Re: Warn about comparing quoted lists (etc) using `eq`

[Mattias Engdegård]

15 dec. 2022 kl. 05.17 skrev Juanma Barranquero <lekktu@gmail.com>:

> Note that your patch has a false positive with the empty string, which is 
> optimized to be always the same object

That's only an implementation aspect that we give no guarantees about on the 
Lisp language level, and in fact as you correctly observe:

> there are two, a unibyte one and a multibyte one

Precisely, and often there's no telling which one we compare against in (eq x 
"") -- that expression may return t or nil when x is an empty string.
Thus it's definitely a well-motivated warning, toy examples like

> (let ((x "")) (eq x ""))

notwithstanding. We simply do not guarantee the identity of literal strings and 
the empty string is no exception.

15 dec. 2022 kl. 07.34 skrev Dr. Arne Babenhauserheide <arne_bab@web.de>:

> I miss a suggestion, though: the warning as it is is not actionable. It
> would be great if it could say something along the lines of
> 
> "... (arg 2). Consider using `equal'."

While this sounds like a useful suggestion on the surface, it's actually 
slightly dangerous: blindly replacing `eq` with `equal` (and `memq` with 
`member`, and so on) may very well break working code (that silently relied on 
the condition not being true). It's also a poor remedy against pratfalls like

  (memq x '(one 'two 'three))

which do occur. What we really want is to make a human look at the code, and 
think.