bug#45576: 28.0.50 [native-comp] some functions in eshell stop functioning after commit 0593f478762437e2a8618f3f874a26424e4590b4

[address@hidden]

Indeed, 
 (functionp (func1 x)) does not imply (functionp x)


Sent from ProtonMail mobile



-------- Original Message --------
On Jan 1, 2021, 15:12, Andrea Corallo < akrl@sdf.org> wrote:

"gliao.tw@pm.me" <gliao.tw@pm.me> writes:

> Hi,
>
> More precisely, in this case, the "eshell-find-alias-function" definded in $EMACS_DIR/share/emacs/28.0.50/lisp/eshell/esh-cmd.el.gz seems where the issue pops up.
>
> Before commit #0593f478762437e2a8618f3f874a26424e4590b4,
>
> (eshell-find-alias-function "ls") shows
>
> "eshell/ls"
>
> but on or after commit #0593f478762437e2a8618f3f874a26424e4590b4, the same command shows
>
> "nil"
>
> Both cases are based on functions loaded from eshell.el.gz and esh-cmd.el's corresponding eln files.
>
> It seems that some part of the type inference code added recently lead to this unexpected result.
>
> Thanks,
> Kiong-Ge.

I see what's going, this is very funny :)

I've used the alist producing `cl-deftype-satisfies' and reversed it to
get the relation type-predicate -> type.

But doing this way satisfying `functionp' implies the argument being a
function (while indeed it can be also a symbol).

Using this information the compiler cleverly proves the output of
`eshell-find-alias-function' being always nil.

So I guess the relation type <-> predicate in the alist used to build
`cl-deftype-satisfies' is correct but not bijective and cannot simply be
reversed to produce what I use.

I'll come up with a fix tomorrow.

Andrea