bug#25088: 25.1; feature-unload and reload of cl-defstruct fails

[Stefan Monnier]

> I think it's safe, but I've been wrong before.

I also think it's safe, and I've also been wrong before.

Another thing: the patch makes sense regardless of the change, since
it's good to define the things before we use them.

> Not sure how much safer this is, I think we would have to record the
> which symbol plists are being modified during `load' so that
> `unload-feature' could find them in `load-history' and reverse them
> along with functions definitions.  This would get rid of the
> compiler-macro entries that were added to cl-defstruct accessor function
> symbols, and so they would load successfully the second time round just
> like the first (presumably).

Yes, that would be good.  Currently unload-feature automatically un-does
defalias and defvar, more or less, but it would be good to extend this
to other top-level operations like `put`.


        Stefan