Debian Bug report logs - #431464
autoconf: files are rebuilt several times on vfat (because "cp -p" can't preserve time?)

Toggle useless messages

Message #5 received at submit@bugs.debian.org (full text, mbox, reply):

From: Vincent Lefevre <vincent@vinc17.org>

To: Debian Bug Tracking System <submit@bugs.debian.org>

Subject: autoconf: files are rebuilt several times on vfat (because "cp -p" can't preserve time?)

Date: Mon, 2 Jul 2007 21:37:42 +0200

Package: autoconf
Version: 2.61-4
Severity: normal

I've built a MPFR tarball on this Debian machine and transfered it
to my Zaurus (a PDA under Linux), which has a SD card with a vfat
file system. Doing "make check" on the Zaurus rebuilds files several
times: "make check" builds the test files twice, run the tests, then
rebuilds MPFR entirely (so that I can't see the test results, in
particular).

I think I've found the problem. The config.log says:

configure:8965: checking whether ln -s works
configure:8972: result: no, using cp -p

I wonder why "-p" is used, as on vfat, the mtime cannot be preserved.
If the Makefile assumes that "cp -p" preserves mtime, then it is no
longer surprising that I get many file rebuilds.

-- System Information:
Debian Release: lenny/sid
  APT prefers unstable
  APT policy: (500, 'unstable'), (500, 'stable')
Architecture: i386 (i686)

Kernel: Linux 2.6.18-4-686-bigmem (SMP w/2 CPU cores)
Locale: LANG=POSIX, LC_CTYPE=en_US.ISO8859-1 (charmap=ISO-8859-1)
Shell: /bin/sh linked to /bin/bash

Versions of packages autoconf depends on:
ii  debianutils                   2.22       Miscellaneous utilities specific t
ii  m4                            1.4.9-2    a macro processing language
ii  perl                          5.8.8-7    Larry Wall's Practical Extraction 

Versions of packages autoconf recommends:
ii  automake1.6 [automaken]   1.6.3-12       A tool for generating GNU Standard
ii  automake1.7 [automaken]   1.7.9-9        A tool for generating GNU Standard
ii  automake1.8 [automaken]   1.8.5+nogfdl-2 A tool for generating GNU Standard
ii  automake1.9 [automaken]   1.9.6+nogfdl-3 A tool for generating GNU Standard

-- no debconf information

Message #10 received at 431464@bugs.debian.org (full text, mbox, reply):

From: Ben Pfaff <blp@cs.stanford.edu>

To: Vincent Lefevre <vincent@vinc17.org>

Cc: 431464@bugs.debian.org

Subject: Re: Bug#431464: autoconf: files are rebuilt several times on vfat (because "cp -p" can't preserve time?)

Date: Thu, 05 Jul 2007 10:01:27 -0700

Thanks for the bug report.  I haven't had a chance to investigate
it yet, but I will do so as soon as I can.  If this is something
that is blocking your progress on something important, please let
me know so that I can try to give it a higher priority.
-- 
Ben Pfaff 
http://benpfaff.org

Message #15 received at 431464@bugs.debian.org (full text, mbox, reply):

From: Vincent Lefevre <vincent@vinc17.org>

To: Ben Pfaff <blp@cs.stanford.edu>

Cc: 431464@bugs.debian.org

Subject: Re: Bug#431464: autoconf: files are rebuilt several times on vfat (because "cp -p" can't preserve time?)

Date: Fri, 6 Jul 2007 00:41:28 +0200

On 2007-07-05 10:01:27 -0700, Ben Pfaff wrote:
> Thanks for the bug report.  I haven't had a chance to investigate
> it yet, but I will do so as soon as I can.  If this is something
> that is blocking your progress on something important, please let
> me know so that I can try to give it a higher priority.

A second (and last) release candidate of MPFR 2.3.0 is planned for the
end of the month. If we can have a fix before this release candidate,
this would really be useful.

-- 
Vincent Lefèvre <vincent@vinc17.org> - Web: <http://www.vinc17.org/>
100% accessible validated (X)HTML - Blog: <http://www.vinc17.org/blog/>
Work: CR INRIA - computer arithmetic / Arenaire project (LIP, ENS-Lyon)

Message #20 received at 431464@bugs.debian.org (full text, mbox, reply):

From: Ben Pfaff <blp@cs.stanford.edu>

To: Vincent Lefevre <vincent@vinc17.org>

Cc: 431464@bugs.debian.org

Subject: Re: Bug#431464: autoconf: files are rebuilt several times on vfat (because "cp -p" can't preserve time?)

