emacs floating point error during dump?

emacs floating point error during dump?

[Miles Bader]

Anyone else see this when compiling CVS emacs?

   LC_ALL=C ./temacs -batch -l loadup dump
   emacs: Invalid hash table rehash size: 5.30239915e-315
   make[1]: *** [emacs] Error 1

-Miles
-- 
I have seen the enemy, and he is us.  -- Pogo

Re: emacs floating point error during dump?

[Luc Teirlinck]

Miles Bader wrote:

   Anyone else see this when compiling CVS emacs?

      LC_ALL=C ./temacs -batch -l loadup dump
      emacs: Invalid hash table rehash size: 5.30239915e-315
      make[1]: *** [emacs] Error 1

   -Miles

No, I just bootstrapped without any problems.  I have LANG set to
en_US.ISO-8859-1, but that should not make any difference (I guess).
There was a problem with bootstrapping a little while ago, which I
fixed in the meantime, but that had nothing to do with hash tables.
(It was a typo in display.texi).

Sincerely,

Luc.

Re: emacs floating point error during dump?

[Simon Josefsson]

Miles Bader <miles@lsi.nec.co.jp> writes:

> Anyone else see this when compiling CVS emacs?
>
>    LC_ALL=C ./temacs -batch -l loadup dump
>    emacs: Invalid hash table rehash size: 5.30239915e-315
>    make[1]: *** [emacs] Error 1

FWIW; yes.  I get the same number too.  Trying "make bootstrap" now
(before it was just a 'cvs upd'ed old installation).

Re: emacs floating point error during dump?

[Stefan Monnier]

> Anyone else see this when compiling CVS emacs?
> 
>    LC_ALL=C ./temacs -batch -l loadup dump
>    emacs: Invalid hash table rehash size: 5.30239915e-315
>    make[1]: *** [emacs] Error 1

My GC changes mostly affect handling of floating point values, so
it sounds like a good candidate.  But the above is useless for any
serious debugging, of course.

I'm trying to install my GC patches in small increments so that binary
search in the version space can point to reasonably small changes.
Please try to find out which patch to alloc.c leads to the above problem
(if any).  Also we need backtrace and stuff, of course.

Also don't forget that the makefile dependency rules for lisp.h are
missing, so you need to `rm src/*.o' because I've changed the representation
of floating point numbers.


	Stefan

Re: emacs floating point error during dump?

[Miles Bader]

"Stefan Monnier" <monnier+gnu/emacs@cs.yale.edu> writes:
> Also don't forget that the makefile dependency rules for lisp.h are
> missing, so you need to `rm src/*.o' because I've changed the representation
> of floating point numbers.

Yes that did it; thanks!

[You really ought to warn people about this sort of thing ... or did I
just miss it?]

-Miles
-- 
Somebody has to do something, and it's just incredibly pathetic that it
has to be us.  -- Jerry Garcia

Re: emacs floating point error during dump?

[Stefan Monnier]

> [You really ought to warn people about this sort of thing ... or did I
> just miss it?]

You mean, we should fix the dependency rules ?  100% agreement,


	Stefan "who uses a `makedepend' thingy"

Re: emacs floating point error during dump?

[Miles Bader]

"Stefan Monnier" <monnier+gnu/emacs@cs.yale.edu> writes:
> > [You really ought to warn people about this sort of thing ... or did I
> > just miss it?]
> 
> You mean, we should fix the dependency rules ?  100% agreement,

It seems like it might be a good idea to split lisp.h into multiple
include files, so that the .c files could have a dependency on the
internal representations, but not on all the fairly uncritical stuff
like external function and variable declarations.

E.g., we could move the internal-rep stuff into `lisp-core.h' (and have
lisp.h #include it), and add a dependency on that for every lisp-using
.c file.

-Miles
-- 
Next to fried food, the South has suffered most from oratory.
  			-- Walter Hines Page

Re: emacs floating point error during dump?

[Andreas Schwab]

Miles Bader <miles@lsi.nec.co.jp> writes:

|> "Stefan Monnier" <monnier+gnu/emacs@cs.yale.edu> writes:
|> > > [You really ought to warn people about this sort of thing ... or did I
|> > > just miss it?]
|> > 
|> > You mean, we should fix the dependency rules ?  100% agreement,
|> 
|> It seems like it might be a good idea to split lisp.h into multiple
|> include files, so that the .c files could have a dependency on the
|> internal representations, but not on all the fairly uncritical stuff
|> like external function and variable declarations.

Or we could just add the dependencies on lisp.h to the makefile.  After
all, this is not much of a burden nowadays any more.

Andreas.

