bug#24682: 26.0.50; Bootstrap failed

bug#24682: 26.0.50; Bootstrap failed

[Eli Zaretskii]

> From: Paul Eggert <eggert@cs.ucla.edu>
> Date: Fri, 14 Oct 2016 23:09:50 -0500
> 
> Does plain Emacs 24.5 build OK on Ubuntu 16.10?
> 
> Which Emacs version ships with Ubuntu 16.10? Emacs 24.5 or 25.1? Either 
> way, do Ubuntu's patches to Emacs seem relevant to your problem?
> 
> Others are reporting similar problems. See:
> 
> http://askubuntu.com/questions/836690/seg-fault-building-emacs25-in-ubuntu-16-10

But since a very similar problem happens with building our master
branch, I guess the Ubuntu's patches to Emacs are off the hook?

I'm guessing this is some problem with the new Ubuntu that is outside
Emacs, perhaps the kernel or the Binutils.

bug#24682: Emacs 25 from GIT build issue under Ubuntu 16.10

[Paul Eggert]

On 10/15/2016 12:56 AM, Eli Zaretskii wrote:
> This is bug#24682, please follow up there. 

configure.ac on emacs-25 knows about this problem, and attempts to build 
Emacs with -nopie. Please investigate why this is not working on Ubuntu 
16.10. I assume './configure' is outputting a string like "checking 
whether gcc accepts -nopie... no"; why is that? Try looking in 
config.log near the string "accepts -nopie".

As noted in configure.ac, this problem has occurred before; see 
Bug#18784 and Bug#20338. Possibly we merely need to tweak configure.ac 
to pacify Ubuntu's variant of GCC.

bug#24682: Emacs 25 from GIT build issue under Ubuntu 16.10

[Cesar Quiroz]

[-- Attachment #1: Type: text/plain, Size: 1173 bytes --]

-no-pie versus -nopie

There was something in the release notes about -no-pie being the standard
negation of -pie, while -nopie was introduced (by Debian?) to deal with the
issue.

Relevant snippet from config.log:

configure:31624: checking whether gcc accepts -nopie
configure:31644: gcc -o conftest -no-pie      -nopie conftest.c   >&5
gcc: error: unrecognized command line option '-nopie'; did you mean
'-no-pie'?
configure:31644: $? = 1


--César

On Sat, Oct 15, 2016 at 9:51 PM, Paul Eggert <eggert@cs.ucla.edu> wrote:

> On 10/15/2016 12:56 AM, Eli Zaretskii wrote:
>
>> This is bug#24682, please follow up there.
>>
>
> configure.ac on emacs-25 knows about this problem, and attempts to build
> Emacs with -nopie. Please investigate why this is not working on Ubuntu
> 16.10. I assume './configure' is outputting a string like "checking whether
> gcc accepts -nopie... no"; why is that? Try looking in config.log near the
> string "accepts -nopie".
>
> As noted in configure.ac, this problem has occurred before; see Bug#18784
> and Bug#20338. Possibly we merely need to tweak configure.ac to pacify
> Ubuntu's variant of GCC.
>

[-- Attachment #2: Type: text/html, Size: 2257 bytes --]

bug#24682: Emacs 25 from GIT build issue under Ubuntu 16.10

[Eli Zaretskii]

> From: Cesar Quiroz <cesar.quiroz@gmail.com>
> Date: Sat, 15 Oct 2016 22:31:46 -0700
> Cc: Michael Albinus <michael.albinus@gmx.de>, 24682@debbugs.gnu.org, 
> 	Óscar Fuentes <ofv@wanadoo.es>, Eli Zaretskii <eliz@gnu.org>, 
> 	Tim Cross <theophilusx@gmail.com>
> 
> -no-pie versus -nopie
> 
> There was something in the release notes about -no-pie being the standard negation of -pie, while -nopie was
> introduced (by Debian?) to deal with the issue.
> 
> Relevant snippet from config.log:
> 
> configure:31624: checking whether gcc accepts -nopie
> configure:31644: gcc -o conftest -no-pie -nopie conftest.c >&5 
> gcc: error: unrecognized command line option '-nopie'; did you mean '-no-pie'?
> configure:31644: $? = 1

I guess the fix, whatever it is, should be backported to the release
branch?

bug#24682: Emacs 25 from GIT build issue under Ubuntu 16.10

[Cesar Quiroz]

[-- Attachment #1: Type: text/plain, Size: 1060 bytes --]

Probably just clone the -nopie AC check macro with -no-pie, and just pass
whichever, if any, wins.

