poll_for_input_1 undefined

poll_for_input_1 undefined

[Richard Stallman]

[[[ To any NSA and FBI agents reading my email: please consider    ]]]
[[[ whether defending the US Constitution against all enemies,     ]]]
[[[ foreign or domestic, requires you to follow Snowden's example. ]]]

In the trunk, updated an hour ago:

xterm.c at line 9012 calls poll_for_input_1,
but this is not defined anywhere.

keyboard.c has the only definition for poll_for_input_1
but it is inside #ifdef POLL_FOR_INPUT.
POLL_FOR_INPUT is not defined.

In fact, config.h does not mention POLL_FOR_INPUT, neither to define
it nor not to define it.  That is because config.in does not mention
POLL_FOR_INPUT either.

It appears to me that the trunk is broken on X on all platforms,
and has been since the June 5 change in keyboard.c.

Can that really be true, with nobody noticing?

I'm compiling on the Yeeloong, which is mipsel,
but that should not affect much.

-- 
Dr Richard Stallman
President, Free Software Foundation
51 Franklin St
Boston MA 02110
USA
www.fsf.org  www.gnu.org
Skype: No way! That's nonfree (freedom-denying) software.
  Use Ekiga or an ordinary phone call.

Re: poll_for_input_1 undefined

[Dmitry Antipov]

On 07/11/2014 08:57 PM, Richard Stallman wrote:

> In fact, config.h does not mention POLL_FOR_INPUT, neither to define
> it nor not to define it.  That is because config.in does not mention
> POLL_FOR_INPUT either.
>
> It appears to me that the trunk is broken on X on all platforms,
> and has been since the June 5 change in keyboard.c.

This is pretty strange if you look at this configure.in fragment:

if test "$window_system" != "none"; then
   AC_DEFINE(HAVE_WINDOW_SYSTEM, 1, [Define if you have a window system.])
   AC_DEFINE(POLL_FOR_INPUT, 1, [Define if you poll periodically to detect C-g.])
   WINDOW_SYSTEM_OBJ="fontset.o fringe.o image.o"
fi

Can you regenerate build stuff with autogen.sh and check whether it helps?

Dmitry

Re: poll_for_input_1 undefined

[Glenn Morris]

Richard Stallman wrote:

> It appears to me that the trunk is broken on X on all platforms,
> and has been since the June 5 change in keyboard.c.
>
> Can that really be true, with nobody noticing?

Since (IIUC) you build infrequently and do not bootstrap, it is very
unlikely you will be the first person to notice such an error. It is
much more likely that you encounter some random glitch because you did
not do a clean rebuild. I assume that's the case here.

Note that there are automated builds on Emacs that run on several
platforms, in theory after every commit, though in practice they often
don't run quite that often. But normally several times a day.

Re: poll_for_input_1 undefined

[Andreas Schwab]

Richard Stallman <rms@gnu.org> writes:

> In fact, config.h does not mention POLL_FOR_INPUT, neither to define
> it nor not to define it.  That is because config.in does not mention
> POLL_FOR_INPUT either.

config.in is a generated file.  Unless your host has an instable clock
this will be regenerated automatically.

Andreas.

-- 
Andreas Schwab, schwab@linux-m68k.org
GPG Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: poll_for_input_1 undefined

[Richard Stallman]

[[[ To any NSA and FBI agents reading my email: please consider    ]]]
[[[ whether defending the US Constitution against all enemies,     ]]]
[[[ foreign or domestic, requires you to follow Snowden's example. ]]]

    config.in is a generated file.  Unless your host has an instable clock
    this will be regenerated automatically.

It wasn't regenerated in my case.

(config.in used to be edited manually, and I did not realize that
was no longer the practice.)

I went to src and ran "make -k emacs".
First it recompiled all the .o files in src.
Then it did a recursive make for a library
which reran configure, but without updating config.in.
Then it recompiled the .o files again.
This does not seem right.

-- 
Dr Richard Stallman
President, Free Software Foundation
51 Franklin St
Boston MA 02110
USA
www.fsf.org  www.gnu.org
Skype: No way! That's nonfree (freedom-denying) software.
  Use Ekiga or an ordinary phone call.

Re: poll_for_input_1 undefined

[Andreas Schwab]

Richard Stallman <rms@gnu.org> writes:

> Then it did a recursive make for a library
> which reran configure, but without updating config.in.

config.in isn't generated by configure.  It is generated by autoheader.

> Then it recompiled the .o files again.
> This does not seem right.

Most likely you have inconsistent timestamps due to instable clock.

Andreas.

-- 
Andreas Schwab, schwab@linux-m68k.org
GPG Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: poll_for_input_1 undefined

[Richard Stallman]

[[[ To any NSA and FBI agents reading my email: please consider    ]]]
[[[ whether defending the US Constitution against all enemies,     ]]]
[[[ foreign or domestic, requires you to follow Snowden's example. ]]]

    > Then it recompiled the .o files again.
    > This does not seem right.