-- 
Andreas Schwab, SuSE Labs, schwab@suse.de
SuSE Linux AG, Deutschherrnstr. 15-19, D-90429 Nürnberg
Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: emacs floating point error during dump?

[Richard Stallman]

The reason I omitted dependencies on lisp.h was that that file changes
quite often, but most changes in it don't require recompiling all of
Emacs.

Perhaps it would make sense to put the prototypes and variable
declarations in a different file, and add dependencies on lisp.h
but not on that other file.

Re: emacs floating point error during dump?

[Miles Bader]

Richard Stallman <rms@gnu.org> writes:
> Perhaps it would make sense to put the prototypes and variable
> declarations in a different file, and add dependencies on lisp.h
> but not on that other file.

Right, that's what I was thinking of...

-Miles
-- 
I'd rather be consing.

Re: emacs floating point error during dump?

[Richard Stallman]

    Or we could just add the dependencies on lisp.h to the makefile.  After
    all, this is not much of a burden nowadays any more.

To have to recompile everything just because of a change in prototype
would still be an annoyance.  We may as well avoid it, since we can
easily do so.

Re: emacs floating point error during dump?

[Sam Steingold]

> * In message <E19aeWP-0008Hh-68@fencepost.gnu.org>
> * On the subject of "Re: emacs floating point error during dump?"
> * Sent on Thu, 10 Jul 2003 12:43:53 -0400
> * Honorable Richard Stallman <rms@gnu.org> writes:
>
> The reason I omitted dependencies on lisp.h was that that file changes
> quite often, but most changes in it don't require recompiling all of
> Emacs.

This is The Wrong Thing to do.
Correctness first, performance later.
The Emacs developers already spent _more_ time handling the issues
stemming from this makefile bug (answering questions, trying to debug
instead of a simple "make clean && make" &c) than would have been spent
rebuilding Emacs as per the "overkill" dependencies.

(Just MHO, as always - see `X-Disclaimer' header :-)

-- 
Sam Steingold (http://www.podval.org/~sds) running w2k
<http://www.camera.org> <http://www.iris.org.il> <http://www.memri.org/>
<http://www.mideasttruth.com/> <http://www.palestine-central.com/links.html>
Software is like sex: it's better when it's free.

Re: emacs floating point error during dump?

[Stefan Monnier]

>     Or we could just add the dependencies on lisp.h to the makefile.  After
>     all, this is not much of a burden nowadays any more.
> 
> To have to recompile everything just because of a change in prototype
> would still be an annoyance.  We may as well avoid it, since we can
> easily do so.

I expect I'm one of the few people who recompiles Emacs (very) regularly
and who uses a 266Mhz PII machine to do those recompiles, and yet I have
gone through the trouble of adding all those dependencies to lisp.h
(via `makedepend') and haven't actually been annoyed by them.

I suffer more from Emacs's own performance problems (that I still haven't
had enough time to really investigate) than from excessive recompilation.


	Stefan

Re: emacs floating point error during dump?

[Richard Stallman]

    > The reason I omitted dependencies on lisp.h was that that file changes
    > quite often, but most changes in it don't require recompiling all of
    > Emacs.

    This is The Wrong Thing to do.
    Correctness first, performance later.

It is silly to be rigid about it.

Re: emacs floating point error during dump?

[Sam Steingold]

> * In message <E19b3zV-0002xW-Vc@fencepost.gnu.org>
> * On the subject of "Re: emacs floating point error during dump?"
> * Sent on Fri, 11 Jul 2003 15:55:37 -0400
> * Honorable Richard Stallman <rms@gnu.org> writes:
>
>     > The reason I omitted dependencies on lisp.h was that that file
>     > changes quite often, but most changes in it don't require
>     > recompiling all of Emacs.
> 
>     This is The Wrong Thing to do.
>     Correctness first, performance later.
> 
> It is silly to be rigid about it.

it is unthinkable to be lax about it.

those who sacrifice freedom for security will lose both.

those who sacrifice correctness for performance, will lose both:

you already wasted more of your time on this discussion than you (or,
rather, _your computer_!) would have spent on un-necessary
recompilations; so your approach is a total loss.

-- 
Sam Steingold (http://www.podval.org/~sds) running w2k
<http://www.camera.org> <http://www.iris.org.il> <http://www.memri.org/>
<http://www.mideasttruth.com/> <http://www.palestine-central.com/links.html>
Of course, I haven't tried it. But it will work.    - Isaak Asimov

Re: emacs floating point error during dump?

