Re: [Sbcl-devel] [Sbcl-commits] master: Fix DELETE-PACKAGE interaction with optimistic symbol loading.

[Christophe R. <cs...@ca...>]

Charles Zhang <cha...@ya...> writes:

> So it seems unlikely to run into a situation where you would've gotten
> no error in older sbcls but would with current HEAD. The cause of the
> error has changed slightly (from the dumper coalescing the symbols to
> the IR1 component making it the same before the dumper gets to it),
> but the fundamental issue remains with using RENAME-PACKAGE and making
> assumptions about how the literal symbols get dumped and reloaded. If
> a real regression with user code pops up with top-level-lambda-max,
> happy to set it to 0 and add a test case. But it seems like we won't
> gain much for this corner of the spec by turning it off. Compile time
> and load time package environments are hard. Thoughts?

Here's my test case, which does have a behaviour difference between
2.2.4 and HEAD; (load (compile-file "foo.lisp")) is enough to show the
difference in behaviour.