EGCS 1.1.1 FYI

EGCS 1.1.1 FYI

[jeramy]

Cygnus released 1.1.1 on Dec. 1 with the following PPC specific
improvements.

Fixed PPC/RS6000 LEGITIMIZE_ADDRESS macro and bug in conversion from
unsigned ints to double precision floats. 
Fixed handling of long longs for varargs/stdarg functions on the ppc. 
Fixed code generation problems with gpr<->fpr copies for 64bit ppc. 
Makes sure target libraries build with -fPIC for PPC targets.

Good to see this much platform specific progress in a x.x.1 release. :)

[[ This message was sent via the linuxppc-dev mailing list. Replies are ]]
[[ not forced back to the list, so be sure to  Cc linuxppc-dev  if your ]]
[[ reply is of general interest. To unsubscribe from linuxppc-dev, send ]]
[[ the message 'unsubscribe' to linuxppc-dev-request@lists.linuxppc.org ]]

Re: EGCS 1.1.1 FYI

[Michael Meissner]

On Sun, Dec 06, 1998 at 02:06:46AM -0600, jeramy wrote:
> 
> Cygnus released 1.1.1 on Dec. 1 with the following PPC specific
> improvements.

Note, while we host the egcs machines (and provide a number of compiler
hackers), EGCS != Cygnus.  EGCS is netwide and has people from many different
organizations (including most of the developers from the FSF GCC, and many more
developers that were never part of the FSF).

> Fixed PPC/RS6000 LEGITIMIZE_ADDRESS macro and bug in conversion from
> unsigned ints to double precision floats. 
> Fixed handling of long longs for varargs/stdarg functions on the ppc. 

Note, this means any code that passes long long should have both the caller and
callee recompiled with the same compiler.

> Fixed code generation problems with gpr<->fpr copies for 64bit ppc. 

That doesn't affect PowerPC Linux right now, since the only 64-bit ppc target
is AIX based (there are a number of issues like gas/ld support, and machine
availability before it will matter to Linux/PPC).

> Makes sure target libraries build with -fPIC for PPC targets.
> 
> Good to see this much platform specific progress in a x.x.1 release. :)

-- 
Michael Meissner, Cygnus Solutions (Massachusetts office)
4th floor, 955 Massachusetts Avenue, Cambridge, MA 02139, USA
meissner@cygnus.com,	617-354-5416 (office),	617-354-7161 (fax)

[[ This message was sent via the linuxppc-dev mailing list. Replies are ]]
[[ not forced back to the list, so be sure to  Cc linuxppc-dev  if your ]]
[[ reply is of general interest. To unsubscribe from linuxppc-dev, send ]]
[[ the message 'unsubscribe' to linuxppc-dev-request@lists.linuxppc.org ]]

Re: EGCS 1.1.1 FYI

[Paul J. Schinder]

On Sun, 06 Dec 1998 02:06:46 -0600, jeramy  wrote:
>
>
>Cygnus released 1.1.1 on Dec. 1 with the following PPC specific
>improvements.
>
>Fixed PPC/RS6000 LEGITIMIZE_ADDRESS macro and bug in conversion from
>unsigned ints to double precision floats. 
>Fixed handling of long longs for varargs/stdarg functions on the ppc. 
>Fixed code generation problems with gpr<->fpr copies for 64bit ppc. 
>Makes sure target libraries build with -fPIC for PPC targets.
>
>Good to see this much platform specific progress in a x.x.1 release. :)

So are Gary Thomas' patches still necessary? I downloaded his src rpm and
applied the patches that would apply to 1.1.1. As I recall, one said it
was aIready there, and a couple applied with offsets. I didn't use the rpm
itself because after bootstrapping the compiler and running make check the
Fortran compiler failed most of its tests, and I need a working Fortran
compiler. (And I nearly gave up on the rpm before stumbling across the
gperf patch in the contrib directory.) The make check on 1.1.1 failed
fewer tests than any egcs build I've done. I'm not that interested in
building kernels, but I do need working Fortran. I'm using 2.1.127 and
glibc-1.99 on a Performa 6400.







