[gentoo-alpha] glibc-2.3.2-decc-compaq.patch

[gentoo-alpha] glibc-2.3.2-decc-compaq.patch

[Marc Giger]

Hi devs,

I'm playing around with the ccc/cxx and found some interesting things.

Can someone of you comment why the following patch is needed?

diff -ruN glibc-2.3.2.orig/include/features.h
glibc-2.3.2/include/features.h--- glibc-2.3.2.orig/include/features.h
2003-06-14 00:28:23.000000000 +0100+++ glibc-2.3.2/include/features.h   
  2003-06-14 00:58:57.000000000 +0100@@ -285,7 +285,8 @@
 #if defined __GNUC__ \
|| (defined __PGI && defined __i386__ ) \
|| (defined __INTEL_COMPILER && (defined __i386__ || defined __ia64__))\
- || (defined __STDC_VERSION__ && __STDC_VERSION__ >= 199901L)
+ || (defined __STDC_VERSION__ && __STDC_VERSION__ >= 199901L)\
+ &&!(defined(__DECC) || defined(__DECCXX)) 
# define__GLIBC_HAVE_LONG_LONG      1 
#endif

The Compaq C Compiler knows the "long long" datatype like gcc does. It's
also of the same size on both compilers 
(long == long int == long long == 8bytes). 

This breaks the compilation (with ccc) of some packages.

Comments?


Another thing is the following...:

printf '%s %s %s\n' \
                "-L${gcc_libs_path}" \
                " -SysIncDir
/usr/lib/compaq/ccc-${ccc_release}/alpha-linux/include" \               
" -SysIncDir /usr/include/linux" \                >
usr/lib/compaq/ccc-${ccc_release}/alpha-linux/bin/comp.config

..declared in the ccc ebuild. That's a big problem. I needed a lot of
time to figure out, why ccc could not compile files which had declared
INT_MAX or something like that. (@see <limits.h>)

The standard include path and the ccc include path must be declared
BEFORE gcc's include paths in comp.config, otherwise the INT_MAX & Co
Macros expands to __INT_MAX__ because the #include_next directives.
__INT_MAX__ is a predefined gcc thing.

Thank you for the time. Please mail me to discuss these.

greets

Marc

--
gentoo-alpha@gentoo.org mailing list

Re: [gentoo-alpha] glibc-2.3.2-decc-compaq.patch

[Aron Griffis]

[-- Attachment #1: Type: text/plain, Size: 1487 bytes --]

Marc Giger wrote:	[Wed Feb 04 2004, 06:13:30PM EST]
> Can someone of you comment why the following patch is needed?
> 
> diff -ruN glibc-2.3.2.orig/include/features.h
> glibc-2.3.2/include/features.h--- glibc-2.3.2.orig/include/features.h
> 2003-06-14 00:28:23.000000000 +0100+++ glibc-2.3.2/include/features.h   
>   2003-06-14 00:58:57.000000000 +0100@@ -285,7 +285,8 @@
>  #if defined __GNUC__ \
> || (defined __PGI && defined __i386__ ) \
> || (defined __INTEL_COMPILER && (defined __i386__ || defined __ia64__))\
> - || (defined __STDC_VERSION__ && __STDC_VERSION__ >= 199901L)
> + || (defined __STDC_VERSION__ && __STDC_VERSION__ >= 199901L)\
> + &&!(defined(__DECC) || defined(__DECCXX)) 
> # define__GLIBC_HAVE_LONG_LONG      1 
> #endif
> 
> The Compaq C Compiler knows the "long long" datatype like gcc does. It's
> also of the same size on both compilers 
> (long == long int == long long == 8bytes). 

You're right, this looks broken to me.

> printf '%s %s %s\n' \
>                 "-L${gcc_libs_path}" \
>                 " -SysIncDir
> /usr/lib/compaq/ccc-${ccc_release}/alpha-linux/include" \               
> " -SysIncDir /usr/include/linux" \                >
> usr/lib/compaq/ccc-${ccc_release}/alpha-linux/bin/comp.config

So do you know how this should be changed exactly?

