[latex3-commits] [git/LaTeX3-latex3-latex2e] apdf161: Simplify a bit \@@_make_name:n (e6a1a69e)

[Phelype Oleinik]

On Wed, Sep 9, 2020 at 11:48 AM Bruno Le Floch <blflatex at gmail.com> wrote:
>
> > -          { \tl_gset:co {@@~#1}
> > -                        {\cs:w g_@@_#1_code_next_tl \cs_end: } }
> > +          { \tl_gset:co { @@~#1 }
> > +                        { \cs:w g_@@_#1_code_next_tl \cs_end: } }
>
> I wonder if it would be nicer to call it \cs:w @@_next~#1 \cs_end: to match the
> @@~#1 (note I was careful to avoid name clashes, you cannot do @@~next_#1 or
> similar).

Hm...  It might appear in error messages, as the main token list, even
if we don't need to do fancy \csname stuff with it.  Might as well go
into the same package.

> > -\cs_new:Npn \@@_use_undefined:w #1 #2 \use:n #3 #4 \cs_end:
> > +\cs_new:Npn \@@_use_undefined:w #1 #2 @@~#3 \cs_end:
>
> Unsafe.  The \use:n was there so that a \cs_end: inside the hook name wouldn't
> mark the end of the argument of \@@_use_undefined:w.

Ouch, I completely overlooked that.  I thought that the \use:n thingy
was just so that the actual hook name could be easily extracted from
within the \cs:w...\cs_end:.

> One safe and fast option would be
>
> \cs_new_eq:NN \@@_use_end: \cs_end:
>
> and use \@@_use_end: as the end-marker.  It looks odd in the code but at least
> is safe and avoids the ugly \use:n.
>
> Alternatively, if we want maximum speed for the case where the hook exists, we
> could accept to always turn the cs to \relax, something like
>
> [...]
>
> Ok, it's really ugly, I don't like it.  But it should be a bit faster since it
> avoids making the cs from its csname twice, and manipulates fewer tokens as a
> result.

I vote for the \@@_use_end: version.  Hook names probably won't have
/that/ much complicated code, so building the csname twice won't have
that much of an impact on performance, and may save adding an useless
csname to the hash table.

Thanks!

Phelype