Re: Policy on pure space-related fixes

[Lynn Winebarger]

On Sat, Jul 30, 2022 at 9:27 AM Eli Zaretskii <eliz@gnu.org> wrote:
>
> > From: Lynn Winebarger <owinebar@gmail.com>
> > Date: Sat, 30 Jul 2022 07:45:21 -0400
> >
> > Generally, are fixes to pure space worth submitting, given the plan to 
> > decommission it, or will they just be
> > ignored?  I'm still waiting on clearance from my employer to submit 
> > patches, unfortunately.
>
> It depends on the purpose of the change.  If the unexec build is
> broken due to that, we will increase the pure space.  But I don't see
> a reason to do that to cater to a build that we don't yet support, so
> I suggest you do that locally (since you are making a lot of local
> changes anyway).

Now that I have a basic grasp of the bootstrap and dumping process,
I've written a script invoked from my site-load (only during a
non-bootstrap dump) that builds a normal emacs dump in situ, moves
that to emacs-1 (and emacs-1.pdmp), then byte-compiles all the
required files in the site-load.el file using emacs-1 and extracts the
docstrings, before proceeding with the usual build.
It's basically the process Stefan M described for a three stage
bootstrap, but without modifying the configure/Makefiles to make it
happen.
I'm able to run the dump without modifying any lisp files, though I've
made a few modifications to alloc.c to fix the segfaults/runaway
allocation issues, and to lread.c to add a "docstring-hack" variable
so I can explicitly turn off the docstring hack in site-load.el.  I
also have to flip the purify flag off on some of the files that modify
variables declared with defconst.
The build makes it up to the dumping step, at least, but then is
either stuck in a loop of some kind or processing an incredibly large
mark-compact GC algorithm.  I stopped it after a couple of hours.  I'm
going to try adding files in smaller chunks (than the 511 or so I
added) to see where the dumping procedure goes off the rails.
I'm going to try an accurate solution to dealing with cyclic
structures by purecopy as well, to see if I can get rid of the other
segfaults as well.

Lynn