Regards,
Aron

--
Aron Griffis
Gentoo Linux Developer (alpha / ia64 / ruby / vim)
Key fingerprint = E3B6 8734 C2D6 B5E5 AE76  FB3A 26B1 C5E3 2010 4EB0


[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-alpha] glibc-2.3.2-decc-compaq.patch

[Marc Giger]

[-- Attachment #1: Type: text/plain, Size: 2613 bytes --]

On Thu, 5 Feb 2004 22:55:36 -0500
Aron Griffis <agriffis@gentoo.org> wrote:

> Marc Giger wrote:	[Wed Feb 04 2004, 06:13:30PM EST]
> > Can someone of you comment why the following patch is needed?
> > 
> > diff -ruN glibc-2.3.2.orig/include/features.h
> > glibc-2.3.2/include/features.h---
> > glibc-2.3.2.orig/include/features.h 2003-06-14 00:28:23.000000000
> > +0100+++ glibc-2.3.2/include/features.h   
> >   2003-06-14 00:58:57.000000000 +0100@@ -285,7 +285,8 @@
> >  #if defined __GNUC__ \
> > || (defined __PGI && defined __i386__ ) \
> > || (defined __INTEL_COMPILER && (defined __i386__ || defined
> > __ia64__))\- || (defined __STDC_VERSION__ && __STDC_VERSION__ >=
> > 199901L)+ || (defined __STDC_VERSION__ && __STDC_VERSION__ >=
> > 199901L)\+ &&!(defined(__DECC) || defined(__DECCXX)) 
> > # define__GLIBC_HAVE_LONG_LONG      1 
> > #endif
> > 
> > The Compaq C Compiler knows the "long long" datatype like gcc does.
> > It's also of the same size on both compilers 
> > (long == long int == long long == 8bytes). 
> 
> You're right, this looks broken to me.

Yeah, it looks like:-) but I was wrong... I removed this check and ran
into troubles. The Macro __GLIBC_HAVE_LONG_LONG must not be defined
with ccc. So this check is totally ok! It has nothing to do with
"long long". Sorry... I was searching for the reason why I couldn't
compile python with ccc.
After some time I found it! The attached patch solves the problem. Now
I'm running python 2.3 compiled with ccc:-)


> 
> > printf '%s %s %s\n' \
> >                 "-L${gcc_libs_path}" \
> >                 " -SysIncDir
> > /usr/lib/compaq/ccc-${ccc_release}/alpha-linux/include" \           
> >    
> > " -SysIncDir /usr/include/linux" \                >
> > usr/lib/compaq/ccc-${ccc_release}/alpha-linux/bin/comp.config
> 
> So do you know how this should be changed exactly?

Hehe, no! 
This seems not to be the right place neither. I think this is useless
anyway. The right place to fix it is in "create-comp-config.sh". I tried
to change it but it seems that a shell-script and regex guru is
needed;-) The same should be changed by cxx (preventing).

With these two fixes, I'm able to compile a lot more packages with the
compaq compiler.


Oh, while we are at it. What do you think if we move the libots libs to
/lib instead of /usr/lib ?

Lastly, I ran into troubles as I compiled gawk with ccc. 
Could you imagine what happens? I've made a RAID array with several
partitions on it. One partition was /usr.
gawk is needed in the init scripts, but libots.so is located in
/usr/lib, which is not mounted yet. Too bad.

greets

Marc



[-- Attachment #2: glibc-alpha-ccc.patch --]
[-- Type: text/x-diff, Size: 533 bytes --]

--- /tmp/usr/include/sys/sysmacros.h	2004-02-06 18:34:59.000000000 +0100
+++ /usr/include/sys/sysmacros.h	2004-02-06 18:36:32.000000000 +0100
@@ -25,7 +25,7 @@
 /* If the compiler does not know long long it is out of luck.  We are
    not going to hack weird hacks to support the dev_t representation
    they need.  */
-#ifdef __GLIBC_HAVE_LONG_LONG
+#if defined __GLIBC_HAVE_LONG_LONG || defined __DECC || defined __DECCXX
 __extension__
 extern __inline unsigned int gnu_dev_major (unsigned long long int __dev)
      __THROW;




[-- Attachment #3: Type: text/plain, Size: 39 bytes --]

--
gentoo-alpha@gentoo.org mailing list

Re: [gentoo-alpha] glibc-2.3.2-decc-compaq.patch

[Aron Griffis]

[-- Attachment #1: Type: text/plain, Size: 1996 bytes --]

Marc Giger wrote:	[Fri Feb 06 2004, 02:14:11PM EST]
> Yeah, it looks like:-) but I was wrong... I removed this check and ran
> into troubles. The Macro __GLIBC_HAVE_LONG_LONG must not be defined
> with ccc. So this check is totally ok! It has nothing to do with
> "long long". Sorry... I was searching for the reason why I couldn't
> compile python with ccc.
> After some time I found it! The attached patch solves the problem. Now
> I'm running python 2.3 compiled with ccc:-)

Oh good!

