Require at least gcc 4.2

Require at least gcc 4.2

[Vladimir 'phcoder' Serbinenko]

[-- Attachment #1: Type: text/plain, Size: 383 bytes --]

It looks like grub2's codebase triggers bugs in old compiler. I don't
feel like we should support these compilers at all. Hence this patch.
If someone wants to use older compiler he's free to patch this check
out but this way he's aware that we don't support gcc prior to 4.2

-- 
Regards
Vladimir 'phcoder' Serbinenko

Personal git repository: http://repo.or.cz/w/grub2/phcoder.git

[-- Attachment #2: gcc42.diff --]
[-- Type: text/plain, Size: 1043 bytes --]

diff --git a/ChangeLog b/ChangeLog
index 84dd6e4..916f581 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,9 @@
+2009-09-01  Vladimir Serbinenko  <phcoder@gmail.com>
+
+	Remove gcc < 4.2 support.
+
+	* configure.ac (grub_cv_cc_recent): New check.
+
 2009-08-30  Vladimir Serbinenko  <phcoder@gmail.com>
 
 	* kern/file.c (grub_file_read): Spelling fix
diff --git a/configure.ac b/configure.ac
index cddd9fe..cb389a5 100644
--- a/configure.ac
+++ b/configure.ac
@@ -213,6 +213,15 @@ CPPFLAGS="$TARGET_CPPFLAGS"
 LDFLAGS="$TARGET_LDFLAGS"
 LIBS=""
 
+AC_CACHE_CHECK([whether gcc is recent enough], grub_cv_cc_recent, [
+  CFLAGS=
+  AC_COMPILE_IFELSE([AC_LANG_PROGRAM([[
+#if defined (__GNUC__) && (__GNUC__ < 4 || __GNUC_MINOR__ < 2)
+#error Use at least GCC 4.2
+#endif
+]], [[]])],[grub_cv_cc_recent=yes],[grub_cv_cc_recent=no]) ])
+test "x$grub_cv_cc_recent" = xyes || AC_MSG_ERROR([GCC >= 4.2 is required])
+
 if test "x$TARGET_CFLAGS" = x; then
   # debug flags.
   TARGET_CFLAGS="-Wall -W -Wshadow -Wpointer-arith -Wmissing-prototypes \

Re: Require at least gcc 4.2

[Bean]

On Tue, Sep 1, 2009 at 9:44 PM, Vladimir 'phcoder'
Serbinenko<phcoder@gmail.com> wrote:
> It looks like grub2's codebase triggers bugs in old compiler. I don't
> feel like we should support these compilers at all. Hence this patch.
> If someone wants to use older compiler he's free to patch this check
> out but this way he's aware that we don't support gcc prior to 4.2

Hi,

I don't think it's gcc's version that's causing the problem, it's the
bug of some local port like apple's gcc. In fact, I've been using
mingw gcc 3.4.5 for a long time and doesn't seem to have issue. It'd
be nice to locate the specific bug and write a small test problem to
check for it (something like NESTED_FUNC_ATTR), instead of reply
blindly on gcc version. (what if apple's gcc-4.2 also contain some bug
?)

-- 
Bean

gitgrub home: http://github.com/grub/grub/
my fork page: http://github.com/bean123/grub/

Re: Require at least gcc 4.2

[Vladimir 'phcoder' Serbinenko]

On Tue, Sep 1, 2009 at 4:04 PM, Bean<bean123ch@gmail.com> wrote:
> On Tue, Sep 1, 2009 at 9:44 PM, Vladimir 'phcoder'
> Serbinenko<phcoder@gmail.com> wrote:
>> It looks like grub2's codebase triggers bugs in old compiler. I don't
>> feel like we should support these compilers at all. Hence this patch.
>> If someone wants to use older compiler he's free to patch this check
>> out but this way he's aware that we don't support gcc prior to 4.2
>
> Hi,
>
> I don't think it's gcc's version that's causing the problem, it's the
> bug of some local port like apple's gcc. In fact, I've been using
> mingw gcc 3.4.5 for a long time and doesn't seem to have issue. It'd
> be nice to locate the specific bug and write a small test problem to
> check for it (something like NESTED_FUNC_ATTR), instead of reply
> blindly on gcc version. (what if apple's gcc-4.2 also contain some bug
> ?)
There was a test to check for it but it was removed because there is
no way to do such check in cross-compiler environment.
Even if gcc<4.2 seem to work ok I think we shouldn't waste resources
on it and officially declare a minimal requirement ("you're free to go
under requirement but don't post bugreports if you do")
>
> --
> Bean
>
> gitgrub home: http://github.com/grub/grub/
> my fork page: http://github.com/bean123/grub/
>
>
> _______________________________________________
> Grub-devel mailing list
> Grub-devel@gnu.org
> http://lists.gnu.org/mailman/listinfo/grub-devel
>



