From: Mark Walters <markwalters1009@gmail.com>
To: notmuch@notmuchmail.org
Subject: Re: [PATCH 0/3] Control notmuch show output
In-Reply-To: <1341673978-6094-1-git-send-email-markwalters1009@gmail.com>
References: <1341673978-6094-1-git-send-email-markwalters1009@gmail.com>
User-Agent: Notmuch/0.13.2+90~g84fa1ef (http://notmuchmail.org) Emacs/23.4.1
	(x86_64-pc-linux-gnu)
Date: Sat, 14 Jul 2012 17:49:29 +0100
Message-ID: <87r4sei8yu.fsf@qmul.ac.uk>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Precedence: list


I wonder if we need as many options as this: I implemented a simpler set
of possibilities: 
no-body:          just from, to, cc, subject and date headers and no bodies
brief:            from, to, cc, subject and date header with bodies
most:             all headers except "received" with bodies
all:              all headers and bodies

I then benchmarked this by timing the command line run. I saved the
output and ran it in a simple emacs lisp script which just ran (and
timed) json-read on the output. I also noted the size of the data.

All were done on the query 

notmuch show --format=json <relevant output> tag:notmuch 

(which contains 11 775 messages in 1 672 threads)

             Cli time      JSON time      Data MB
no-body:      10.0            7               7    
brief:        11.9           55              70
most:         12.6           73              91          
all:          13.0           85             107

(note the cli was actually run on a different (slower) machine than the
json)

So do we want all sorts of header options, or would the above be
sufficient? Note that for notmuch show itself the difference is probably
negligible in most cases as the data size is *much* smaller. The
datasize may be significant for some uses though.

Best wishes

Mark