From: Tomi Ollila <tomi.ollila@iki.fi>
To: Dmitry Kurochkin <dmitry.kurochkin@gmail.com>, notmuch@notmuchmail.org
Subject: Re: [PATCH 1/3] test: add functions to count how much times notmuch
	was called
In-Reply-To: <87aa7evdy8.fsf@gmail.com>
References: <1322271878-32614-1-git-send-email-dmitry.kurochkin@gmail.com>
	<1322450895-32523-1-git-send-email-dmitry.kurochkin@gmail.com>
	<1322450895-32523-2-git-send-email-dmitry.kurochkin@gmail.com>
	<yf6aa7gj7w5.fsf@taco2.nixu.fi> <87hb1ovsz4.fsf@gmail.com>
	<yf6mxbfxezr.fsf@taco2.nixu.fi> <87aa7evdy8.fsf@gmail.com>
User-Agent: Notmuch/0.10+15~gb5803e9 (http://notmuchmail.org) Emacs/23.3.1
	(i686-pc-linux-gnu)
Date: Wed, 30 Nov 2011 13:53:24 +0200
Message-ID: <yf662i1n7wr.fsf@taco2.nixu.fi>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Precedence: list

On Wed, 30 Nov 2011 01:03:27 +0400, Dmitry Kurochkin <dmitry.kurochkin@gmail.com> wrote:
> Hi Tomi.
> 
> On Tue, 29 Nov 2011 14:58:00 +0200, Tomi Ollila <tomi.ollila@iki.fi> wrote:
> > Hi Dmitry.

[ ... ]

> > 
> > The (posix) shell command language defines 'Arithmetic Expansion' in
> > 
> > http://pubs.opengroup.org/onlinepubs/007908799/xcu/chap2.html#tag_001_006_004
> > 
> > I.e. using format $(( expression )) makes shell doing the arithetic itself
> > instead of forking a process (or two!) to do so.
> > 
> 
> I though expr was a builtin.  Now I agree that it is better to replace
> it with $(()), even though I still prefer the expr syntax.

Actually, I thought also that expr was a builtin. That makes the resolution
'forks subshell to execute builtin expr' below wrong. If it were a builtin
then bash would also fork only once (to get details right). I re-tested
with zsh using 'builtin pwd' and '/bin/pwd' instead of 'expr' -- only one fork 
in each case. So, those who examined my tests with deep interest also note
this correction.


> > Normally in this case it is not so big deal (and still it isn't, but...)
> > In this  particular case the shell wrapper counting notmuch launches and
> > exec'ing it the wrapper could do this without fork(2)ing a single time
> > (i.e. keep the process count unchanged compared to execing notmuch
> > directly)
> > 
> > Anyway, many opinions; as far as it works I'm fine with it :) 
> > 
> > Now that you feel relaxed, check the results of some further
> > experimentation ;) :
> > 
> > excerpt from man strace:
> > 
> >        -ff         If the -o filename option is in effect,  each  processes
> >                    trace  is  written  to  filename.pid  where  pid  is the
> >                    numeric process id of each process.
> > 
> > Executing  rm -f forked.*; strace -ff -o forked bash -c 'echo $(( 5 + 5 ))' 
> > 
> > will output '10' and create just one 'forked.<pid>' file
> > 
> > Executing  rm -f forked.*; strace -ff -o forked bash -c 'echo $(expr 5 + 5)' 
> > 
> > output 10 as expected, but there is now *3* forked.<pid> files !
> > 
> > bash does not optmize; it forks subshell to execute $(...) and then
> > there just works as usual (forks subshell to execute builtin expr))
> > 
> > Executing  rm -f forked.*; strace -ff -o forked bash -c 'echo $(exec expr 5 + 5)' 
> > 
> > (the added 'exec' takes off one fork -- just 2 forked.<pid> files appear).
> > 
> > I did the same tests using dash, ksh & zsh on linux system, and every one
> > of these managed to optimize one fork out in the above 3 fork case.
> > 
> 
> Thanks for details.
> 
> Regards,
>   Dmitry
> 
> > 
> > Tomi

Tomi