Date: Sat, 07 Jul 2007 11:12:59 -0700

Vincent Lefevre <vincent@vinc17.org> writes:

> I've built a MPFR tarball on this Debian machine and transfered it
> to my Zaurus (a PDA under Linux), which has a SD card with a vfat
> file system. Doing "make check" on the Zaurus rebuilds files several
> times: "make check" builds the test files twice, run the tests, then
> rebuilds MPFR entirely (so that I can't see the test results, in
> particular).

I've now attempted to reproduce this problem.  I created a 128 MB
file, used mkdosfs to create a DOS file system on it, mounted it
as vfat, untarred the mpfr candidate from
http://www.mpfr.org/mpfr-2.3.0/mpfr-2.3.0-rc1.tar.gz on it, ran
"./configure; make; make check; make".  But I didn't observe the
behavior you mention.  In particular, "make check" doesn't appear
to rebuild the library, and the final "make" doesn't (re)build
anything at all.

There is some potential for confusion in running "make check", in
that it builds many test programs that "make" does not build.
Thus, the first time that you run "make check", it will spend a
while compiling before it runs the tests, which may be
surprising.  It does not actually rebuild the library, though.
The second time you run "make check", it doesn't rebuild anything
and runs the tests immediately.  (This is what should happen and
what I observe on my vfat file system.)

My kernel is Linux version 2.6.18-4-686 (Debian
2.6.18.dfsg.1-11), "make" is version 3.81-3, and the vfat
partition is mounted as:
        /home/blp/tmp/vfat.bin on /mnt type vfat (rw,loop=/dev/loop0,uid=1000,gid=1000)


> I think I've found the problem. The config.log says:
>
> configure:8965: checking whether ln -s works
> configure:8972: result: no, using cp -p
>
> I wonder why "-p" is used, as on vfat, the mtime cannot be preserved.
> If the Makefile assumes that "cp -p" preserves mtime, then it is no
> longer surprising that I get many file rebuilds.

I do not think that this is the problem.  It looks to me that "ln
-s" or "cp -p" is used only by libtool, and only in final link
steps.  Thus, if this was the problem, I think that the only
visible problem would be the possibility of extra re-linkings,
not recompilations.

I'm out of my own ideas at this point, but I'm willing to
continue investigating if you have any leads.
-- 
Ben Pfaff 
http://benpfaff.org

Message #25 received at 431464@bugs.debian.org (full text, mbox, reply):

From: Vincent Lefevre <vincent@vinc17.org>

To: Ben Pfaff <blp@cs.stanford.edu>

Cc: 431464@bugs.debian.org

Subject: Re: Bug#431464: autoconf: files are rebuilt several times on vfat (because "cp -p" can't preserve time?)

Date: Sat, 7 Jul 2007 21:00:17 +0200

On 2007-07-07 11:12:59 -0700, Ben Pfaff wrote:
> There is some potential for confusion in running "make check", in
> that it builds many test programs that "make" does not build.
> Thus, the first time that you run "make check", it will spend a
> while compiling before it runs the tests, which may be
> surprising.

This is normal. The problem on the Zaurus is that after the tests are
run, the library is rebuilt.

> My kernel is Linux version 2.6.18-4-686 (Debian
> 2.6.18.dfsg.1-11), "make" is version 3.81-3, and the vfat
> partition is mounted as:
>         /home/blp/tmp/vfat.bin on /mnt type vfat (rw,loop=/dev/loop0,uid=1000,gid=1000)

On the Zaurus:

zaurus:~> uname -a
Linux zaurus 2.4.18-rmk7-pxa3-embedix #1 Thu, 6 Nov 2003 09:29:23 +0900 armv5tel unknown

GNU Make version 3.77, by Richard Stallman and Roland McGrath.

> I do not think that this is the problem.  It looks to me that "ln
> -s" or "cp -p" is used only by libtool, and only in final link
> steps.  Thus, if this was the problem, I think that the only
> visible problem would be the possibility of extra re-linkings,
> not recompilations.

I'll have to check again, and perhaps try on the internal Flash
(which is not vfat) if I have enough disk space there.

-- 
Vincent Lefèvre <vincent@vinc17.org> - Web: <http://www.vinc17.org/>
100% accessible validated (X)HTML - Blog: <http://www.vinc17.org/blog/>
Work: CR INRIA - computer arithmetic / Arenaire project (LIP, ENS-Lyon)

Message #30 received at 431464@bugs.debian.org (full text, mbox, reply):

From: Ben Pfaff <blp@cs.stanford.edu>

To: Vincent Lefevre <vincent@vinc17.org>

Cc: 431464@bugs.debian.org

Subject: Re: Bug#431464: autoconf: files are rebuilt several times on vfat (because "cp -p" can't preserve time?)

