Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0 (Apple Message framework v1081)
References: <905C5ED0-6639-4DEB-95AC-A2FCB2C4491D@gmail.com>
            <AANLkTinSd4KS9gbk_cBPseFjbTFfSmSJUgYJUUP_vpBZ@mail.gmail.com>
            <E8A13BAD-6790-4966-9BAF-379E21DD249C@gmail.com>
            <AANLkTin1PWjtDuxug+ZjGKKg=VBbuz7Pf12rCkAo9+ex@mail.gmail.com>
            <AANLkTinh=zgNFCYWuGwO4mCsL6a7Yezxec264DmWwabk@mail.gmail.com>
Content-Transfer-Encoding: 8bit
Message-ID:  <935A1EA2-C971-4A3A-8DF4-6F890E0CFB32@gmail.com>
Date:         Mon, 14 Mar 2011 08:44:37 +1030
Reply-To: Mailing list for the LaTeX3 project
              <LATEX-L@listserv.uni-heidelberg.de>
Sender: Mailing list for the LaTeX3 project <LATEX-L@listserv.uni-heidelberg.de>
From: Will Robertson <wspr81@GMAIL.COM>
Subject: Re: expl3's seq_pop_right etc.
To: LATEX-L@listserv.uni-heidelberg.de
In-Reply-To:  <AANLkTinh=zgNFCYWuGwO4mCsL6a7Yezxec264DmWwabk@mail.gmail.com>
Precedence: list
Status: R

Hi Bruno,

Great work!

On 14/03/2011, at 6:44 AM, Bruno Le Floch wrote:

> The only reason to not brace items is to be able to find an item
> faster using delimited arguments, and that this will only work for
> "nice" items (i.e., no brace, no `#`).

I was aware that the \seq_if_in functions had this restriction, but I didn't realise the internal structure of the seq was chosen for this reason; IMO braces turn up far too easily in input text to restrict them from such a data structure. (Imagine pushing a list of names to a stack, say, and someone is called "Hef{}feron"!)

> So I have the following more
> radical proposal: brace the arguments, and propose two searches,

We like radical proposals :)

> - one true search, which goes through each element and compares with
> `\tl_if_eq:nn`,
> 
> - a `str`-like one, which detokenizes the seq, and searches for the
> detokenized version of `\seq_elt:n {<item>}` with a delimited argument
> (now ok, since everything is detokenized).
> 
> That proposal would make the code much, much clearer in my opinion.
> Also slightly faster, but not more than two-fold.

I did a quick comparison, and it seems like the delimited-argument search is around about 10-20 times faster than mapping, even for relatively large seqs (several thousand (small) elements, which takes a surprisingly long time to create using \seq_push:Nn).

So I'm generally in favour of this suggestion. But I wonder if there is some downside to using braces that I'm not experienced enough to see. (I can't think off the top of my head of another stack-like data structures in TeX that uses braces.) Can anyone comment further on this?

-- Will