-- 
Regards
Vladimir 'phcoder' Serbinenko

Personal git repository: http://repo.or.cz/w/grub2/phcoder.git

Re: Require at least gcc 4.2

[Bean]

On Tue, Sep 1, 2009 at 10:12 PM, Vladimir 'phcoder'
Serbinenko<phcoder@gmail.com> wrote:
> On Tue, Sep 1, 2009 at 4:04 PM, Bean<bean123ch@gmail.com> wrote:
>> On Tue, Sep 1, 2009 at 9:44 PM, Vladimir 'phcoder'
>> Serbinenko<phcoder@gmail.com> wrote:
>>> It looks like grub2's codebase triggers bugs in old compiler. I don't
>>> feel like we should support these compilers at all. Hence this patch.
>>> If someone wants to use older compiler he's free to patch this check
>>> out but this way he's aware that we don't support gcc prior to 4.2
>>
>> Hi,
>>
>> I don't think it's gcc's version that's causing the problem, it's the
>> bug of some local port like apple's gcc. In fact, I've been using
>> mingw gcc 3.4.5 for a long time and doesn't seem to have issue. It'd
>> be nice to locate the specific bug and write a small test problem to
>> check for it (something like NESTED_FUNC_ATTR), instead of reply
>> blindly on gcc version. (what if apple's gcc-4.2 also contain some bug
>> ?)
> There was a test to check for it but it was removed because there is
> no way to do such check in cross-compiler environment.
> Even if gcc<4.2 seem to work ok I think we shouldn't waste resources
> on it and officially declare a minimal requirement ("you're free to go
> under requirement but don't post bugreports if you do")

Hi,

Could you post a small test problem that illustrate the bug ?

-- 
Bean

gitgrub home: http://github.com/grub/grub/
my fork page: http://github.com/bean123/grub/

Re: Require at least gcc 4.2

[Robert Millan]

On Tue, Sep 01, 2009 at 03:44:18PM +0200, Vladimir 'phcoder' Serbinenko wrote:
>  
> +AC_CACHE_CHECK([whether gcc is recent enough], grub_cv_cc_recent, [
> +  CFLAGS=
> +  AC_COMPILE_IFELSE([AC_LANG_PROGRAM([[
> +#if defined (__GNUC__) && (__GNUC__ < 4 || __GNUC_MINOR__ < 2)
> +#error Use at least GCC 4.2
> +#endif

This would fail with e.g. GCC 5.0.  Other than this, the patch is fine.

If Bean wants to add a configure option (e.g. like QEMU's --disable-gcc-check)
so that this requirement can be overriden, that's ok too.

-- 
Robert Millan

  The DRM opt-in fallacy: "Your data belongs to us. We will decide when (and
  how) you may access your data; but nobody's threatening your freedom: we
  still allow you to remove your data and not access it at all."

Re: Require at least gcc 4.2

[Vladimir 'phcoder' Serbinenko]

