`evil-bounds-of-not-thing-at-point`/`evil-forward-end` assume `forward-thing` returns a number, which breaks with Emacs 30.1's `forward-thing-provider-alist`

[dschrempf]

Emacs 30.1 added forward-thing-provider-alist, so a mode can hook in its own forward-thing behavior for a given thing (including Evil's own things, like evil-paragraph).

Problem: on that provider branch, forward-thing returns nil/t (moved / stopped early), not the integer count you get from the old forward-op branch. Evil has a couple of spots that assume it's always a number:

• evil-bounds-of-not-thing-at-point
• evil-forward-end

Both do (zerop (forward-thing thing ...)), so once a provider is registered you get:

(zerop nil) => (wrong-type-argument number-or-marker-p nil)

which crashes plain paragraph objects/motions (a p, i p, }, {).

Minimal reproducer: (Emacs 30.2 + Evil 1.15.0, still broken on both master branches as of 2026-07-16):

(setq-local forward-thing-provider-alist
            '((evil-paragraph . (lambda (bw) (forward-paragraph (if bw -1 1))))))
(evil-select-an-object 'evil-paragraph nil nil 'inclusive 1 t)
;; => (wrong-type-argument number-or-marker-p nil)

I ran into this trying to swap some fragile advice-based paragraph-object narrowing (in a Tree-sitter Haskell mode) for a clean buffer-local provider, the one hook Emacs added for exactly this use case immediately crashes Evil.

For context, I also raised the return-value shape upstream with Emacs: https://lists.gnu.org/archive/html/bug-gnu-emacs/2026-07/msg00767.html

The fix is probably just: don't assume forward-thing returns a number, treat nil as "moved" and t as "no progress".

[dschrempf]

FYI, I have submitted a patch to the Emacs bug tracker, resolving the inconsistent return value of forward-thing. If you like, review the patch, and see if it suits the need of evil!