> > > printf '%s %s %s\n' \
> > >                 "-L${gcc_libs_path}" \
> > >                 " -SysIncDir
> > > /usr/lib/compaq/ccc-${ccc_release}/alpha-linux/include" \           
> > >    
> > > " -SysIncDir /usr/include/linux" \                >
> > > usr/lib/compaq/ccc-${ccc_release}/alpha-linux/bin/comp.config
> > 
> > So do you know how this should be changed exactly?
> 
> Hehe, no! 
> This seems not to be the right place neither. I think this is useless
> anyway. The right place to fix it is in "create-comp-config.sh". I tried
> to change it but it seems that a shell-script and regex guru is
> needed;-) The same should be changed by cxx (preventing).
> 
> With these two fixes, I'm able to compile a lot more packages with the
> compaq compiler.

I'll take a look at create-comp-config.sh

> Oh, while we are at it. What do you think if we move the libots libs to
> /lib instead of /usr/lib ?

Sounds good to me.  Done.  Try out 2.2.7-r1.  If it works for you then
I'll mark it stable.

> Lastly, I ran into troubles as I compiled gawk with ccc. 
> Could you imagine what happens? I've made a RAID array with several
> partitions on it. One partition was /usr.
> gawk is needed in the init scripts, but libots.so is located in
> /usr/lib, which is not mounted yet. Too bad.

Ugh :-(

Regards,
Aron

--
Aron Griffis
Gentoo Linux Developer (alpha / ia64 / ruby / vim)
Key fingerprint = E3B6 8734 C2D6 B5E5 AE76  FB3A 26B1 C5E3 2010 4EB0


[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-alpha] glibc-2.3.2-decc-compaq.patch

[Marc Giger]

On Fri, 6 Feb 2004 19:40:51 -0500
Aron Griffis <agriffis@gentoo.org> wrote:

> Marc Giger wrote:	[Fri Feb 06 2004, 02:14:11PM EST]
> > Yeah, it looks like:-) but I was wrong... I removed this check and
> > ran into troubles. The Macro __GLIBC_HAVE_LONG_LONG must not be
> > defined with ccc. So this check is totally ok! It has nothing to do
> > with"long long". Sorry... I was searching for the reason why I
> > couldn't compile python with ccc.
> > After some time I found it! The attached patch solves the problem.
> > Now I'm running python 2.3 compiled with ccc:-)
> 
> Oh good!
> 
> > > > printf '%s %s %s\n' \
> > > >                 "-L${gcc_libs_path}" \
> > > >                 " -SysIncDir
> > > > /usr/lib/compaq/ccc-${ccc_release}/alpha-linux/include" \       
> > > >    
> > > >    
> > > > " -SysIncDir /usr/include/linux" \                >
> > > > usr/lib/compaq/ccc-${ccc_release}/alpha-linux/bin/comp.config
> > > 
> > > So do you know how this should be changed exactly?
> > 
> > Hehe, no! 
> > This seems not to be the right place neither. I think this is
> > useless anyway. The right place to fix it is in
> > "create-comp-config.sh". I tried to change it but it seems that a
> > shell-script and regex guru is needed;-) The same should be changed
> > by cxx (preventing).
> > 
> > With these two fixes, I'm able to compile a lot more packages with
> > the compaq compiler.
> 
> I'll take a look at create-comp-config.sh

It seems there has nothing changed. ccc-6.5.9.001-r1.ebuild gives me the
same result. Do I something wrong?

To illustrate the resulting comp.config should look as follows:

-SD/usr/lib/compaq/ccc-6.5.9.001-6/alpha-linux/include -SD/usr/include
-SD/usr/lib/gcc-lib/alpha-unknown-linux-gnu/3.3.2/include  | -SysIncDir
/usr/lib/compaq/ccc-6.5.9.001-6/alpha-linux/include -SysIncDir
/usr/include  -SysIncDir
/usr/lib/gcc-lib/alpha-unknown-linux-gnu/3.3.2/include 
-L/usr/lib/compaq/ccc-6.5.9.001-6/alpha-linux/lib  
-L/usr/lib/gcc-lib/alpha-unknown-linux-gnu/3.3.2
-L/usr/lib/gcc-lib/alpha-unknown-linux-gnu/3.3.2/../../../../alpha-unkn
own-linux-gnu/lib
-L/usr/lib/gcc-lib/alpha-unknown-linux-gnu/3.3.2/../../..