On Tue, Sep 1, 2009 at 4:19 PM, Bean<bean123ch@gmail.com> wrote:
> On Tue, Sep 1, 2009 at 10:12 PM, Vladimir 'phcoder'
> Serbinenko<phcoder@gmail.com> wrote:
>> On Tue, Sep 1, 2009 at 4:04 PM, Bean<bean123ch@gmail.com> wrote:
>>> On Tue, Sep 1, 2009 at 9:44 PM, Vladimir 'phcoder'
>>> Serbinenko<phcoder@gmail.com> wrote:
>>>> It looks like grub2's codebase triggers bugs in old compiler. I don't
>>>> feel like we should support these compilers at all. Hence this patch.
>>>> If someone wants to use older compiler he's free to patch this check
>>>> out but this way he's aware that we don't support gcc prior to 4.2
>>>
>>> Hi,
>>>
>>> I don't think it's gcc's version that's causing the problem, it's the
>>> bug of some local port like apple's gcc. In fact, I've been using
>>> mingw gcc 3.4.5 for a long time and doesn't seem to have issue. It'd
>>> be nice to locate the specific bug and write a small test problem to
>>> check for it (something like NESTED_FUNC_ATTR), instead of reply
>>> blindly on gcc version. (what if apple's gcc-4.2 also contain some bug
>>> ?)
>> There was a test to check for it but it was removed because there is
>> no way to do such check in cross-compiler environment.
>> Even if gcc<4.2 seem to work ok I think we shouldn't waste resources
>> on it and officially declare a minimal requirement ("you're free to go
>> under requirement but don't post bugreports if you do")
>
> Hi,
>
> Could you post a small test problem that illustrate the bug ?
Yves Blusseau reported to me that modules compiled with gcc-4.0 based
Apple's CC failed whereas the ones compiled with gcc-4.2 worked well.
As MacOSX is the only OS I'm aware of which ships pre-4.2 gcc by
default (the actually ship both 4.0 and 4.2) I thought it may be just
a gcc 4.0 bug. Even if it's not so I still think dropping support for
old building environments is a good thing since it decreases bug
possibilities and so maintainance work too. If you want to add
--disable-gcc-check it's ok with me. Just take my patch as base,
replace the check line with
#if defined (__GNUC__) && (__GNUC__ < 4 || (__GNUC__ == 4 &&
__GNUC_MINOR__ < 2))
add --disable-gcc-check and commit (since it's approved by Robert
Millan in another mail)
>
> --
> Bean
>
> gitgrub home: http://github.com/grub/grub/
> my fork page: http://github.com/bean123/grub/
>
>
> _______________________________________________
> Grub-devel mailing list
> Grub-devel@gnu.org
> http://lists.gnu.org/mailman/listinfo/grub-devel
>



-- 
Regards
Vladimir 'phcoder' Serbinenko

Personal git repository: http://repo.or.cz/w/grub2/phcoder.git

Re: Require at least gcc 4.2

[Seth Goldberg]

>>
>> Hi,
>>
>> Could you post a small test problem that illustrate the bug ?
> Yves Blusseau reported to me that modules compiled with gcc-4.0 based 
> Apple's CC failed whereas the ones compiled with gcc-4.2 worked well. As 
> MacOSX is the only OS I'm aware of which ships pre-4.2 gcc by default (the

   Unfortunately, Solaris currently ships with

% /usr/sfw/bin/gcc -v
gcc version 3.4.3 (csl-sol210-3_4-20050802)

   I'm not sure when we're updating to 4.x (hopefully soon, for many reasons, 
but I just don't know).

  --S

Re: Require at least gcc 4.2

[David Miller]

From: "Vladimir 'phcoder' Serbinenko" <phcoder@gmail.com>
Date: Tue, 1 Sep 2009 16:12:20 +0200

> Even if gcc<4.2 seem to work ok I think we shouldn't waste resources
> on it and officially declare a minimal requirement ("you're free to go
> under requirement but don't post bugreports if you do")

You're making it such that I won't be able to do development on my
main sparc development machine, which is using gcc-4.1.3

In my view, this is going a bit overboard as a method to address this
problem.

Re: Require at least gcc 4.2

[David Miller]

From: "Vladimir 'phcoder' Serbinenko" <phcoder@gmail.com>
Date: Tue, 1 Sep 2009 19:02:39 +0200

> Even if it's not so I still think dropping support for old building
> environments is a good thing since it decreases bug possibilities
> and so maintainance work too.

It also decreases your testing base.

I think this change is short-sighted.

If anything, only make this restriction on x86 platforms,
which is the only place where this issue even happens.

Re: Require at least gcc 4.2

[Robert Millan]

On Tue, Sep 01, 2009 at 03:23:01PM -0700, David Miller wrote:
> From: "Vladimir 'phcoder' Serbinenko" <phcoder@gmail.com>
> Date: Tue, 1 Sep 2009 16:12:20 +0200
> 
> > Even if gcc<4.2 seem to work ok I think we shouldn't waste resources
> > on it and officially declare a minimal requirement ("you're free to go
> > under requirement but don't post bugreports if you do")
> 
> You're making it such that I won't be able to do development on my
> main sparc development machine, which is using gcc-4.1.3
> 
> In my view, this is going a bit overboard as a method to address this
> problem.

Why not just skip the check?  (e.g. --disable-gcc-check)

-- 
Robert Millan

  The DRM opt-in fallacy: "Your data belongs to us. We will decide when (and
  how) you may access your data; but nobody's threatening your freedom: we
  still allow you to remove your data and not access it at all."