-------
Paul J. Schinder
schinder@pobox.com


[[ This message was sent via the linuxppc-dev mailing list. Replies are ]]
[[ not forced back to the list, so be sure to  Cc linuxppc-dev  if your ]]
[[ reply is of general interest. To unsubscribe from linuxppc-dev, send ]]
[[ the message 'unsubscribe' to linuxppc-dev-request@lists.linuxppc.org ]]

Re: EGCS 1.1.1 FYI

[David Edelsohn]

>>>>> "Paul J Schinder" writes:

Paul> So are Gary Thomas' patches still necessary?

	The trampoline patch now has become a shared library compatibility
issue for glibc-1.99 support.  It only is necessary until LinuxPPC cuts
over to glibc-2.1 which will require other binary incompatible changes at
the same time.

David

[[ This message was sent via the linuxppc-dev mailing list. Replies are ]]
[[ not forced back to the list, so be sure to  Cc linuxppc-dev  if your ]]
[[ reply is of general interest. To unsubscribe from linuxppc-dev, send ]]
[[ the message 'unsubscribe' to linuxppc-dev-request@lists.linuxppc.org ]]

Re: EGCS 1.1.1 FYI

[Ralf Weidemann]

>>>>> "David" == David Edelsohn <dje@watson.ibm.com> writes:
>>>>> "Paul J Schinder" writes:
    Paul> So are Gary Thomas' patches still necessary?

    David> 	The trampoline patch now has become a shared library
    David> compatibility issue for glibc-1.99 support.  It only is
    David> necessary until LinuxPPC cuts over to glibc-2.1 which will
    David> require other binary incompatible changes at the same time.

Can someone clarify where the problem is exact with native
egcs and libc-1.99 ? I use egcs-1.1.1 since the weekend and
it seems to work fine. I had no trouble to build some shared
lib stuff like in the latest jade or sgml-tools. This packages
didn't compile with older version I tried (also Gary's egcs-1.1b),
but that were generell compiler issues, I think.

But on the other side I also tried to run the SeaMonkeyBrowser
(Mozilla) from recent cvs-sources. Building went fine, but the
beast didn't run. I get only a simple "Aborted" message on the
console and it seems to crash somewhere in the startup code.
So maybe this is somehow related ?

Had someone luck to run a recent version of Mozilla at all on
LinuxPPC ?

Thanks,
Ralf




[[ This message was sent via the linuxppc-dev mailing list. Replies are ]]
[[ not forced back to the list, so be sure to  Cc linuxppc-dev  if your ]]
[[ reply is of general interest. To unsubscribe from linuxppc-dev, send ]]
[[ the message 'unsubscribe' to linuxppc-dev-request@lists.linuxppc.org ]]

Re: EGCS 1.1.1 FYI

[Michael Meissner]

On Mon, Dec 07, 1998 at 11:22:42PM +0100, Ralf Weidemann wrote:
> 
> >>>>> "David" == David Edelsohn <dje@watson.ibm.com> writes:
> >>>>> "Paul J Schinder" writes:
>     Paul> So are Gary Thomas' patches still necessary?
> 
>     David> 	The trampoline patch now has become a shared library
>     David> compatibility issue for glibc-1.99 support.  It only is
>     David> necessary until LinuxPPC cuts over to glibc-2.1 which will
>     David> require other binary incompatible changes at the same time.
> 
> Can someone clarify where the problem is exact with native
> egcs and libc-1.99 ? I use egcs-1.1.1 since the weekend and
> it seems to work fine. I had no trouble to build some shared
> lib stuff like in the latest jade or sgml-tools. This packages
> didn't compile with older version I tried (also Gary's egcs-1.1b),
> but that were generell compiler issues, I think.