--César

On Sat, Oct 15, 2016 at 11:54 PM, Eli Zaretskii <eliz@gnu.org> wrote:

> > From: Cesar Quiroz <cesar.quiroz@gmail.com>
> > Date: Sat, 15 Oct 2016 22:31:46 -0700
> > Cc: Michael Albinus <michael.albinus@gmx.de>, 24682@debbugs.gnu.org,
> >       Óscar Fuentes <ofv@wanadoo.es>, Eli Zaretskii <eliz@gnu.org>,
> >       Tim Cross <theophilusx@gmail.com>
> >
> > -no-pie versus -nopie
> >
> > There was something in the release notes about -no-pie being the
> standard negation of -pie, while -nopie was
> > introduced (by Debian?) to deal with the issue.
> >
> > Relevant snippet from config.log:
> >
> > configure:31624: checking whether gcc accepts -nopie
> > configure:31644: gcc -o conftest -no-pie -nopie conftest.c >&5
> > gcc: error: unrecognized command line option '-nopie'; did you mean
> '-no-pie'?
> > configure:31644: $? = 1
>
> I guess the fix, whatever it is, should be backported to the release
> branch?
>

[-- Attachment #2: Type: text/html, Size: 1845 bytes --]

bug#24682: Emacs 25 from GIT build issue under Ubuntu 16.10

[Paul Eggert]

[-- Attachment #1: Type: text/plain, Size: 349 bytes --]

Cesar Quiroz wrote:
> There was something in the release notes about -no-pie being the standard
> negation of -pie, while -nopie was introduced (by Debian?) to deal with the
> issue.

Which release notes?

I installed the attached into the Emacs master branch; does it fix the problem? 
If so, I'll backport it to emacs-25 and ask you to test that.

[-- Attachment #2: 0001-Port-to-Ubuntu-16.10-which-needs-gcc-nopie.patch --]
[-- Type: text/x-diff, Size: 2613 bytes --]

From e5f26dd7f1af4bf54091a52adc3587f6a228e641 Mon Sep 17 00:00:00 2001
From: Paul Eggert <eggert@cs.ucla.edu>
Date: Sun, 16 Oct 2016 16:25:47 -0700
Subject: [PATCH] Port to Ubuntu 16.10, which needs gcc -nopie

* configure.ac (emacs_cv_prog_cc_no_pie): Rename from
emacs_cv_prog_cc_nopie.  All usages changed.  Check for -no-pie in
preference to -nopie (Bug#24682).
---
 configure.ac | 32 ++++++++++++++++++--------------
 1 file changed, 18 insertions(+), 14 deletions(-)

diff --git a/configure.ac b/configure.ac
index cd11b10..3f06eff 100644
--- a/configure.ac
+++ b/configure.ac
@@ -5123,25 +5123,29 @@ AC_DEFUN
   *) LD_SWITCH_SYSTEM_TEMACS= ;;
 esac
 
-# -nopie fixes a temacs segfault on Gentoo, OpenBSD, and other systems
-# with "hardened" GCC configurations for some reason (Bug#18784).
-# We don't know why -nopie works, but not segfaulting is better than
-# segfaulting.  Use ac_c_werror_flag=yes when trying -nopie, otherwise
-# clang keeps warning that it does not understand -nopie, and pre-4.6
-# GCC has a similar problem (Bug#20338).
-AC_CACHE_CHECK([whether $CC accepts -nopie],
-  [emacs_cv_prog_cc_nopie],
+# -no-pie or -nopie fixes a temacs segfault on Gentoo, OpenBSD,
+# Ubuntu, and other systems with "hardened" GCC configurations for
+# some reason (Bug#18784).  We don't know why this works, but not
+# segfaulting is better than segfaulting.  Use ac_c_werror_flag=yes
+# when trying the option, otherwise clang keeps warning that it does
+# not understand it, and pre-4.6 GCC has a similar problem
+# (Bug#20338).  Prefer -no-pie to -nopie, as -no-pie is the
+# spelling used by GCC 6.1.0 and later (Bug#24682).
+AC_CACHE_CHECK(
+  [for $CC option to disable position independent executables],
+  [emacs_cv_prog_cc_no_pie],
   [emacs_save_c_werror_flag=$ac_c_werror_flag
    emacs_save_LDFLAGS=$LDFLAGS
    ac_c_werror_flag=yes
-   LDFLAGS="$LDFLAGS -nopie"
-   AC_LINK_IFELSE([AC_LANG_PROGRAM([], [])],
-     [emacs_cv_prog_cc_nopie=yes],
-     [emacs_cv_prog_cc_nopie=no])
+   for emacs_cv_prog_cc_no_pie in -no-pie -nopie no; do
+     test $emacs_cv_prog_cc_no_pie = no && break
+     LDFLAGS="$emacs_save_LDFLAGS $emacs_cv_prog_cc_no_pie"
+     AC_LINK_IFELSE([AC_LANG_PROGRAM([], [])], [break])
+   done
    ac_c_werror_flag=$emacs_save_c_werror_flag
    LDFLAGS=$emacs_save_LDFLAGS])
-if test "$emacs_cv_prog_cc_nopie" = yes; then
-  LD_SWITCH_SYSTEM_TEMACS="$LD_SWITCH_SYSTEM_TEMACS -nopie"
+if test "$emacs_cv_prog_cc_no_pie" != no; then
+  LD_SWITCH_SYSTEM_TEMACS="$LD_SWITCH_SYSTEM_TEMACS $emacs_cv_prog_cc_no_pie"
 fi
 
 if test x$ac_enable_profiling != x ; then
-- 
2.7.4

bug#24682: Emacs 25 from GIT build issue under Ubuntu 16.10

[Cesar Quiroz]

