bug#30626: 26.0.91; Crash when traversing a `stream-of-directory-files'

[Eli Zaretskii]

> Date: Sat, 17 Mar 2018 18:10:24 +0200
> From: Eli Zaretskii <eliz@gnu.org>
> Cc: michael_heerdegen@web.de, john.b.mastro@gmail.com, nicolas@petton.fr,
>       30626@debbugs.gnu.org
> 
> > (--stream-- #[256 "\211\203\303\242\207\303\242\204 
> > \304\300\242\305\300\242\302\242\\\301\302\242#B\240\210\303\306\240\210\304\242\207"
> >  
> >               [(1) 1000000 (1) (nil) (nil) stream-range t] 7 "
> >               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> > 
> > (fn &optional CHECK)"])
> > 
> > Not sure where to go next with this.
> 
> The goal was to find out which variable holds a reference to the
> entire long stream, right?  So it sounds like a pointer to it is kept
> in an automatic variable on the stack of exec_byte_code, right?  Which
> kinda makes sense, since the stream is still being processed, I think.
> 
> Or am I confused?

Actually, there's still some mystery: if this object is a 7-element
vector, where do all the other GC frame come from?  Hmm... how
long/deep is each of the cons cells in elements 1 through 4 of the
vector?  If they are deeply nested, then that's the answer we've been
looking for, I think.