There is a difference of opinion on the difference of the size of the
instruction block for trampolines coupled with a coding error in the trampoline
support module.  Some version of the compiler other than the egcs/fsf compiler
has upped the size of the trampolines to be larger than the 40 bytes checked
for, and the tramp.asm module check for too small of a block aborts due to the
test being backwards.  If you don't use trampolines (including running the
compiler testsuite), you won't see the problem.  If you don't know what
trampolines are then you probably aren't using them.

<arrogant opinion>
Trampolines are a botch, and I have encounted problems with them in just above
every single GCC platform I have worked on in the last 9 years, but RMS loves
them, so they have remained.  What they are is a chunk of stack that the
compiler puts instructions to load up the static chain, and jump to the real
function in the location.  This address is then passed when you take the
address of a nested (lexically scoped) function.
</arrogant opinion>

> But on the other side I also tried to run the SeaMonkeyBrowser
> (Mozilla) from recent cvs-sources. Building went fine, but the
> beast didn't run. I get only a simple "Aborted" message on the
> console and it seems to crash somewhere in the startup code.
> So maybe this is somehow related ?

If SeaMonkeyBrowser calls through a trampoline inside of static constructors,
then yes, this is a symptom.

-- 
Michael Meissner, Cygnus Solutions (Massachusetts office)
4th floor, 955 Massachusetts Avenue, Cambridge, MA 02139, USA
meissner@cygnus.com,	617-354-5416 (office),	617-354-7161 (fax)

[[ This message was sent via the linuxppc-dev mailing list. Replies are ]]
[[ not forced back to the list, so be sure to  Cc linuxppc-dev  if your ]]
[[ reply is of general interest. To unsubscribe from linuxppc-dev, send ]]
[[ the message 'unsubscribe' to linuxppc-dev-request@lists.linuxppc.org ]]

Re: EGCS 1.1.1 FYI

[Elgin Lee]

>>>>> In <19981207200522.A4510@tiktok.cygnus.com> 
>>>>>	Michael Meissner <meissner@cygnus.com> wrote:

> test being backwards.  If you don't use trampolines (including running the
> compiler testsuite), you won't see the problem.  If you don't know what
> trampolines are then you probably aren't using them.

glibc-0.961212 uses trampolines for the nsswitch-based name resolver
functions (e.g., gethostbyname()).  If you use an unpatched egcs-1.1.1
to compile a program using one of those routines, the program will
crash on that call.

--Elgin

[[ This message was sent via the linuxppc-dev mailing list. Replies are ]]
[[ not forced back to the list, so be sure to  Cc linuxppc-dev  if your ]]
[[ reply is of general interest. To unsubscribe from linuxppc-dev, send ]]
[[ the message 'unsubscribe' to linuxppc-dev-request@lists.linuxppc.org ]]

Re: EGCS 1.1.1 FYI

[Gary Thomas]

On 08-Dec-98 Michael Meissner wrote:
> 
> On Mon, Dec 07, 1998 at 11:22:42PM +0100, Ralf Weidemann wrote:
>> But on the other side I also tried to run the SeaMonkeyBrowser
>> (Mozilla) from recent cvs-sources. Building went fine, but the
>> beast didn't run. I get only a simple "Aborted" message on the
>> console and it seems to crash somewhere in the startup code.
>> So maybe this is somehow related ?
> 
> If SeaMonkeyBrowser calls through a trampoline inside of static constructors,
> then yes, this is a symptom.
> 

I'm sure that the problem is here.  If you've built code with a compiler
that *does not have* the trampoline patch (which is what I think you've
done with Mozilla) then the startup for 'libstdc++' will fail (amoung
others).

I've built a new EGCS-1.1.1 which can be used with existing code until
the new GLIBC-2.1.xx is ready.  You can get this from:
  ftp://ftp.linuxppc.org/pub/gary/redhat/RPMS/ppc
Pick up the 'egcs-1.1-1b*' packages.