[Kim F. Storm]

Sam Steingold <sds@gnu.org> writes:

> > * In message <E19b3zV-0002xW-Vc@fencepost.gnu.org>
> > * On the subject of "Re: emacs floating point error during dump?"
> > * Sent on Fri, 11 Jul 2003 15:55:37 -0400
> > * Honorable Richard Stallman <rms@gnu.org> writes:
> >
> >     > The reason I omitted dependencies on lisp.h was that that file
> >     > changes quite often, but most changes in it don't require
> >     > recompiling all of Emacs.
> > 
> >     This is The Wrong Thing to do.
> >     Correctness first, performance later.
> > 
> > It is silly to be rigid about it.
> 
> it is unthinkable to be lax about it.

I agree with Sam -- breaking dependencies on purpose is beyond me.

I really don't understand why changing a prototype should not result
in recompiling things depending on that header file and I really fail
to see why you deliberately want compilation to fail from time to
time.  After all, CVS emacs evolves daily with many people involved --
and having broken dependencies causing strange behaviour is simply a
waste of everybody's time.

And I don't see why you think lisp.h changes very often.  If we ignore
the recent changes by Stefan, it only changes every 14 days, which IMO
doesn't warrent for breaking dependencies.

-- 
Kim F. Storm <storm@cua.dk> http://www.cua.dk

Re: emacs floating point error during dump?

[Richard Stallman]

    it is unthinkable to be lax about it.

I have no difficulty thinking it.  You need to stretch your mind.  My
decision is to do what Miles suggested.

    you already wasted more of your time on this discussion than you (or,
    rather, _your computer_!) would have spent on un-necessary
    recompilations; so your approach is a total loss.

Recompilation of all of Emacs seems to take around ten minutes on my
computer.  I have not spent anywhere near ten minutes on this
discussion.  But I agree, let's not spend more time on it.

Re: emacs floating point error during dump?

[Sam Steingold]

> * In message <E19bUSe-0003wB-S2@fencepost.gnu.org>
> * On the subject of "Re: emacs floating point error during dump?"
> * Sent on Sat, 12 Jul 2003 20:11:28 -0400
> * Honorable Richard Stallman <rms@gnu.org> writes:
>
>     it is unthinkable to be lax about it.
> 
> I have no difficulty thinking it.  You need to stretch your mind.

You need to learn either English or manners.

If your problem is indeed English, let me rephrase:

When talking about code, you rightfully insist on having the correct
solution, not just "half correct" or "mostly correct".

Why do you settle for half-solutions with the build infrastructure?

>     you already wasted more of your time on this discussion than you
>     (or, rather, _your computer_!) would have spent on un-necessary
>     recompilations; so your approach is a total loss.
> 
> Recompilation of all of Emacs seems to take around ten minutes on my
> computer.

For a $1,000 you can get a machine which will do that in 1 min.
I am sure FSF can afford that.
Note that the recompilation is done by the machine, so the time is
cheap, as opposed to the time spent by humans debugging and reporting
these errors that stem from your intentional sloppiness.

> I have not spent anywhere near ten minutes on this discussion.

... yet.  This issue will come back to haunt you until it is fixed.

Actually, this is at least the third time in the last year when this is
being debated.  Did you factor in those instance too?

-- 
Sam Steingold (http://www.podval.org/~sds) running w2k
<http://www.camera.org> <http://www.iris.org.il> <http://www.memri.org/>
<http://www.mideasttruth.com/> <http://www.palestine-central.com/links.html>
People hear what they want to hear and discard the rest.

Re: emacs floating point error during dump?

[Miles Bader]

On Sat, Jul 12, 2003 at 08:30:36PM -0400, Sam Steingold wrote:
> You need to learn either English or manners.

You do realize that being a twit isn't going to help you make your case,
don't you?

-Miles
-- 
Fast, small, soon; pick any 2.

Re: emacs floating point error during dump?

[Sam Steingold]

> * In message <20030713013256.GA26136@gnu.org>
> * On the subject of "Re: emacs floating point error during dump?"
> * Sent on Sat, 12 Jul 2003 21:32:56 -0400
> * Honorable Miles Bader <miles@gnu.org> writes:
>
> On Sat, Jul 12, 2003 at 08:30:36PM -0400, Sam Steingold wrote:
> > You need to learn either English or manners.
> 
> You do realize that being a twit isn't going to help you make your
> case, don't you?

I won't swallow insults to "help my case".  I am not a politician.

I don't think that being "the big cheese" entitles people to being rude.

-- 
Sam Steingold (http://www.podval.org/~sds) running w2k
<http://www.camera.org> <http://www.iris.org.il> <http://www.memri.org/>
<http://www.mideasttruth.com/> <http://www.palestine-central.com/links.html>
If you're passed on the right, you're in the wrong lane.

