Re: Changes in GC and in pure space

[Paul Eggert]

Stefan Monnier wrote:
> > > As it happens, the issue with hash tables is bound up with the unexec thingy, in
> > > that portable dumping currently can screw up hash tables with user-defined
> > > tests.
>
> Note that AFAIK we currently don't dump any such hash-tables, so if
> there's a bug there it only affects the case of the user *re*dumping
> with a special config, which AFAIK we do not intend to officially
> support yet for 27.1.

Although I vaguely recall an email conversation to that effect, I don't see it 
in the Emacs documentation. And since unexec is now disabled by default, if 
people want to use any dumping method at all then by default they'll be 
redumping in the way that you suggest. If we really intend to not support that, 
we should be clearer about it in NEWS and documentation.

At any rate the simplest workaround is to avoid pdumping the problematic hash 
tables, and I installed the attached patch to do that. We can add support for 
pdumping those hash tables later, if anybody cares about it.

0001-Do-not-pdump-user-defined-hashtabs.patch
Description: Text Data