Re: RFC/WIP: Delete SET_TOP_LEVEL ....

[Andrew Cagney <ac131313 at cygnus dot com>]

Todd Whitesel wrote:
> 
> >       o       have catch_errors() always
> >               call do_cleanups()
> 
> I think this is the simplest valid migration plan.

What I'll do is make it ``bug compatible'' and have the main.c specific
cases do cleanups.

> As for functions using cleanups but not calling do_cleanup before returning,
> there is a least one place where this is relied upon for correct operation.
> 
> The code in coffread.c:coff_locate_sections() which reads in .stab[0-9]*
> sections does something evil. It xmalloc's buffers for the section contents
> and reads them in, then uses a cleanup to free them (here, one buffer, "n"):
> 
>           /* This will be run after coffstab_build_psymtabs is called
>              in coff_symfile_read, at which point we no longer need
>              the information.  */
>           make_cleanup (free, n);
> 
> The actual consumer of these buffers is dbxread.c:fill_symbuf().
> 
> This code is of course scary because if anyone adds a make_cleanup/do_cleanups
> pair between this code and last use of the memory, it will cause this fragile
> algorithm to break.

Thanks for pointing this out.  I was ``rationally paranoid''.

	Andrew