Date: Sat, 07 Jul 2007 14:17:51 -0700

Vincent Lefevre <vincent@vinc17.org> writes:

> On 2007-07-07 11:12:59 -0700, Ben Pfaff wrote:
>> There is some potential for confusion in running "make check", in
>> that it builds many test programs that "make" does not build.
>> Thus, the first time that you run "make check", it will spend a
>> while compiling before it runs the tests, which may be
>> surprising.
>
> This is normal. The problem on the Zaurus is that after the tests are
> run, the library is rebuilt.

Do you have a suggestion for how I might reproduce the problem on
a desktop system?  If I can reproduce it, I can probably debug it.
-- 
Ben Pfaff 
http://benpfaff.org

Message #35 received at 431464@bugs.debian.org (full text, mbox, reply):

From: Vincent Lefevre <vincent@vinc17.org>

To: Ben Pfaff <blp@cs.stanford.edu>

Cc: 431464@bugs.debian.org

Subject: Re: Bug#431464: autoconf: files are rebuilt several times on vfat (because "cp -p" can't preserve time?)

Date: Mon, 9 Jul 2007 14:37:19 +0200

On 2007-07-07 14:17:51 -0700, Ben Pfaff wrote:
> Do you have a suggestion for how I might reproduce the problem on
> a desktop system?  If I can reproduce it, I can probably debug it.

I've now found the problem. This is unrelated to the use of "cp -p".
On my Zaurus, GCC and some of header files are in a directory mounted
with CRAMFS. The timestamps of these files are meaningless, e.g.

-rwxr-xr-x    1 root     root          808 Jul  9 14:30 exception*
-rwxr-xr-x    1 root     root         3360 Jan  1  1970 float.h*
-rwxr-xr-x    1 root     root          275 Jan  1  1970 iso646.h*
-rwxr-xr-x    1 root     root         3264 Jul  9 13:42 limits.h*
-rwxr-xr-x    1 root     root         1207 Jan  1  1970 new*
-rwxr-xr-x    1 root     root          161 Jan  1  1970 new.h*
-rwxr-xr-x    1 root     root          137 Jan  1  1970 proto.h*
-rwxr-xr-x    1 root     root          495 Jan  1  1970 readme*
-rwxr-xr-x    1 root     root         6038 Jul  9 13:41 stdarg.h*
-rwxr-xr-x    1 root     root          502 Jan  1  1970 stdbool.h*
-rwxr-xr-x    1 root     root         9834 Jul  9 13:40 stddef.h*
-rwxr-xr-x    1 root     root          330 Jan  1  1970 syslimits.h*
[...]

As such headers are listed in the .deps/*.Plo files, I probably get
unnecessary rebuilds because of that. I wonder if it is possible to
prevent such header files from appearing in these .deps/*.Plo files.

-- 
Vincent Lefèvre <vincent@vinc17.org> - Web: <http://www.vinc17.org/>
100% accessible validated (X)HTML - Blog: <http://www.vinc17.org/blog/>
Work: CR INRIA - computer arithmetic / Arenaire project (LIP, ENS-Lyon)

Message #40 received at 431464@bugs.debian.org (full text, mbox, reply):

From: Ralf Wildenhues <Ralf.Wildenhues@gmx.de>

To: Vincent Lefevre <vincent@vinc17.org>, 431464@bugs.debian.org

Subject: Re: Bug#431464: autoconf: files are rebuilt several times on vfat (because "cp -p" can't preserve time?)

Date: Mon, 9 Jul 2007 18:50:46 +0200

Hello Vincent,

* Vincent Lefevre wrote on Mon, Jul 09, 2007 at 02:37:19PM CEST:
> 
> On my Zaurus, GCC and some of header files are in a directory mounted
> with CRAMFS. The timestamps of these files are meaningless, e.g.
> [...]
> 
> As such headers are listed in the .deps/*.Plo files, I probably get
> unnecessary rebuilds because of that. I wonder if it is possible to
> prevent such header files from appearing in these .deps/*.Plo files.

Why don't you simply
  configure --disable-dependency-tracking
?

Cheers,
Ralf

Message #45 received at 431464@bugs.debian.org (full text, mbox, reply):

From: Vincent Lefevre <vincent@vinc17.org>

To: Ralf Wildenhues <Ralf.Wildenhues@gmx.de>

Cc: 431464@bugs.debian.org

Subject: Re: Bug#431464: autoconf: files are rebuilt several times on vfat (because "cp -p" can't preserve time?)

