bug#62009: 29.0.60; Emacs crashes on setf symbol-name

[Daniel Mendler <mail@daniel-mendler.de>]

On 3/10/23 12:50, Eli Zaretskii wrote:
> Why bother?  Emacs is not in the business of preventing Lisp
> programmers from shooting themselves in the foot, certainly not when
> that incurs runtime overhead, even a small one.

Of course Elisp is in the business of preventing programmers from
shooting themselves in the foot, otherwise we would extend Emacs in C.
It is not only that Elisp is easier to write thanks to macros and other
conveniences, but also because it is safer!

I fully agree with you that we should not introduce a performance
regression, in particular not one which increases GC pressure badly.
Furthermore I agree that this is a minor bug which only occurs as an
edge case when some specific strings are mutated.

However the cost of fixing this bug is minor, since Elisp already
supports read-only data stuctures as pointed out by Robert. We may only
need an additional check in aset which won't introduce significant costs
thanks to the branch predictor of CPUs. As Augusto pointed out, aset on
strings can already be very costly because the string must be
reallocated. In comparison, the additional check will be free.

Daniel