Re: mark_object crash in 22.1 and latest CVS (as of tonight)

[Stefan Monnier]

>> The `evaporate' option is inherently dangerous since it reclaims the
>> marker object forcefully without checking that nobody else is holding on
>> to it, even tho it comes from a plain normal argument and the caller may
>> very well have kept another ref to it somewhere.

> Indeed.  I faintly recall a discussion about restricting the access to
> match data.

Using an uninterned Qevaporate would kind of do that, but as we're now
seeing it's not sufficient since those markers may end up in the
undo-log anyway.

> Maybe that option was an ill-fated attempt based on the assumption
> that no one was allowed to copy a reference to match data or any of
> its components.

Sounds like it.

>> I suggest we kill it,

> Don't forget to kill it in emacs_base too.

Indeed,


        Stefan