Re: RFC: String interpolation

[Joost Kremers]

On Thu, Dec 08 2016, Clément Pit--Claudel wrote:
> On 2016-12-08 09:43, Joost Kremers wrote:
> > > (let ((password "letmein")) (fmt "The password is 
> > > ${password}")) ⇒
> > > "The password is letmein"
> >
> > Why the {}? Is there any context in which $<expr> is ambiguous 
> > such
> > that ${<expr>} is needed to resolve it? Other than vectors, I 
> > mean
>
> Yes, unfortunately: you can include virtually anything in a Lisp 
> variable name.
>
> (let* ((a 1)
>        (b 2.0)
>        (a/b (/ a b)))
>   (fmt "$a/$b = ${a/b}"))

My initial thinking was that you could simply take everything up 
to the next whitespace character to be part of the symbol, but 
obviously I didn't realise that it should be possible *not* to 
have whitespace immediately after the value of a symbol. But I see 
your point now.

> > I honestly must say I find these last two very confusing. I 
> > would
> > simply write:
> >
> > (fmt "Today is $(format-time-string \"%Y-%m-%d\" 
> > current-time)).")
> >
> > The escaped double quotes are not perfect, but I find this much 
> > more
> > readable and more pleasant to look at.
>
> I really dislike the escaped quotes, but it might just be me :)

Oh, I'm sure you're not the only one. :-) I'm not a big fan, 
either, but I can live with them. It'd be nice to have two 
different string delimiters, as in Python (actually, Python has 
three...) but that's not something a package can fix.

Anyway, your current implementation handles 
"$[date|%Y-%m-%d](current-time)" as well as the format that I 
prefer, so no-one loses out.

> This was inspired from Python's per-type format string, but I'm 
> open to better syntax suggestions :)

I was hoping to provide some, but looking at it now, I doubt I 
was... :-)

> > Perhaps you should also look at abo-abo's hydra package, since 
> > he
> > does basically the same thing in his hydra doc strings. AFAICT, 
> > it
> > works this way: % starts a placeholder, ` marks the start of 
> > the
> > expression. If the expression is an s-expr, you can leave out 
> > the `.
>
> Thanks for the reference!  Do you know how it knows where to 
> stop after a backtick?

> From a quick look at the code, I believe the assumption is that 
symbols only contain lower and uppercase letters, digits and the 
characters - and / . So that won't work. The relevant code seems 
to be here:

https://github.com/abo-abo/hydra/blob/master/hydra.el#L636

> I'm a bit worried about the lack of delimiter after `, though; 
maybe it would help to see what the example above (with $a/$b and 
$a/b) would look like in that syntax.

AFAICT it won't work. "%`a/b" would refer to the symbol `a/b', 
which is fine, but "%`a/%`b" would refer to a symbol `a/' and a 
symbol `b', but the former doesn't exist. And `a/b' only happens 
to work because / is included in the regexp used to parse hydra 
doc strings. Other characters won't work at all.

> > > fmt expands the strings at compile time (so they must be 
> > > constant
> > > strings).
> >
> > I've occasionally appreciated the fact that I can pass a 
> > dynamically
> > created format string to `format'... ;-) Ok, I could always use
> > `format' in such cases.
>
> Yes; otherwise, fmt becomes a thinly veiled eval, and that isn't 
> great.

True.

--
Joost Kremers
Life has its moments