Date: Mon, 9 Jul 2007 21:24:36 +0200

On 2007-07-09 18:50:46 +0200, Ralf Wildenhues wrote:
> Why don't you simply
>   configure --disable-dependency-tracking
> ?

Yes, I could do that on the Zaurus (I didn't know this feature as it
usually doesn't appear in "configure --help" of other software). I'll
try that.

-- 
Vincent Lefèvre <vincent@vinc17.org> - Web: <http://www.vinc17.org/>
100% accessible validated (X)HTML - Blog: <http://www.vinc17.org/blog/>
Work: CR INRIA - computer arithmetic / Arenaire project (LIP, ENS-Lyon)

Message #50 received at 431464@bugs.debian.org (full text, mbox, reply):

From: Ralf Wildenhues <Ralf.Wildenhues@gmx.de>

To: Vincent Lefevre <vincent@vinc17.org>, 431464@bugs.debian.org

Subject: Re: Bug#431464: autoconf: files are rebuilt several times on vfat (because "cp -p" can't preserve time?)

Date: Mon, 9 Jul 2007 23:32:55 +0200

* Vincent Lefevre wrote on Mon, Jul 09, 2007 at 09:24:36PM CEST:
> On 2007-07-09 18:50:46 +0200, Ralf Wildenhues wrote:
> > Why don't you simply
> >   configure --disable-dependency-tracking
> > ?
> 
> Yes, I could do that on the Zaurus (I didn't know this feature as it
> usually doesn't appear in "configure --help" of other software). I'll
> try that.

Unless the package uses the Automake option no-dependencies globally (in
which case dependencies are not used),
  ./configure --help

will show both of the following lines, and if it doesn't, please report
a bug against Automake, thanks.
  --disable-dependency-tracking  speeds up one-time build
  --enable-dependency-tracking   do not reject slow dependency extractors

And the manual describes all this in more detail:
<http://sources.redhat.com/automake/automake.html#Dependency-Tracking>

Cheers,
Ralf

Message #55 received at 431464@bugs.debian.org (full text, mbox, reply):

From: Ben Pfaff <blp@cs.stanford.edu>

To: Vincent Lefevre <vincent@vinc17.org>

Cc: 431464@bugs.debian.org

Subject: Re: Bug#431464: autoconf: files are rebuilt several times on vfat (because "cp -p" can't preserve time?)

Date: Mon, 09 Jul 2007 18:49:29 -0700

Vincent Lefevre <vincent@vinc17.org> writes:

> On 2007-07-07 14:17:51 -0700, Ben Pfaff wrote:
>> Do you have a suggestion for how I might reproduce the problem on
>> a desktop system?  If I can reproduce it, I can probably debug it.
>
> I've now found the problem. This is unrelated to the use of "cp -p".
> On my Zaurus, GCC and some of header files are in a directory mounted
> with CRAMFS. The timestamps of these files are meaningless, e.g.
>
> -rwxr-xr-x    1 root     root          808 Jul  9 14:30 exception*
> -rwxr-xr-x    1 root     root         3360 Jan  1  1970 float.h*

Is this a bug in cramfs?  You might well decide to bring this up
on the linux-kernel mailing list to figure out what's going on.

I don't think that this is a bug in Autoconf.  Do you agree?  If
so, then I will close the bug.
-- 
"I admire him, I frankly confess it; and when his time comes
 I shall buy a piece of the rope for a keepsake."
--Mark Twain

Message #60 received at 431464@bugs.debian.org (full text, mbox, reply):

From: Vincent Lefevre <vincent@vinc17.org>

To: Ben Pfaff <blp@cs.stanford.edu>

Cc: 431464@bugs.debian.org

Subject: Re: Bug#431464: autoconf: files are rebuilt several times on vfat (because "cp -p" can't preserve time?)

Date: Tue, 10 Jul 2007 08:57:42 +0200

On 2007-07-09 18:49:29 -0700, Ben Pfaff wrote:
> Vincent Lefevre <vincent@vinc17.org> writes:
> > I've now found the problem. This is unrelated to the use of "cp -p".
> > On my Zaurus, GCC and some of header files are in a directory mounted
> > with CRAMFS. The timestamps of these files are meaningless, e.g.
> >
> > -rwxr-xr-x    1 root     root          808 Jul  9 14:30 exception*
> > -rwxr-xr-x    1 root     root         3360 Jan  1  1970 float.h*
> 
> Is this a bug in cramfs?  You might well decide to bring this up
> on the linux-kernel mailing list to figure out what's going on.

It is probably a bug in cramfs. It may have already been fixed as the
Linux kernel version is quite old (this is SHARP's original ROM, and
the only way to change it is to reflash it with an unofficial one).