------------------------------------------------------------------------
Gary Thomas                              |
email: gdt@linuxppc.org                  | "Fine wine is a necessity of
   ... opinions expressed here are mine  |        life for me"
       and no one else would claim them! |
                                         |      Thomas Jefferson
------------------------------------------------------------------------



[[ This message was sent via the linuxppc-dev mailing list. Replies are ]]
[[ not forced back to the list, so be sure to  Cc linuxppc-dev  if your ]]
[[ reply is of general interest. To unsubscribe from linuxppc-dev, send ]]
[[ the message 'unsubscribe' to linuxppc-dev-request@lists.linuxppc.org ]]

Re: EGCS 1.1.1 FYI

[hagopiar]

[-- Attachment #1: Type: TEXT/PLAIN, Size: 1594 bytes --]

FYI, I get the same problem with glibc-2.0.104... I have an strace, last
50 lines are attached... No time to do anything anymore... <sigh>
								-Rob

On Mon, 7 Dec 1998, Ralf Weidemann wrote:

> Date: Mon, 7 Dec 1998 23:22:42 +0100 (CET)
> From: Ralf Weidemann <RalfW@Scalar.de>
> To: LinuxPPC Dev <linuxppc-dev@lists.linuxppc.org>
> Subject: Re: EGCS 1.1.1 FYI 
> 
> 
> >>>>> "David" == David Edelsohn <dje@watson.ibm.com> writes:
> >>>>> "Paul J Schinder" writes:
>     Paul> So are Gary Thomas' patches still necessary?
> 
>     David> 	The trampoline patch now has become a shared library
>     David> compatibility issue for glibc-1.99 support.  It only is
>     David> necessary until LinuxPPC cuts over to glibc-2.1 which will
>     David> require other binary incompatible changes at the same time.
> 
> Can someone clarify where the problem is exact with native
> egcs and libc-1.99 ? I use egcs-1.1.1 since the weekend and
> it seems to work fine. I had no trouble to build some shared
> lib stuff like in the latest jade or sgml-tools. This packages
> didn't compile with older version I tried (also Gary's egcs-1.1b),
> but that were generell compiler issues, I think.
> 
> But on the other side I also tried to run the SeaMonkeyBrowser
> (Mozilla) from recent cvs-sources. Building went fine, but the
> beast didn't run. I get only a simple "Aborted" message on the
> console and it seems to crash somewhere in the startup code.
> So maybe this is somehow related ?
> 
> Had someone luck to run a recent version of Mozilla at all on
> LinuxPPC ?
> 
> Thanks,
> Ralf
> 
> 
> 
> 

[-- Attachment #2: Type: TEXT/PLAIN, Size: 2681 bytes --]

