egcs-1.0.2 stuff

egcs-1.0.2 stuff

[Jeffrey A Law]

I've started putting together some patches for the egcs-1.0.2 release. 
Like egcs-1.0.1 we want to fix critical bugs.

We don't have any kind of a schedule yet, but I figured it was worth
letting folks go ahead and poke at the changes.

Right now these are only available on the cvs server; I'll bundle
them up for ftp after I take care of a few more critical fixes.

To check out the release branch, go into an empty directory and
issue the command:

cvs -z 9 co -regcs_1_00_branch egcs

I don't recommend trying to use a single checked out tree to move
back and forth between the mainline sources and the release branch.
Instead keep a tree for the mainline sources and a tree for the branch.

The changes I've already installed on the release branch:

  * Fixes for some of the memory explosion problems:

    * Lazy allocation of label maps.

    * Don't call save_for_inline_copy if we just want -Wreturn-type.

  * libstdc++ should be up to date (read link compatable) with
  libstdc++ 2.8.0.

  * Fix to avoid duplicate symbol problems on systems that don't
  support weak symbols.

  * Patch for loop optimizer bug found on rs6000/ppc port.

  * Patch to fix stdarg bug for irix6.

  * Fix compiler crash on ommitted bound in fortran CASE statement.

  * Add missing entries to g77 lang-options.

  * Fix problems with -fpedantic in the g77 compiler.

  * Fix problem with static linking on sco5.

  * Fix problem bootstrapping with native "cc" on sco5.

  * Patch for compiler abort on very long thunk names.


Stuff still to take care of:

  * I thought there was a 3rd patch for the memory explosion
  problems.  Is that not the case?

  * 386 backend problem with literals vs PIC (or whatever it was that
  rth recently fixed :-)

  * Some problems with glibc2.1.

  * Collect2 fixes to deal with various aix library problems.

  * Toon also reported a nasty g77 problem; I don't know if it
  was fixed by one of the g77 changes I brought into the branch
  or not.

  * Other critical bugfixes that I've missed (such as...?)

Re: egcs-1.0.2 stuff

[Toon Moene]

>  * Fix compiler crash on ommitted bound in fortran CASE statement.

>  * Add missing entries to g77 lang-options.

>  * Fix problems with -fpedantic in the g77 compiler.

These are the only g77-related patches that you mention.  Are you  
sure my fix to the backspace problem on Alpha's is in ?  See  
http://www.cygnus.com/ml/egcs-bugs/1998-Jan/0339.html .  This one is  
sanctioned by David M Gay, the f2c maintainer.

> Stuff still to take care of:

>  * Toon also reported a nasty g77 problem; I don't know if it
>  was fixed by one of the g77 changes I brought into the branch
>  or not.

I haven't heard from anyone that s/he fixed it.  rth promised to  
look into it, but it might have overflowed his work-stack ;-)

For the problem description and a simple example, see  
http://www.cygnus.com/ml/egcs-bugs/1998-Feb/0015.html .

Cheers,
Toon.

Re: egcs-1.0.2 stuff

[Richard Hadsell]

Jeffrey A Law wrote:
> 
>   * Other critical bugfixes that I've missed (such as...?)

This is not critical, but it is so simple, it will be annoying to not
see it get into your compiler:

[Re: egcs-1.0.1 can't find headers in -I./]

Richard Hadsell writes:
 > It seems that egcs-1.0.1 does not recognize -I./ as the current
 > directory.  This was true of egcs-1.0 as well.  -I. works fine.  It
 > would be nice to get -I./ working, because we have some make rules
that
 > generate the include path automatically from discovered header files,
 > and they give "./" as the directory for any header file that is named
 > without a path.
 > 
 > I think that this test case is telling us that it misinterprets
"-I./"
 > as "-I/":

Here is a fix for the above problem.  In cccp.c the diff of
cccp.c.orig  and cccp.c (fixed file) is :

9815c9815
<     if (dir->fname[len - 1] != '/') {
---
>     if (len && dir->fname[len - 1] != '/') {

-- 
Dick Hadsell			914-381-8400  Fax: 914-381-9790
Reply-to:			hadsell@blueskystudios.com
Blue Sky | VIFX			http://www.blueskystudios.com
1 South Road, Harrison, NY 10528

Re: egcs-1.0.2 stuff

[Joern Rennecke]

> To check out the release branch, go into an empty directory and
> issue the command:
> 
> cvs -z 9 co -regcs_1_00_branch egcs
> 
> I don't recommend trying to use a single checked out tree to move
> back and forth between the mainline sources and the release branch.
> Instead keep a tree for the mainline sources and a tree for the branch.

Hmmm, but AFAIK it's all right to copy an entire tree and then update
it to another version.  This can safe some time if you are on a slow link.