> I don't think that this is a bug in Autoconf.  Do you agree?  If
> so, then I will close the bug.

I agree concerning the problems I've reported. But concerning the
"cp -p", is the "-p" useful or just cosmetic?

-- 
Vincent Lefèvre <vincent@vinc17.org> - Web: <http://www.vinc17.org/>
100% accessible validated (X)HTML - Blog: <http://www.vinc17.org/blog/>
Work: CR INRIA - computer arithmetic / Arenaire project (LIP, ENS-Lyon)

Message #65 received at 431464@bugs.debian.org (full text, mbox, reply):

From: Ralf Wildenhues <Ralf.Wildenhues@gmx.de>

To: Vincent Lefevre <vincent@vinc17.org>

Subject: Re: Bug#431464: autoconf: files are rebuilt several times on vfat (because "cp -p" can't preserve time?)

Date: Tue, 10 Jul 2007 09:25:55 +0200

Hello Vincent,

* Vincent Lefevre wrote on Tue, Jul 10, 2007 at 08:57:42AM CEST:
> 
> I agree concerning the problems I've reported. But concerning the
> "cp -p", is the "-p" useful or just cosmetic?

It's useful, as it resembles the functionality it emulates more closely:
with symlinked files, 'make' considers the time stamp of the linked-to
file.

Cheers,
Ralf

Message #70 received at 431464@bugs.debian.org (full text, mbox, reply):

From: Vincent Lefevre <vincent@vinc17.org>

To: Ralf Wildenhues <Ralf.Wildenhues@gmx.de>

Cc: 431464@bugs.debian.org

Subject: Re: Bug#431464: autoconf: files are rebuilt several times on vfat (because "cp -p" can't preserve time?)

Date: Tue, 10 Jul 2007 10:31:50 +0200

On 2007-07-10 09:25:55 +0200, Ralf Wildenhues wrote:
> * Vincent Lefevre wrote on Tue, Jul 10, 2007 at 08:57:42AM CEST:
> > I agree concerning the problems I've reported. But concerning the
> > "cp -p", is the "-p" useful or just cosmetic?
> 
> It's useful, as it resembles the functionality it emulates more
> closely: with symlinked files, 'make' considers the time stamp of
> the linked-to file.

But if this timestamp is important to preserve, then there may be
problems on file systems where the timestamp cannot be preserved.

-- 
Vincent Lefèvre <vincent@vinc17.org> - Web: <http://www.vinc17.org/>
100% accessible validated (X)HTML - Blog: <http://www.vinc17.org/blog/>
Work: CR INRIA - computer arithmetic / Arenaire project (LIP, ENS-Lyon)

Message #75 received at 431464-done@bugs.debian.org (full text, mbox, reply):

From: Ben Pfaff <blp@cs.stanford.edu>

To: 431464-done@bugs.debian.org

Subject: Bug #431464: autoconf: files are rebuilt several times on vfat (because "cp -p" can't preserve time?)

Date: Mon, 27 Jul 2009 20:27:09 -0700

Version: 2.61-4

In 2007, Vincent Lefevre reported:

    Doing "make check" on the Zaurus rebuilds files several
    times: "make check" builds the test files twice, run the tests, then
    rebuilds MPFR entirely (so that I can't see the test results, in
    particular).

    I think I've found the problem. The config.log says:

    configure:8965: checking whether ln -s works
    configure:8972: result: no, using cp -p

    I wonder why "-p" is used, as on vfat, the mtime cannot be preserved.
    If the Makefile assumes that "cp -p" preserves mtime, then it is no
    longer surprising that I get many file rebuilds.

I was unable to reproduce the problem on a vfat file system on my
own system.

Vincent discovered the real problem:

    I've now found the problem. This is unrelated to the use of "cp -p".
    On my Zaurus, GCC and some of header files are in a directory mounted
    with CRAMFS. The timestamps of these files are meaningless...

    As such headers are listed in the .deps/*.Plo files, I probably get
    unnecessary rebuilds because of that. I wonder if it is possible to
    prevent such header files from appearing in these .deps/*.Plo files.

which caused me to query:

    I don't think that this is a bug in Autoconf.  Do you agree?  If
    so, then I will close the bug.

Vincent replied in the affirmative:

    I agree concerning the problems I've reported.

but I forgot to ever close the bug, so I'm doing so now (with
this email).
-- 
"...In the UNIX world, people tend to interpret `non-technical user'
 as meaning someone who's only ever written one device driver."
--Daniel Pead

Send a report that this bug log contains spam.