The paths must be in the following order (for -SD and -SysIncDir):

1. ccc paths
2. /usr/include
3. gcc paths

> 
> > Oh, while we are at it. What do you think if we move the libots libs
> > to/lib instead of /usr/lib ?
> 
> Sounds good to me.  Done.  Try out 2.2.7-r1.  If it works for you then
> I'll mark it stable.

This one is fine now. Thank you!

greets

Marc

--
gentoo-alpha@gentoo.org mailing list

Re: [gentoo-alpha] glibc-2.3.2-decc-compaq.patch

[Aron Griffis]

[-- Attachment #1: Type: text/plain, Size: 441 bytes --]

Marc Giger wrote:	[Sat Feb 07 2004, 07:09:56AM EST]
> It seems there has nothing changed. ccc-6.5.9.001-r1.ebuild gives me the
> same result. Do I something wrong?

No, I just didn't really understand the problem before.  Thanks for
explaining it.  I think it should work now in -r2  :-)

Regards,
Aron

--
Aron Griffis
Gentoo Linux Developer (alpha / ia64 / ruby / vim)
Key fingerprint = E3B6 8734 C2D6 B5E5 AE76  FB3A 26B1 C5E3 2010 4EB0


[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-alpha] glibc-2.3.2-decc-compaq.patch

[Marc Giger]

On Sun, 8 Feb 2004 00:37:47 -0500
Aron Griffis <agriffis@gentoo.org> wrote:

> Marc Giger wrote:	[Sat Feb 07 2004, 07:09:56AM EST]
> > It seems there has nothing changed. ccc-6.5.9.001-r1.ebuild gives me
> > the same result. Do I something wrong?
> 
> No, I just didn't really understand the problem before.  Thanks for
> explaining it.  I think it should work now in -r2  :-)

Hiya, it is right now! Thank you

Sorry for not explaining it correctly the first time but sometimes I
have to fight with my English;-)

If you are near Switzerland one day just let me know, I will offer
you a beer (or two) :-)

greets

Marc

--
gentoo-alpha@gentoo.org mailing list

Re: [gentoo-alpha] glibc-2.3.2-decc-compaq.patch

[Aron Griffis]

[-- Attachment #1: Type: text/plain, Size: 552 bytes --]

Marc Giger wrote:	[Sun Feb 08 2004, 05:14:32AM EST]
> Sorry for not explaining it correctly the first time but sometimes I
> have to fight with my English;-)

It's no problem at all.  Just keep filing bugs and telling me about
things that are wrong with Gentoo/Alpha. :-)

> If you are near Switzerland one day just let me know, I will offer
> you a beer (or two) :-)

I'm looking forward to it :-)

Regards,
Aron

--
Aron Griffis
Gentoo Linux Developer (alpha / ia64 / ruby / vim)
Key fingerprint = E3B6 8734 C2D6 B5E5 AE76  FB3A 26B1 C5E3 2010 4EB0


[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-alpha] glibc-2.3.2-decc-compaq.patch

[Marc Giger]

On Sun, 8 Feb 2004 00:37:47 -0500
Aron Griffis <agriffis@gentoo.org> wrote:

> Marc Giger wrote:	[Sat Feb 07 2004, 07:09:56AM EST]
> > It seems there has nothing changed. ccc-6.5.9.001-r1.ebuild gives me
> > the same result. Do I something wrong?
> 
> No, I just didn't really understand the problem before.  Thanks for
> explaining it.  I think it should work now in -r2  :-)

Can you do the same thing on cxx? I think it needs the same fix.

Thanks and greets

Marc

--
gentoo-alpha@gentoo.org mailing list

Re: [gentoo-alpha] glibc-2.3.2-decc-compaq.patch

[Aron Griffis]

[-- Attachment #1: Type: text/plain, Size: 1545 bytes --]

Marc Giger wrote:	[Fri Feb 06 2004, 02:14:11PM EST]
> Yeah, it looks like:-) but I was wrong... I removed this check and ran
> into troubles. The Macro __GLIBC_HAVE_LONG_LONG must not be defined
> with ccc. 