fcntl(4, F_SETFL, O_RDONLY|O_NONBLOCK)  = 0
fcntl(5, F_GETFL)                       = 0x1 (flags O_WRONLY)
fcntl(5, F_SETFL, O_WRONLY|O_NONBLOCK)  = 0
brk(0)                                  = 0x1877000
brk(0x1878000)                          = 0x1878000
SYS_173(0xd, 0x7fffe408, 0, 0x8, 0x20)  = 0
SYS_173(0xe, 0x7fffe408, 0, 0x8, 0x20)  = 0
open("/root/.mozilla/cookies", O_RDONLY) = -1 ENOENT (No such file or directory)
stat("/root/.mozilla/index.db", 0x7ffff298) = -1 ENOENT (No such file or directory)
stat("/root/.mozilla/index.db", 0x7ffff298) = -1 ENOENT (No such file or directory)
brk(0)                                  = 0x1878000
brk(0x1879000)                          = 0x1879000
brk(0)                                  = 0x1879000
brk(0x187a000)                          = 0x187a000
brk(0)                                  = 0x187a000
brk(0x187b000)                          = 0x187b000
brk(0)                                  = 0x187b000
brk(0x187c000)                          = 0x187c000
brk(0)                                  = 0x187c000
brk(0x187d000)                          = 0x187d000
brk(0)                                  = 0x187d000
brk(0x187e000)                          = 0x187e000
brk(0)                                  = 0x187e000
brk(0x187f000)                          = 0x187f000
brk(0)                                  = 0x187f000
brk(0x1880000)                          = 0x1880000
brk(0)                                  = 0x1880000
brk(0x1881000)                          = 0x1881000
brk(0)                                  = 0x1881000
brk(0x1882000)                          = 0x1882000
brk(0)                                  = 0x1882000
brk(0x1883000)                          = 0x1883000
brk(0)                                  = 0x1883000
brk(0x1884000)                          = 0x1884000
brk(0)                                  = 0x1884000
brk(0x1885000)                          = 0x1885000
brk(0)                                  = 0x1885000
brk(0x1886000)                          = 0x1886000
brk(0)                                  = 0x1886000
brk(0x1887000)                          = 0x1887000
brk(0)                                  = 0x1887000
brk(0x1888000)                          = 0x1888000
brk(0)                                  = 0x1888000
brk(0x1889000)                          = 0x1889000
brk(0)                                  = 0x1889000
brk(0x188a000)                          = 0x188a000
brk(0)                                  = 0x188a000
brk(0x188b000)                          = 0x188b000
--- SIGSEGV (Segmentation fault) ---
+++ killed by SIGSEGV +++

Re: EGCS 1.1.1 FYI

[David T. McWherter]

I've tried compiling the mozilla code too, but it aborts due
to a failing dynamic linking call in normal startup code...
it's particularly odd, as it's trying to dlopen a file with
a name parameter of NULL, and it's not even trying to hide it
(the last code I saw had a dlopen(NULL,...).  I'm *guessing*
that on other Linux platforms, such code is well-defined, 
perhaps to use the "current executable file" to do the opening,
or something...but I'm not sure...still kinda odd to me...

-David

On Mon, 7
Dec
1998 hagopiar@vuser.vu.union.edu wrote:

> FYI, I get the same problem with glibc-2.0.104... I have an strace, last
> 50 lines are attached... No time to do anything anymore... <sigh>
> 								-Rob
> 
> On Mon, 7 Dec 1998, Ralf Weidemann wrote:
> 
> > Date: Mon, 7 Dec 1998 23:22:42 +0100 (CET)
> > From: Ralf Weidemann <RalfW@Scalar.de>
> > To: LinuxPPC Dev <linuxppc-dev@lists.linuxppc.org>
> > Subject: Re: EGCS 1.1.1 FYI 
> > 
> > 
> > >>>>> "David" == David Edelsohn <dje@watson.ibm.com> writes:
> > >>>>> "Paul J Schinder" writes:
> >     Paul> So are Gary Thomas' patches still necessary?
> > 
> >     David> 	The trampoline patch now has become a shared library
> >     David> compatibility issue for glibc-1.99 support.  It only is
> >     David> necessary until LinuxPPC cuts over to glibc-2.1 which will
> >     David> require other binary incompatible changes at the same time.
> > 
> > Can someone clarify where the problem is exact with native
> > egcs and libc-1.99 ? I use egcs-1.1.1 since the weekend and
> > it seems to work fine. I had no trouble to build some shared
> > lib stuff like in the latest jade or sgml-tools. This packages
> > didn't compile with older version I tried (also Gary's egcs-1.1b),
> > but that were generell compiler issues, I think.
> > 
> > But on the other side I also tried to run the SeaMonkeyBrowser
> > (Mozilla) from recent cvs-sources. Building went fine, but the
> > beast didn't run. I get only a simple "Aborted" message on the
> > console and it seems to crash somewhere in the startup code.
> > So maybe this is somehow related ?
> > 
> > Had someone luck to run a recent version of Mozilla at all on
> > LinuxPPC ?
> > 
> > Thanks,
> > Ralf
> > 
> > 
> > 
> > 
> 