Re: Require at least gcc 4.2

[Robert Millan]

On Tue, Sep 01, 2009 at 06:07:04PM +0200, Robert Millan wrote:
> On Tue, Sep 01, 2009 at 03:44:18PM +0200, Vladimir 'phcoder' Serbinenko wrote:
> >  
> > +AC_CACHE_CHECK([whether gcc is recent enough], grub_cv_cc_recent, [
> > +  CFLAGS=
> > +  AC_COMPILE_IFELSE([AC_LANG_PROGRAM([[
> > +#if defined (__GNUC__) && (__GNUC__ < 4 || __GNUC_MINOR__ < 2)
> > +#error Use at least GCC 4.2
> > +#endif
> 
> This would fail with e.g. GCC 5.0.  Other than this, the patch is fine.
> 
> If Bean wants to add a configure option (e.g. like QEMU's --disable-gcc-check)
> so that this requirement can be overriden, that's ok too.

Btw, INSTALL file still says 2.95.  If we do this, it should be adjusted as
well.

-- 
Robert Millan

  The DRM opt-in fallacy: "Your data belongs to us. We will decide when (and
  how) you may access your data; but nobody's threatening your freedom: we
  still allow you to remove your data and not access it at all."

Re: Require at least gcc 4.2

[Vladimir 'phcoder' Serbinenko]

On Thu, Sep 3, 2009 at 4:58 PM, Robert Millan<rmh@aybabtu.com> wrote:
> On Tue, Sep 01, 2009 at 06:07:04PM +0200, Robert Millan wrote:
>> On Tue, Sep 01, 2009 at 03:44:18PM +0200, Vladimir 'phcoder' Serbinenko wrote:
>> >
>> > +AC_CACHE_CHECK([whether gcc is recent enough], grub_cv_cc_recent, [
>> > +  CFLAGS=
>> > +  AC_COMPILE_IFELSE([AC_LANG_PROGRAM([[
>> > +#if defined (__GNUC__) && (__GNUC__ < 4 || __GNUC_MINOR__ < 2)
>> > +#error Use at least GCC 4.2
>> > +#endif
>>
>> This would fail with e.g. GCC 5.0.  Other than this, the patch is fine.
>>
>> If Bean wants to add a configure option (e.g. like QEMU's --disable-gcc-check)
>> so that this requirement can be overriden, that's ok too.
>
> Btw, INSTALL file still says 2.95.  If we do this, it should be adjusted as
> well.
>
Cosidering recent gcc's still have nested function bug I consider
forcing 4.2 is absolutely unnecessary.
However INSTALL needs to be upgraded and mention a newer GCC since
grub2 isn't compileable with gcc 2.95
> --
> Robert Millan
>
>  The DRM opt-in fallacy: "Your data belongs to us. We will decide when (and
>  how) you may access your data; but nobody's threatening your freedom: we
>  still allow you to remove your data and not access it at all."
>
>
> _______________________________________________
> Grub-devel mailing list
> Grub-devel@gnu.org
> http://lists.gnu.org/mailman/listinfo/grub-devel
>



-- 
Regards
Vladimir 'phcoder' Serbinenko

Personal git repository: http://repo.or.cz/w/grub2/phcoder.git

Re: Require at least gcc 4.2

[Robert Millan]

On Sat, Sep 05, 2009 at 11:01:32PM +0200, Vladimir 'phcoder' Serbinenko wrote:
> On Thu, Sep 3, 2009 at 4:58 PM, Robert Millan<rmh@aybabtu.com> wrote:
> > On Tue, Sep 01, 2009 at 06:07:04PM +0200, Robert Millan wrote:
> >> On Tue, Sep 01, 2009 at 03:44:18PM +0200, Vladimir 'phcoder' Serbinenko wrote:
> >> >
> >> > +AC_CACHE_CHECK([whether gcc is recent enough], grub_cv_cc_recent, [
> >> > +  CFLAGS=
> >> > +  AC_COMPILE_IFELSE([AC_LANG_PROGRAM([[
> >> > +#if defined (__GNUC__) && (__GNUC__ < 4 || __GNUC_MINOR__ < 2)
> >> > +#error Use at least GCC 4.2
> >> > +#endif
> >>
> >> This would fail with e.g. GCC 5.0.  Other than this, the patch is fine.
> >>
> >> If Bean wants to add a configure option (e.g. like QEMU's --disable-gcc-check)
> >> so that this requirement can be overriden, that's ok too.
> >
> > Btw, INSTALL file still says 2.95.  If we do this, it should be adjusted as
> > well.
> >
> Cosidering recent gcc's still have nested function bug I consider
> forcing 4.2 is absolutely unnecessary.
> However INSTALL needs to be upgraded and mention a newer GCC since
> grub2 isn't compileable with gcc 2.95

David et al, can we agree on a minimal version of GCC that we should recommend?

