lain.khirnov.net>
References: <20260307160342.16482-1-anton@khirnov.net>
 <20260307160342.16482-4-anton@khirnov.net> <87341b9mza.fsf@tethera.net>
 <177649733250.3642.1582452438689018762@lain.khirnov.net>
Date: Fri, 19 Jun 2026 09:20:27 +0900
Message-ID: <87h5mzwfck.fsf@tethera.net>
MIME-Version: 1.0
Message-ID-Hash: AERCEEVTJ6JP6VPHBJ24FFKJRUNK327M
Precedence: list
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit

Anton Khirnov <anton@khirnov.net> writes:

> Quoting David Bremner (2026-04-04 09:57:29)
>> Anton Khirnov <anton@khirnov.net> writes:
>> 
>> > ---
>> >  test/T681-index-filter.sh | 84 +++++++++++++++++++++++++++++++++++++++
>> >  1 file changed, 84 insertions(+)
>> >  create mode 100755 test/T681-index-filter.sh
>> >
>> 
>> BTW, I think we talked about performance tests for the filtering at some
>> point.  I just checked, and the enron corpus has some
>> attachements. There are about 6600 word documents and about 3500 excell
>> documents. I guess we probably should not trust those attachments.
>
> The question is, are measuring performance of just the libnotmuch
> interface, or of the whole package including the sample filter.py,
> whatever programs it invokes, and the sandbox?

We chatted about this briefly on IRC. To summarize my comments from
then,

1) Existing performance tests are enough for making sure the
theoretically negligible performance impact for people not using the new
feature really is negligible.

2) It seems likely that users will have performance problems to debg
with this new feature, and it would be helpful to have some kind of
baseline test so that we can establish that some trivial filtering
is within expected an performance on their machine. This will also help
guard against performance regressions in attachment indexing introduced
by other changes to libnotmuch.

I don't think we really discussed how much effort was involved with
implementing (and I guess maintaining) such a test, since we didn't
really get past motivation(s) for doing so.

d

_______________________________________________
notmuch mailing list -- notmuch@notmuchmail.org
To unsubscribe send an email to notmuch-leave@notmuchmail.org