-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
       David T. McWherter     dtm@waterw.com


[[ This message was sent via the linuxppc-dev mailing list. Replies are ]]
[[ not forced back to the list, so be sure to  Cc linuxppc-dev  if your ]]
[[ reply is of general interest. To unsubscribe from linuxppc-dev, send ]]
[[ the message 'unsubscribe' to linuxppc-dev-request@lists.linuxppc.org ]]

Re: EGCS 1.1.1 FYI

[Daniel Jacobowitz]

On Mon, Dec 07, 1998 at 11:22:42PM +0100, Ralf Weidemann wrote:
> 
> >>>>> "David" == David Edelsohn <dje@watson.ibm.com> writes:
> >>>>> "Paul J Schinder" writes:
>     Paul> So are Gary Thomas' patches still necessary?
> 
>     David> 	The trampoline patch now has become a shared library
>     David> compatibility issue for glibc-1.99 support.  It only is
>     David> necessary until LinuxPPC cuts over to glibc-2.1 which will
>     David> require other binary incompatible changes at the same time.
> 
> Can someone clarify where the problem is exact with native
> egcs and libc-1.99 ? I use egcs-1.1.1 since the weekend and
> it seems to work fine. I had no trouble to build some shared
> lib stuff like in the latest jade or sgml-tools. This packages
> didn't compile with older version I tried (also Gary's egcs-1.1b),
> but that were generell compiler issues, I think.
> 
> But on the other side I also tried to run the SeaMonkeyBrowser
> (Mozilla) from recent cvs-sources. Building went fine, but the
> beast didn't run. I get only a simple "Aborted" message on the
> console and it seems to crash somewhere in the startup code.
> So maybe this is somehow related ?
> 
> Had someone luck to run a recent version of Mozilla at all on
> LinuxPPC ?

No.  There's a bug that at least one person (Anthony Tong) is working
on finding, something with the shared libraries, I think.  SeaMonkey
has its own set of problems, in addition - I recommend playing with
classic mozilla instead.

Dan

/--------------------------------\  /--------------------------------\
|       Daniel Jacobowitz        |__|     CMU, CS class of 2002      |
|   Debian GNU/Linux Developer    __   Part-Time Systems Programmer  |
|         dan@debian.org         |  |        drow@cs.cmu.edu         |
\--------------------------------/  \--------------------------------/

[[ This message was sent via the linuxppc-dev mailing list. Replies are ]]
[[ not forced back to the list, so be sure to  Cc linuxppc-dev  if your ]]
[[ reply is of general interest. To unsubscribe from linuxppc-dev, send ]]
[[ the message 'unsubscribe' to linuxppc-dev-request@lists.linuxppc.org ]]

Re: EGCS 1.1.1 FYI

[Elgin Lee]

>>>>> In <Pine.LNX.3.96.981208202419.4464D-100000@zero> 
>>>>>	Tom Vier <nester@nester.dyn.ml.org> wrote:

> anyone know if egcs-1.1.1-1a.ppc.rpm from pre-R5 is patched?

I don't know, but I doubt it.  You can check the SRPM to see.  IIRC,
LinuxPPC pre-R5 is based on an egcs-1.1.1-compiled glibc 2.1, which

    1. works fine with an unpatched egcs
    2. would not work with a patched egcs

The patches are only for compatibility with the currently deployed
shared libraries.

--Elgin

[[ This message was sent via the linuxppc-dev mailing list. Replies are ]]
[[ not forced back to the list, so be sure to  Cc linuxppc-dev  if your ]]
[[ reply is of general interest. To unsubscribe from linuxppc-dev, send ]]
[[ the message 'unsubscribe' to linuxppc-dev-request@lists.linuxppc.org ]]