    Most likely you have inconsistent timestamps due to instable clock.

Could you explain what you mean by "instable clock"?
My machine's clock seems to advance steadily and properly.

All the files are local on this machine.

-- 
Dr Richard Stallman
President, Free Software Foundation
51 Franklin St
Boston MA 02110
USA
www.fsf.org  www.gnu.org
Skype: No way! That's nonfree (freedom-denying) software.
  Use Ekiga or an ordinary phone call.

Re: poll_for_input_1 undefined

[Richard Stallman]

[[[ To any NSA and FBI agents reading my email: please consider    ]]]
[[[ whether defending the US Constitution against all enemies,     ]]]
[[[ foreign or domestic, requires you to follow Snowden's example. ]]]

    > Then it did a recursive make for a library
    > which reran configure, but without updating config.in.

    config.in isn't generated by configure.  It is generated by autoheader.

The makefile should update config.in before using it.
And it should run configure before compiling the C sources,
if it is going to run configure at all.

-- 
Dr Richard Stallman
President, Free Software Foundation
51 Franklin St
Boston MA 02110
USA
www.fsf.org  www.gnu.org
Skype: No way! That's nonfree (freedom-denying) software.
  Use Ekiga or an ordinary phone call.

Re: poll_for_input_1 undefined

[Glenn Morris]

Richard Stallman wrote:

> The makefile should update config.in before using it.

It does.

> And it should run configure before compiling the C sources,
> if it is going to run configure at all.

It does.

You can test this eg by: touch configure.ac; make.

If you can give a reproducible recipe showing how to get into the state
that you did, things could be improved. But I imagine that even with
a slow machine, it would be quicker and more efficient to just do a
clean rebuild, rather than us all trying to figure it exactly what
happened.

Re: poll_for_input_1 undefined

[Andreas Schwab]

Richard Stallman <rms@gnu.org> writes:

> Could you explain what you mean by "instable clock"?

http://lists.gnu.org/archive/html/emacs-devel/2013-11/msg00421.html

Andreas.

-- 
Andreas Schwab, schwab@linux-m68k.org
GPG Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: poll_for_input_1 undefined

[Paul Eggert]

I think Richard's problem this time was caused not by clock skew, but 
because 'bzr up' significantly changed the build procedure, and then 'cd 
src; make -k' did not propagate those changes properly.

I took a shot at fixing this and installed the fix as trunk bzr 117526. 
  This fixes only the src subdirectory; I expect lib-src, lisp, and nt 
need similar fixes, but one step at a time.

It is a pain that this build-procedure stuff needs to be semi-copied 
around from one makefile to another.  One possibility to avoid this 
pain, if someone has the time, would be to switch Emacs to a 
non-recursive 'make' style [1], which we've already done in other GNU 
projects (e.g., coreutils) to good effect.  Another possibility would be 
to use Automake.

[1] Miller P. Recursive make considered harmful. AUUGN 1998 
Feb;19(1):14-25. http://aegis.sourceforge.net/auug97.pdf

Re: poll_for_input_1 undefined

[Richard Stallman]

[[[ To any NSA and FBI agents reading my email: please consider    ]]]
[[[ whether defending the US Constitution against all enemies,     ]]]
[[[ foreign or domestic, requires you to follow Snowden's example. ]]]

    > The makefile should update config.in before using it.

    It does.

    > And it should run configure before compiling the C sources,
    > if it is going to run configure at all.

    It does.

It didn't for me yesterday.

I don't have the output saved.  Next time I hope I will remember
to save it.

    If you can give a reproducible recipe showing how to get into the state
    that you did, things could be improved. But I imagine that even with
    a slow machine, it would be quicker and more efficient to just do a
    clean rebuild, rather than us all trying to figure it exactly what
    happened.

This is a bug and it should be fixed.
(I am not sure exactly what "a clean rebuild" means.
'make bootstrap'?  That takes ages.)

-- 
Dr Richard Stallman
President, Free Software Foundation
51 Franklin St
Boston MA 02110
USA
www.fsf.org  www.gnu.org
Skype: No way! That's nonfree (freedom-denying) software.
  Use Ekiga or an ordinary phone call.

Re: poll_for_input_1 undefined

[Paul Eggert]

Richard Stallman wrote:
> Next time I hope I will remember to save it.

My guess is that you could reproduce the bug as follows:

cd emacs
touch configure.ac
cd src
make -k

However, please don't bother.  Instead, please do this:

cd emacs
bzr up
make

This may take a while, but after that the bug should be fixed, because 
of the patch that I contributed in trunk bzr 117526.