bug#61880: Native compilation fails to generate trampolines on certain scenarios

[Eli Zaretskii]

> From: Al Haji-Ali <abdo.haji.ali@gmail.com>
> Cc: 61880@debbugs.gnu.org, akrl@sdf.org
> Date: Tue, 14 Mar 2023 13:22:13 +0000
> 
> 
> On 13/03/2023, Richard Stallman wrote:
> > I'm very glad that we now warn users about this pitfall.
> > We have had bug reports over the years from users who fell
> > into it.  To "fix" it would be a mistake, but a warning
> > is fine.
> 
> Connecting this to bug#61917, is the compilation error there perhaps caused 
> by the advice to the primitive function `delete-region`?
> 
> Is it always wrong to advice primitives?

Not wrong, but risky.  Especially if the advice basically subverts the
primitive, like makes it always fail or something -- this is
frequently done in various mocking frameworks for testing purposes.

> Or only those that are "involved in producing primitives" as Eli put it?

Only those, yes.  And only in a way that significantly changes what
the original primitives do.

> If so, how can one tell which is which?

The warning knows.  It only warns against those we know could be
harmful.