>   * Other critical bugfixes that I've missed (such as...?)

Depends on your definition of critical... this one fixes a problem
with undefined symbols when using -gdwarf for some kind of source.

Mon Jan 26 18:11:30 1998  J"orn Rennecke <amylaar@cygnus.co.uk>

        * c-decl.c (grokdeclarator): Get parameter tags from
        last_function_parm_tags.
        * dwarfout.c (output_formal_types): Set TREE_ASM_WRITTEN before
        traversing the parameter types.
        (output_type): No early exit for FUNCTION_TYPE / METHOD_TYPE context.

Re: egcs-1.0.2 stuff

[Jeffrey A Law]

  In message < 34DF7543.60E3@blueskystudios.com >you write:
  > Here is a fix for the above problem.  In cccp.c the diff of
  > cccp.c.orig  and cccp.c (fixed file) is :
  > 
  > 9815c9815
  > <     if (dir->fname[len - 1] != '/') {
  > ---
  > >     if (len && dir->fname[len - 1] != '/') {
Please send this as a context diff -- using line numbers like this isn't
reliable.  I'd also be interested to know what version of the compiler
you made that diff from -- I think all explicit uses of '/' have been
changed to DIR_SEPARATOR. 

(use diff -c to create a context diffs).

jeff

Re: egcs-1.0.2 stuff

[Jeffrey A Law]

  In message < 199802100017.AAA20573@phal.cygnus.co.uk >you write:
  > Depends on your definition of critical... this one fixes a problem
  > with undefined symbols when using -gdwarf for some kind of source.
  > 
  > Mon Jan 26 18:11:30 1998  J"orn Rennecke <amylaar@cygnus.co.uk>
  > 
  >         * c-decl.c (grokdeclarator): Get parameter tags from
  >         last_function_parm_tags.
  >         * dwarfout.c (output_formal_types): Set TREE_ASM_WRITTEN before
  >         traversing the parameter types.
  >         (output_type): No early exit for FUNCTION_TYPE / METHOD_TYPE context.
Added.
jeff

Re: egcs-1.0.2 stuff

[Jeffrey A Law]

  In message < 199802100017.AAA20573@phal.cygnus.co.uk >you write:
  > > I don't recommend trying to use a single checked out tree to move
  > > back and forth between the mainline sources and the release branch.
  > > Instead keep a tree for the mainline sources and a tree for the branch.
  > 
  > Hmmm, but AFAIK it's all right to copy an entire tree and then update
  > it to another version.  This can safe some time if you are on a slow link.
The reason I don't recommend this is because of some directory reorganization
that has occured.

While I certainly agree it *can* be done, my recommendation is to check
a separate tree from the repository.

Jeff
 

Re: egcs-1.0.2 stuff

[Richard Henderson]

On Mon, Feb 09, 1998 at 07:20:19PM +0100, Toon Moene wrote:
> >  * Toon also reported a nasty g77 problem; I don't know if it
> >  was fixed by one of the g77 changes I brought into the branch
> >  or not.
> 
> I haven't heard from anyone that s/he fixed it.  rth promised to  
> look into it, but it might have overflowed his work-stack ;-)

Ack, it did. ;-)   Thanks for reminding me.


r~

Re: egcs-1.0.2 stuff

[Richard Henderson]

On Mon, Feb 09, 1998 at 07:20:19PM +0100, Toon Moene wrote:
> I haven't heard from anyone that s/he fixed it.  rth promised to  
> look into it, but it might have overflowed his work-stack ;-)

At issue is that a negative SImode lower array bound is being
converted to UDImode, makes for a program that doesn't work
quite so well as one might hope.

How does this look?  


r~

Re: egcs-1.0.2 stuff

[Jeffrey A Law]

  In message < 9802091820.AA17476@moene.indiv.nluug.nl >you write:
  > >  * Fix compiler crash on ommitted bound in fortran CASE statement.
  > 
  > >  * Add missing entries to g77 lang-options.
  > 
  > >  * Fix problems with -fpedantic in the g77 compiler.
  > 
  > These are the only g77-related patches that you mention.  Are you  
  > sure my fix to the backspace problem on Alpha's is in ?  See  
  > http://www.cygnus.com/ml/egcs-bugs/1998-Jan/0339.html .  This one is  
  > sanctioned by David M Gay, the f2c maintainer.
That list was complete.  So it wasn't included.

However, I just added it.


