Optimization for notmuch tag by implicit filters

[Pieter Praet]

On Tue, 26 Apr 2011 23:31:09 +0200, Florian Friesdorf <flo at chaoflow.net> wrote:
> On Sat, 16 Apr 2011 14:59:34 +0200, Pieter Praet <pieter at praet.org> wrote:
> > On Thu, 14 Apr 2011 10:23:46 +0200, Florian Friesdorf <flo at chaoflow.net> wrote:
> > > 
> > > With 60k messages and 12k tagged as sent:
> > > 
> > > $ time notmuch tag +sent -- from:flo at chaoflow.net
> > > 
> > > real	0m8.561s
> > > user	0m8.069s
> > > sys	0m0.212s
> > > 
> > > $ time notmuch tag +sent -- from:flo at chaoflow.net and not tag:sent
> > > 
> > > real	0m0.043s
> > > user	0m0.036s
> > > sys	0m0.006s
> > > 
> > > 
> > > This could be made implicit:
> > > 
> > > notmuch tag +A +B -- <filter>
> > > -->
> > > notmuch tag +A +B -- <filter> and not \(tag:A and tag:B\)
> > > 
> > > Apply command, if one of the tags is not set.
> > > 
> > > 
> > > notmuch tag -C -D -- <filter>
> > > -->
> > > notmuch tag -C -D -- <filter> and \(tag:C or tag:D\)
> > > 
> > > Apply command, if one of the tags is set.
> > > 
> > > 
> > > notmuch tag +A +B -C -D -- <filter>
> > > -->
> > > notmuch tag +A +B -C -D -- <filter> and \(not tag:A or not tag:B\ or tag:C or tag:D\)
> 
> The second '\' after 'B' is not supposed to be there.
> 
> > > In order to enforce tagging and disable the filter there could be a
> > > flag.
> > > 
> > > I lack the knowledge/time to implement it, but I think it's at least
> > > worth documenting it.
> > 
> > Most of us already do this explicitly in our tagging scripts, so no harm
> > in making it standard behaviour, I guess.
> > 
> > Though to keep the implementation nice & clean, I'd advise against the
> > use of parens: no need for escape chars, no messing with De Morgan's
> > law, simply map the tag operations to their inverse in conjunctively
> > joined filters:
> > 
> > notmuch tag +A +B -- <filter> and not tag:A or not tag:B
> > notmuch tag -C -D -- <filter> and tag:C or tag:D
> > notmuch tag +A +B -C -D -- <filter> and not tag:A or not tag:B or tag:C or tag:D
> 
> I am not sure whether I understand what you mean.
> 
> The parens are already supported by notmuch and also needed. Your second
> line for example would remove C and D, if D is set, independently of
> <filter>. Without parens, based on `and` taking precedence over `or`:
> 
> F and (C or D) = F and C or F and D.
> 
> -- 
> Florian Friesdorf <flo at chaoflow.net>
>   GPG FPR: 7A13 5EEE 1421 9FC2 108D  BAAF 38F8 99A3 0C45 F083
> Jabber/XMPP: flo at chaoflow.net
> IRC: chaoflow on freenode,ircnet,blafasel,OFTC
Non-text part: application/pgp-signature


Correct.

I wasn't taking the user-defined <filter> into consideration, which
should indeed be conjunctive with the *entire* following expression.

I was commenting on the use of parens *inside* our implicit filter,
which was absolutely superfluous since you did no such thing. :D

Also, I said "simply map the tag operations to their inverse in
conjunctively joined filters", but what I meant was "DISjunctively
joined filters".

Perhaps I should go over my textbook on propositional logic again one
of these days :)


Peace

-Pieter