Re: emacs floating point error during dump?

[David Kastrup]

Sam Steingold <sds@gnu.org> writes:

> > * In message <20030713013256.GA26136@gnu.org>
> > * On the subject of "Re: emacs floating point error during dump?"
> > * Sent on Sat, 12 Jul 2003 21:32:56 -0400
> > * Honorable Miles Bader <miles@gnu.org> writes:
> >
> > On Sat, Jul 12, 2003 at 08:30:36PM -0400, Sam Steingold wrote:
> > > You need to learn either English or manners.
> > 
> > You do realize that being a twit isn't going to help you make your
> > case, don't you?
> 
> I won't swallow insults to "help my case".  I am not a politician.

Dishing out insults is not going to "help your case" whether or not
your are a politician.

> I don't think that being "the big cheese" entitles people to being
> rude.

So what was it instead that entitled you to being rude?

-- 
David Kastrup, Kriemhildstr. 15, 44793 Bochum

Re: emacs floating point error during dump?

[Miles Bader]

On Sat, Jul 12, 2003 at 10:40:40PM -0400, Sam Steingold wrote:
> I won't swallow insults to "help my case".  I am not a politician.

Clearly not.

> I don't think that being "the big cheese" entitles people to being rude.

No, but though rms's comment was rather borderline (I suspect he was trying
to be funny), it actually said something relevant to the dicussion -- you
need to consider other people's environments as well as your own when
deciding whether compiling everything is `fast enough.'

What you said, on the other hand, had no relevance at all, it was simply a
flat-out insult.

See the difference?

-Miles
-- 
The secret to creativity is knowing how to hide your sources.
  --Albert Einstein

Re: emacs floating point error during dump?

[David Kastrup]

Miles Bader <miles@gnu.org> writes:

> On Sat, Jul 12, 2003 at 10:40:40PM -0400, Sam Steingold wrote:
> > I won't swallow insults to "help my case".  I am not a politician.
> 
> Clearly not.

Please don't confuse "politician" with "diplomat", even though it
would be nice to have a higher correlation between the two sometimes.

-- 
David Kastrup, Kriemhildstr. 15, 44793 Bochum

Re: emacs floating point error during dump?

[Thien-Thi Nguyen]

Sam Steingold <sds@gnu.org> writes:

   Actually, this is at least the third time in the last year when this
   is being debated.  Did you factor in those instance too?

perhaps whoever takes on emacs maintainership after rms will address
this.  if rms wants to save that person some work he will think ahead.
if he wants to save everyone else some confusion in the meantime he will
think abroad.  if he wants to show his thoughts as they change he will
think aloud.  time waits for no one, however.

thi

Re: emacs floating point error during dump?

[Richard Stallman]

Sam, I am sorry if "you need to stretch your mind" seemed rude.  I
meant it as a humorous comeback to "...unthinkable".  I do appreciate
your contributions to GNU.

Re: emacs floating point error during dump?

[David Kastrup]

Richard Stallman <rms@gnu.org> writes:

> Sam, I am sorry if "you need to stretch your mind" seemed rude.  I
> meant it as a humorous comeback to "...unthinkable".  I do
> appreciate your contributions to GNU.

For what it is worth, I think that dependencies should not cut short
for convenience.  People that really want their convenience, can
manage it by touching the various object files.

Of course, splitting lisp.h into more parts might also make sense.

-- 
David Kastrup, Kriemhildstr. 15, 44793 Bochum

Re: emacs floating point error during dump?

[Sam Steingold]

> * In message <E19bkjx-0005gO-Vv@fencepost.gnu.org>
> * On the subject of "Re: emacs floating point error during dump?"
> * Sent on Sun, 13 Jul 2003 13:34:25 -0400
> * Honorable Richard Stallman <rms@gnu.org> writes:
>
> Sam, I am sorry if "you need to stretch your mind" seemed rude.  I
> meant it as a humorous comeback to "...unthinkable".  I do appreciate
> your contributions to GNU.

Thank you.
I am sorry I took this so far.


-- 
Sam Steingold (http://www.podval.org/~sds) running w2k
<http://www.camera.org> <http://www.iris.org.il> <http://www.memri.org/>
<http://www.mideasttruth.com/> <http://www.palestine-central.com/links.html>
I want Tamagochi! -- What for?  Your pet hamster is still alive!

Re: emacs floating point error during dump?

[Stefan Monnier]

> When talking about code, you rightfully insist on having the correct
> solution, not just "half correct" or "mostly correct".
> 
> Why do you settle for half-solutions with the build infrastructure?