  > > Stuff still to take care of:
  > 
  > >  * Toon also reported a nasty g77 problem; I don't know if it
  > >  was fixed by one of the g77 changes I brought into the branch
  > >  or not.
  > 
  > I haven't heard from anyone that s/he fixed it.  rth promised to  
  > look into it, but it might have overflowed his work-stack ;-)
  > 
  > For the problem description and a simple example, see  
  > http://www.cygnus.com/ml/egcs-bugs/1998-Feb/0015.html .
We'll make a decision on this one when we've got a fix.

jeff

Re: egcs-1.0.2 stuff

[H.J. Lu]

Hi, Jeff,

Could you please put my libio patch in snapshot to 1.0.2? Otherwise
glibc 2.1 won't work. Also Andreas Schwab's libio patch should
be in 1.0.2.

BTW, we still need to fix STL for glibc 2.

Thanks.

H.J.

Re: egcs-1.0.2 stuff

[Jeffrey A Law]

  In message < m0y2IHX-0004ecC@ocean.lucon.org >you write:
  > Hi, Jeff,
  > 
  > Could you please put my libio patch in snapshot to 1.0.2? Otherwise
  > glibc 2.1 won't work. Also Andreas Schwab's libio patch should
  > be in 1.0.2.
What patch are you referring to.  Please provide a ChangeLog entry or some
other means by which to uniquely identify the patch.

  > BTW, we still need to fix STL for glibc 2.
Yes.  I know.  My understanding is we're waiting to hear back from external
folks on this patch.

jeff

Re: egcs-1.0.2 stuff

[Richard Hadsell]

[egcs-1.0.1 can't find headers in -I./]

Jeffrey A Law wrote:
> 
> Please send this as a context diff -- using line numbers like this isn't
> reliable.  I'd also be interested to know what version of the compiler
> you made that diff from -- I think all explicit uses of '/' have been
> changed to DIR_SEPARATOR.

Here is the context diff:

7% diff -c cccp.c.orig cccp.c
*** cccp.c.orig Wed Oct  1 02:22:46 1997
--- cccp.c      Fri Jan 30 15:34:47 1998
***************
*** 9812,9818 ****
      len = simplify_filename (dir->fname);
  
      /* Convert directory name to a prefix.  */
!     if (dir->fname[len - 1] != '/') {
        if (len == 1 && dir->fname[len - 1] == '.')
        len = 0;
        else
--- 9812,9818 ----
      len = simplify_filename (dir->fname);
  
      /* Convert directory name to a prefix.  */
!     if (len && dir->fname[len - 1] != '/') {
        if (len == 1 && dir->fname[len - 1] == '.')
        len = 0;
        else


And here is the compiler output from the original message:

39% ECC -v -I./ -c egcsbug2.cc
Reading specs from
/usr/local/lib/gcc-lib/alphaev5-unknown-linux-gnu/egcs-2.90.23/specs
gcc version egcs-2.90.23 980102 (egcs-1.0.1 release)
 /usr/local/lib/gcc-lib/alphaev5-unknown-linux-gnu/egcs-2.90.23/cpp
-lang-c++ -v -I./ -undef -D__GNUC__=2 -D__GNUG__=2 -D__cplusplus
-D__GNUC_MINOR__=90 -D__alpha -D__alpha__ -D__linux__ -D__linux
-D_LONGLONG -Dlinux -Dunix -D__ELF__ -D__alpha -D__alpha__ -D__linux__
-D__linux -D_LONGLONG -D__linux__ -D__unix__ -D__ELF__ -D__linux
-D__unix -Asystem(linux) -Acpu(alpha) -Amachine(alpha) -D__EXCEPTIONS
-D__LANGUAGE_C__ -D__LANGUAGE_C -DLANGUAGE_C -D__LANGUAGE_C_PLUS_PLUS__
-D__LANGUAGE_C_PLUS_PLUS -D__cplusplus egcsbug2.cc /tmp/cca30792.ii
GNU CPP version egcs-2.90.23 980102 (egcs-1.0.1 release) (Alpha
Linux/ELF)
#include "..." search starts here:
#include <...> search starts here:
 /
 /usr/local/include/g++
 /usr/local/include
 /usr/local/alphaev5-unknown-linux-gnu/include
 /usr/local/lib/gcc-lib/alphaev5-unknown-linux-gnu/egcs-2.90.23/include
 /usr/include
End of search list.
egcsbug2.cc:2: egcsbug2.h: No such file or directory
loki1 studio++/tests 40% ECC -v -I. -c egcsbug2.cc
Reading specs from
/usr/local/lib/gcc-lib/alphaev5-unknown-linux-gnu/egcs-2.90.23/specs
gcc version egcs-2.90.23 980102 (egcs-1.0.1 release)
 /usr/local/lib/gcc-lib/alphaev5-unknown-linux-gnu/egcs-2.90.23/cpp
-lang-c++ -v -I. -undef -D__GNUC__=2 -D__GNUG__=2 -D__cplusplus
-D__GNUC_MINOR__=90 -D__alpha -D__alpha__ -D__linux__ -D__linux
-D_LONGLONG -Dlinux -Dunix -D__ELF__ -D__alpha -D__alpha__ -D__linux__
-D__linux -D_LONGLONG -D__linux__ -D__unix__ -D__ELF__ -D__linux
-D__unix -Asystem(linux) -Acpu(alpha) -Amachine(alpha) -D__EXCEPTIONS
-D__LANGUAGE_C__ -D__LANGUAGE_C -DLANGUAGE_C -D__LANGUAGE_C_PLUS_PLUS__
-D__LANGUAGE_C_PLUS_PLUS -D__cplusplus egcsbug2.cc /tmp/cca30794.ii
GNU CPP version egcs-2.90.23 980102 (egcs-1.0.1 release) (Alpha
Linux/ELF)
#include "..." search starts here:
#include <...> search starts here:
 .
 /usr/local/include/g++
 /usr/local/include
 /usr/local/alphaev5-unknown-linux-gnu/include
 /usr/local/lib/gcc-lib/alphaev5-unknown-linux-gnu/egcs-2.90.23/include
 /usr/include
End of search list.
 /usr/local/lib/gcc-lib/alphaev5-unknown-linux-gnu/egcs-2.90.23/cc1plus
/tmp/cca30794.ii -quiet -dumpbase egcsbug2.cc -version -o
/tmp/cca30794.s
GNU C++ version egcs-2.90.23 980102 (egcs-1.0.1 release)
(alphaev5-unknown-linux-gnu) compiled by GNU C version egcs-2.90.23
980102 (egcs-1.0.1 release).
 as -nocpp -o egcsbug2.o /tmp/cca30794.s