For mentioning in the INSTALL file only.  I assume nobody's interested in
gcc 2.95.

-- 
Robert Millan

  The DRM opt-in fallacy: "Your data belongs to us. We will decide when (and
  how) you may access your data; but nobody's threatening your freedom: we
  still allow you to remove your data and not access it at all."

Re: Require at least gcc 4.2

[David Miller]

From: Robert Millan <rmh@aybabtu.com>
Date: Sat, 5 Sep 2009 23:26:11 +0200

> David et al, can we agree on a minimal version of GCC that we should
> recommend?
> 
> For mentioning in the INSTALL file only.  I assume nobody's interested in
> gcc 2.95.

I think gcc-4.1.x, maybe specifically gcc-4.1.3 is a reasonable
base point.

If we want to go back as far as gcc-3.4.x, probably gcc-3.4.6

Re: Require at least gcc 4.2

[Neil Cafferkey]

On Thu, Sep 03, 2009 at 04:21:55PM +0200, Robert Millan wrote:
> On Tue, Sep 01, 2009 at 03:23:01PM -0700, David Miller wrote:
> > From: "Vladimir 'phcoder' Serbinenko" <phcoder@gmail.com>
> > Date: Tue, 1 Sep 2009 16:12:20 +0200
> > 
> > > Even if gcc<4.2 seem to work ok I think we shouldn't waste resources
> > > on it and officially declare a minimal requirement ("you're free to go
> > > under requirement but don't post bugreports if you do")
> > 
> > You're making it such that I won't be able to do development on my
> > main sparc development machine, which is using gcc-4.1.3
> > 
> > In my view, this is going a bit overboard as a method to address this
> > problem.
> 
> Why not just skip the check?  (e.g. --disable-gcc-check)

The central issue IMO isn't whether this is enforced by configure, but
whether it's legitimate to report failure to build with older compilers
as a bug.

No offence, but this proposal strikes me as slightly lazy. In other words,
I think these bugs should be identified instead of being "fixed" by
declaring "your compiler is too old".

FWIW, I still use i386-elf-gcc 3.4.3 under Mac OS X, because (last time I
checked) it's the most recent version available from macports. The fact is
that a lot of people outside those who install a new version of Linux
every six months or so are still using these old compilers.

Regards,
Neil

Re: Require at least gcc 4.2

[Robert Millan]

On Sat, Sep 05, 2009 at 03:12:14PM -0700, David Miller wrote:
> From: Robert Millan <rmh@aybabtu.com>
> Date: Sat, 5 Sep 2009 23:26:11 +0200
> 
> > David et al, can we agree on a minimal version of GCC that we should
> > recommend?
> > 
> > For mentioning in the INSTALL file only.  I assume nobody's interested in
> > gcc 2.95.
> 
> I think gcc-4.1.x, maybe specifically gcc-4.1.3 is a reasonable
> base point.

Ok.  For the time being it'll list GCC 4.1.3.

-- 
Robert Millan

  The DRM opt-in fallacy: "Your data belongs to us. We will decide when (and
  how) you may access your data; but nobody's threatening your freedom: we
  still allow you to remove your data and not access it at all."

Re: Require at least gcc 4.2

[Robert Millan]

On Sun, Sep 06, 2009 at 04:03:28PM +0100, Neil Cafferkey wrote:
> 
> No offence, but this proposal strikes me as slightly lazy. In other words,
> I think these bugs should be identified instead of being "fixed" by
> declaring "your compiler is too old".

In this case, we weren't speaking about bugs we had, which would need to be
fixed, but about bugs in GCC, which had been presumably fixed in GCC, and for
which we were being dragged into keeping workarounds.

It turns out the problem in GCC is not really considered to be "fixable", and
therefore might not be seen as a bug.  So this discussion is now moot, but the
general principle stands.

> FWIW, I still use i386-elf-gcc 3.4.3 under Mac OS X, because (last time I
> checked) it's the most recent version available from macports.

I just checked.  There's now GCC 4.4, and even 4.5 snapshots.

> The fact is
> that a lot of people outside those who install a new version of Linux
> every six months or so are still using these old compilers.

Debian etch was released more than 2 years ago, and it already had GCC 4.1.

-- 
Robert Millan

  The DRM opt-in fallacy: "Your data belongs to us. We will decide when (and
  how) you may access your data; but nobody's threatening your freedom: we
  still allow you to remove your data and not access it at all."