While I completely agree that I'd prefer to have full (and auto-generated)
dependencies between .c and .h files in the makefile, the build
infrastructure has many more "incorrect" parts than that and
I doubt we will ever have a "correct" solution.  There is always
a tradeoff between convenience and excessive recompilation.


	Stefan

PS: examples that I have bumped into over the last few months:
- modifying an elisp macro will not automatically force a rebuild of all
  the .el files that make use of it.
- changing from `make' to `make CFLAGS=-DENABLE_CHECKING' does not
  force recompilation of the .o files.
- modifying `Makefile.in' will correctly cause rebuilding of `Makefile'
  but will fail to reconstruct the targets that the Makefile describes.
- upgrading libc will not automatically cause `configure' to be rerun.
- deinstalling libpng does not automatically cause `configure' to be rerun.

Re: emacs floating point error during dump?

[Sam Steingold]

> * In message <200307132306.h6DN6Wpq004158@rum.cs.yale.edu>
> * On the subject of "Re: emacs floating point error during dump? "
> * Sent on Sun, 13 Jul 2003 19:06:31 -0400
> * Honorable "Stefan Monnier" <monnier+gnu/emacs@cs.yale.edu> writes:
>
> - modifying an elisp macro will not automatically force a rebuild of
>   all the .el files that make use of it.

1. all *.elc should depend on the dumped few files like subr.el.

2. if a foo.el uses a macro defined in bar.el, then it should (require
   'bar) and a simple grep+sed cen generate the dependency.  I did
   something similar for a Commol Lisp package 5 years ago (before
   switching to defsystem), it worked just fine.

> - changing from `make' to `make CFLAGS=-DENABLE_CHECKING' does not
>   force recompilation of the .o files.

This is not a dependency issue.
If you type a make argument, you can as well type "make clean", right?

> - modifying `Makefile.in' will correctly cause rebuilding of
>   `Makefile' but will fail to reconstruct the targets that the
>   Makefile describes.

I am afraid you lost me here.
Do you mean that every target should depend on the Makefile?
If yes, I am sure you can easily add Makefile to each target line.

> - upgrading libc will not automatically cause `configure' to be rerun.

this should not be too hard to fix, I guess:

config.status: ${srcdir}/configure /lib/libc.a
	./config.status --recheck

[of course the location and the name of libc will have to be somehow
determined...]

but how often do you upgrade libc?

> - deinstalling libpng does not automatically cause `configure' to be
> rerun.

again, this can probably be fixed...

The bottom line on these last two items is that host system
dependencies are probably less important than the Emacs
interdependencies because the host system can be presumed to be more
stable than the Emacs development tree.

-- 
Sam Steingold (http://www.podval.org/~sds) running w2k
<http://www.camera.org> <http://www.iris.org.il> <http://www.memri.org/>
<http://www.mideasttruth.com/> <http://www.palestine-central.com/links.html>
If you think big enough, you'll never have to do it.

Re: emacs floating point error during dump?

[Eli Zaretskii]

> From: Sam Steingold <sds@gnu.org>
> Date: 13 Jul 2003 19:58:15 -0400
> 
> > - modifying `Makefile.in' will correctly cause rebuilding of
> >   `Makefile' but will fail to reconstruct the targets that the
> >   Makefile describes.
> 
> I am afraid you lost me here.
> Do you mean that every target should depend on the Makefile?
> If yes, I am sure you can easily add Makefile to each target line.

This problem exists in every GNU project (and elsewhere).  AFAIU, the
solution is that when Makefile.in changes, one does a "make clean" and
then "configure" and "make".

Re: emacs floating point error during dump?

[Kim F. Storm]

"Stefan Monnier" <monnier+gnu/emacs@cs.yale.edu> writes:

> > When talking about code, you rightfully insist on having the correct
> > solution, not just "half correct" or "mostly correct".
> > 
> > Why do you settle for half-solutions with the build infrastructure?
> 
> While I completely agree that I'd prefer to have full (and auto-generated)
> dependencies between .c and .h files in the makefile, the build
> infrastructure has many more "incorrect" parts than that and
> I doubt we will ever have a "correct" solution.  There is always
> a tradeoff between convenience and excessive recompilation.

What about ./configure --with-lax-dependencies for those who _knows_
what they are doing and thus can be lax about dependencies...  and
let the rest of us have full dependencies by default.


-- 
Kim F. Storm <storm@cua.dk> http://www.cua.dk

Re: emacs floating point error during dump?

[Richard Stallman]

    What about ./configure --with-lax-dependencies for those who _knows_
    what they are doing and thus can be lax about dependencies...  and
    let the rest of us have full dependencies by default.

It is ok with me.