[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Preview: portable dumper

From: Eli Zaretskii
Subject: Re: Preview: portable dumper
Date: Mon, 28 Nov 2016 23:14:32 +0200

> From: John Wiegley <address@hidden>
> Cc: Paul Eggert <address@hidden>,  address@hidden,  address@hidden
> Date: Mon, 28 Nov 2016 12:47:10 -0800
> >>>>> "EZ" == Eli Zaretskii <address@hidden> writes:
> EZ> I'm not going to argue. If enough people think I'm mistaken and want to go
> EZ> the portable dumper way, I will resign right here and now. It is very easy
> EZ> to convince me to step down, because I hesitated to take this job to begin
> EZ> with.
> EZ> Your call, crowd.
> OK, please let's take a few steps back and not make any rash decisions
> today. :)

It's no longer my decision.

> If Daniel's proposing an internal optimization and is willing to maintain it
> -- and if it doesn't make work for other contributors -- I suggest we accept
> it as long as he, or someone else, is willing to support it.

Emacs future cannot depend on a single person.  We had more than once
a situation where key developer(s) left leaving a feature or a whole
area unmaintained.  If some development isn't a one-time job -- and
this one surely isn't -- then experience shows that any arcane code in
C that loses its experts will become unmaintained and unmaintainable.
The only hope for us to escape this fate is to do it in a way that
most contemporary contributors understand and can easily learn to
hack, which means do it mostly in Lisp, reusing the existing C
machinery that is proven by time and doesn't need maintenance --
'load' and its cousins.  That was what I proposed.  The advantage of
that is that Emacs then becomes a normal program, one that doesn't
dump itself in any way, shape, or form, but instead reads in a
byte-compiled file, using the normal memory allocation routines.  No
more need to save any internal state, worry about relocations, ASLR,
etc.  It amazes me that the advantages of that are not immediately
clear to everyone, but I guess I'm biased.

> At the very least, it frees us from dependence on a glibc feature
> that is doomed to disappear.

No one is suggesting to stay with unexec.  The disagreement is about
where to go to get rid of it.

> If later the work becomes difficult to support, or if Daniel disappears, won't
> we just be returning to the same position as now? If that happeans, we can
> take what we've learned and try another approach, such as speeding up .elc
> loading. Or, if someone comes up with a better alternative in the meantime, we
> can just switch to it.

I don't think this is practical.  We cannot go back, certainly not
when unexec requires the expertise it does.  No one will agree to go
back.  Like with unexec, the only way the portable dumper will be
thrown away is if/when it turns out it is an obstacle to Emacs
development, by which time it will be too late.

> Accepting this patch doesn't mean we don't try the .elc speedup idea. The only
> thing we have to make sure is that it doesn't unduly increase the difficulty
> of adding new Lisp objects.

Accepting this patch means we believe in this direction, consider it a
perspective one, enough so that we are willing to invest a non-trivial
amount of energy into making it happen, fix the fallout, debug the
results, etc.  We are talking about at least one major release, maybe
more, and definitely several minor releases.  IOW, we are talking
several years.  I don't think we should gamble with such high stakes,
not unless we sure we want to go there, because this is the best
alternative of ditching unexec.

Anyway, I already wrote all this; there was a long discussion about
these issues, it's in the archives.  I have nothing else to add to
what I wrote back then.  This project needs to decide where it wants
to go, and the rest will be history.

reply via email to

[Prev in Thread] Current Thread [Next in Thread]