[-- Attachment #1: Type: text/plain, Size: 884 bytes --]

--César

On Sun, Oct 16, 2016 at 4:56 PM, Paul Eggert <eggert@cs.ucla.edu> wrote:

> Cesar Quiroz wrote:
>
>> There was something in the release notes about -no-pie being the standard
>> negation of -pie, while -nopie was introduced (by Debian?) to deal with
>> the
>> issue.
>>
>
> Which release notes?
>

I meant 16.10:  https://wiki.ubuntu.com/YakketyYak/ReleaseNotes, and from
there specifically:

https://wiki.ubuntu.com/SecurityTeam/PIE


> I installed the attached into the Emacs master branch; does it fix the
> problem? If so, I'll backport it to emacs-25 and ask you to test that.
>

Works for master; ready to test emacs-25.

For reference, I did a 'git clean -fdx' in the git workdir, autogen, and
then I configured the build area with

$ ./config.status --config
'--with-x-toolkit=gtk3' '--with-file-notification=yes' '--with-modules'
$

[-- Attachment #2: Type: text/html, Size: 1866 bytes --]

bug#24682: Emacs 25 from GIT build issue under Ubuntu 16.10

[Cesar Quiroz]

[-- Attachment #1: Type: text/plain, Size: 1707 bytes --]

I applied the patch to my local copy of emacs-25, and it built fine.  Will
test the official port when available.

Going back to the master branch build, the log shows a warning, unrelated
to the -no-pie issue:

  CC       gmalloc.o
/work/emacs/master/src/gmalloc.c:1760:1: warning: no previous prototype for
‘hybrid_aligned_alloc’ [-Wmissing-prototypes]
 hybrid_aligned_alloc (size_t alignment, size_t size)
 ^~~~~~~~~~~~~~~~~~~~

This is not happening in the emacs-25 branch. Maybe the conditional
compilation is hiding a prototype in master but not in emacs-25. I will
take another look, and report it tomorrow if I can confirm the difference.


--César

On Sun, Oct 16, 2016 at 8:04 PM, Cesar Quiroz <cesar.quiroz@gmail.com>
wrote:

>
>
> --César
>
> On Sun, Oct 16, 2016 at 4:56 PM, Paul Eggert <eggert@cs.ucla.edu> wrote:
>
>> Cesar Quiroz wrote:
>>
>>> There was something in the release notes about -no-pie being the standard
>>> negation of -pie, while -nopie was introduced (by Debian?) to deal with
>>> the
>>> issue.
>>>
>>
>> Which release notes?
>>
>
> I meant 16.10:  https://wiki.ubuntu.com/YakketyYak/ReleaseNotes, and from
> there specifically:
>
> https://wiki.ubuntu.com/SecurityTeam/PIE
>
>
>> I installed the attached into the Emacs master branch; does it fix the
>> problem? If so, I'll backport it to emacs-25 and ask you to test that.
>>
>
> Works for master; ready to test emacs-25.
>
> For reference, I did a 'git clean -fdx' in the git workdir, autogen, and
> then I configured the build area with
>
> $ ./config.status --config
> '--with-x-toolkit=gtk3' '--with-file-notification=yes' '--with-modules'
> $
>
>
>

[-- Attachment #2: Type: text/html, Size: 3426 bytes --]

bug#24682: Emacs 25 from GIT build issue under Ubuntu 16.10

[Paul Eggert]

Cesar Quiroz wrote:
> I applied the patch to my local copy of emacs-25, and it built fine.  Will
> test the official port when available.

Thanks, I have installed the patch into the emacs-25 branch and am marking this 
bug as done.

bug#24682: Emacs 25 from GIT build issue under Ubuntu 16.10

[Cesar Quiroz]

[-- Attachment #1: Type: text/plain, Size: 428 bytes --]

Paul, thanks to you. I just rechecked after a pull, and all looks fine
still.

--César

On Sun, Oct 16, 2016 at 11:18 PM, Paul Eggert <eggert@cs.ucla.edu> wrote:

> Cesar Quiroz wrote:
>
>> I applied the patch to my local copy of emacs-25, and it built fine.  Will
>> test the official port when available.
>>
>
> Thanks, I have installed the patch into the emacs-25 branch and am marking
> this bug as done.
>

[-- Attachment #2: Type: text/html, Size: 949 bytes --]