But why not?  I thought your earlier logic made sense.

> So this check is totally ok! It has nothing to do with
> "long long". Sorry... I was searching for the reason why I couldn't
> compile python with ccc.
> After some time I found it! The attached patch solves the problem. Now
> I'm running python 2.3 compiled with ccc:-)

I understand what the patch is doing, but I don't understand why
defining __GLIBC_HAVE_LONG_LONG wouldn't be a better solution?

> This seems not to be the right place neither. I think this is useless
> anyway. The right place to fix it is in "create-comp-config.sh". I tried
> to change it but it seems that a shell-script and regex guru is
> needed;-) The same should be changed by cxx (preventing).

Okay, I think I fixed it.  Try out ccc-6.5.9.001-r1.ebuild when it
arrives at your nearest mirror...  Let me know what you think.

> With these two fixes, I'm able to compile a lot more packages with the
> compaq compiler.

I'd like to be more certain about the sysmacros.h fix before putting it
in since it modifies glibc.  It seems like if that fix is needed then
other header files would need a similar fix!

Regards,
Aron

--
Aron Griffis
Gentoo Linux Developer (alpha / ia64 / ruby / vim)
Key fingerprint = E3B6 8734 C2D6 B5E5 AE76  FB3A 26B1 C5E3 2010 4EB0


[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-alpha] glibc-2.3.2-decc-compaq.patch

[Marc Giger]

On Fri, 6 Feb 2004 21:26:34 -0500
Aron Griffis <agriffis@gentoo.org> wrote:

> Marc Giger wrote:	[Fri Feb 06 2004, 02:14:11PM EST]
> > Yeah, it looks like:-) but I was wrong... I removed this check and
> > ran into troubles. The Macro __GLIBC_HAVE_LONG_LONG must not be
> > defined with ccc. 
> 
> But why not?  I thought your earlier logic made sense.

The following output shows the problems if __GLIBC_HAVE_LONG_LONG is
defined:

make: *** [Parser/acceler.o] Error 1
make: *** Waiting for unfinished jobs....
cc: Error: /usr/include/sys/types.h, line 167: In this declaration,
"int64_t" has no linkage and has a prior declaration in this scope at
line number 27 in file
/usr/lib/compaq/ccc-6.5.9.001-6/alpha-linux/include/sys/types.h.
(nolinkage)__extension__ typedef long long int int64_t;
------------------------------------^
cc: Error: /usr/include/sys/types.h, line 176: In this declaration,
"u_int64_t" has no linkage and has a prior declaration in this scope at
line number 28 in file
/usr/lib/compaq/ccc-6.5.9.001-6/alpha-linux/include/sys/types.h.
(nolinkage)__extension__ typedef unsigned long long int u_int64_t;
---------------------------------------------^
make: *** [Modules/python.o] Error 1

Do you know how to solve it if __GLIBC_HAVE_LONG_LONG is defined?

> 
> > So this check is totally ok! It has nothing to do with
> > "long long". Sorry... I was searching for the reason why I couldn't
> > compile python with ccc.
> > After some time I found it! The attached patch solves the problem.
> > Now I'm running python 2.3 compiled with ccc:-)
> 
> I understand what the patch is doing, but I don't understand why
> defining __GLIBC_HAVE_LONG_LONG wouldn't be a better solution?

See above. I forgot to mention that it is a fix for python only and I
don't know if it should be included in the glibc. I think for other
packages to compile it would need a similar fix in other places, too.
And- so we end nowhere just because we try to solve ccc problems...
This patch was ment to demonstrate only.

> 
> > This seems not to be the right place neither. I think this is
> > useless anyway. The right place to fix it is in
> > "create-comp-config.sh". I tried to change it but it seems that a
> > shell-script and regex guru is needed;-) The same should be changed
> > by cxx (preventing).
> 
> Okay, I think I fixed it.  Try out ccc-6.5.9.001-r1.ebuild when it
> arrives at your nearest mirror...  Let me know what you think.

Cool! I will try it. Thank you!

> 
> > With these two fixes, I'm able to compile a lot more packages with
> > the compaq compiler.
> 
> I'd like to be more certain about the sysmacros.h fix before putting
> it in since it modifies glibc.  It seems like if that fix is needed
> then other header files would need a similar fix!

greets

Marc

--
gentoo-alpha@gentoo.org mailing list