Date: Wed, 9 Oct 2013 12:19:08 -0400
From: Austin Clements <amdragon@MIT.EDU>
To: Mark Walters <markwalters1009@gmail.com>
Subject: Re: [PATCH 00/11] Fix search tagging races
Message-ID: <20131009161907.GS21611@mit.edu>
References: <1381185201-25197-1-git-send-email-amdragon@mit.edu>
	<87ob70kxg1.fsf@qmul.ac.uk>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <87ob70kxg1.fsf@qmul.ac.uk>
User-Agent: Mutt/1.5.21 (2010-09-15)
Cc: notmuch@notmuchmail.org
Precedence: list

Quoth Mark Walters on Oct 08 at  8:56 am:
> 
> Hello
> 
> It's great that this might finally get done. But there is one problem
> currently.
> 
> If you open a large search buffer and then do *-<tab> it will die as the
> tagging routine runs notmuch search to find a completion-list for the
> tag. (it runs notmuch search --output=tags <query>)

Hmm.  If we implement docid queries (described in the TODO added by
this series), we should be able to get away with what we're doing now
without any serious performance problems...

> We could just return all tags in this case. Or we could do something
> like the series
> id:1354263691-19715-1-git-send-email-markwalters1009@gmail.com
> which makes completion happen based on the tags visible to the user, not
> the tags actually in the database.

OTOH, I think what you were going for in this series is the right
thing to do from a UI perspective anyway.  I'll try implementing
something along these lines, though I've got an idea that I think will
by more Elispy.  Currently `notmuch-tag' has this strange interface
where it can interactively prompt in some cases.  This isn't the right
way to do this.  `notmuch-tag' should be non-interactive and the
interactive tagging commands should have an interactive specification
that prompts for tags to change, right at the interactive entry point.
This should give us a clean place to provide a list of existing tags,
and would also let us do other nice things like specify a different
tag prompt for * commands, and maybe add a y/n prompt to confirm a *
tagging command.  It would also provide a convenient place to wait for
search results to finish coming in after prompting in the * command,
which would be awkward to do right now.

> There is also a little discussion of this in my earlier attempt at
> fixing this: eg id:87mwy4smad.fsf@qmul.ac.uk
> 
> Best wishes
> 
> Mark
> 
> On Mon, 07 Oct 2013, Austin Clements <amdragon@MIT.EDU> wrote:
> > I was hacking on undo support for notmuch-emacs and sort of
> > accidentally wrote this instead.  This series fixes a set of
> > well-known races where tagging from search-mode unexpectedly affects
> > messages that arrived after the search was performed (and hence the
> > user doesn't know they're tagging them).  We've attacked this a few
> > times before, but have always run up against something that was
> > missing.  It turns out the pieces are finally all in place.
> >
> > The first five patches just clean various things up in preparation.
> > Patches 6 and 7 add support for tagging large queries, which would
> > otherwise become a problem when later patches start using explicit
> > message ID-based queries for tagging.  The remaining four patches
> > actually fix the search tagging races using explicit message ID-based
> > queries.
> >
> > It's a fairly long series, but none of the patches are very big.
> >
> > _______________________________________________
> > notmuch mailing list
> > notmuch@notmuchmail.org
> > http://notmuchmail.org/mailman/listinfo/notmuch