-- 
Dick Hadsell			914-381-8400  Fax: 914-381-9790
Reply-to:			hadsell@blueskystudios.com
Blue Sky | VIFX			http://www.blueskystudios.com
1 South Road, Harrison, NY 10528

Re: egcs-1.0.2 stuff

[Toon Moene]

On Mon, Feb 09, 1998 at 07:20:19PM +0100, Toon Moene wrote:
> I haven't heard from anyone that s/he fixed it.  rth promised to  
> look into it, but it might have overflowed his work-stack ;-)

At issue is that a negative SImode lower array bound is being
converted to UDImode, makes for a program that doesn't work
quite so well as one might hope.

How does this look?

[ Patch deleted for brevity ]

Yes, that does it.  It also doesn't seem to have a negative impact  
on the generated code.

BTW, perhaps the comment in stor-layout.c can be repeated in  
f/com.c ?  Otherwise the link (that's now there via the ChangeLog  
entry) might get lost over time.

Thanks,
Toon.

Re: egcs-1.0.2 stuff

[H.J. Lu]

> 
> 
>   In message < m0y2IHX-0004ecC@ocean.lucon.org >you write:
>   > Hi, Jeff,
>   > 
>   > Could you please put my libio patch in snapshot to 1.0.2? Otherwise
>   > glibc 2.1 won't work. Also Andreas Schwab's libio patch should
>   > be in 1.0.2.
> What patch are you referring to.  Please provide a ChangeLog entry or some
> other means by which to uniquely identify the patch.
> 

Here they are. BTW, I didn't check the EOF flag in isgetline.cc
because it might be set during the previous read. It can happen
on socket, pipe and tty.

Thanks.

H.J.
-----
Fri Jan 30 07:47:57 1998  H.J. Lu  (hjl@gnu.org)

	* isgetline.cc (istream::getline, istream::get): Call
	streambuf::in_avail () to check EOF when _IO_getline ()
	returns 0.

1998-01-20  Andreas Schwab  (schwab@issan.informatik.uni-dortmund.de)

	* iostream.cc (istream::operator>>(long double&))
	[!_G_HAVE_LONG_DOUBLE_IO]: Scan value into separate variable, in
	case long double is bigger than double.
	(ostream::operator<<(double)) [_G_HAVE_PRINTF_FP]: Fix order of
	initializers of struct printf_info to match declaration order,
	to work around g++ bug.
	(ostream::operator<<(long double)) [_G_HAVE_PRINTF_FP]: Likewise.

	* gen-params: Add missing quotes.  Avoid useless use of command
	substitution.

Sun Feb  1 13:29:47 1998  H.J. Lu  (hjl@gnu.org)

	* filebuf.cc (filebuf::open): Call _IO_file_open if
	_G_HAVE_IO_FILE_OPEN is 1.

	* libio.h (_IO_fpos64_t, _IO_off64_t): Defined if
	_G_IO_IO_FILE_VERSION == 0x20001.

	* libioP.h (_IO_file_open): New declaration.

	* libio.h (_IO_FILE, _IO_stdin_, _IO_stdout_, _IO_stderr_,
	_IO_seekoff, _IO_seekpos): Add support for libio in glibc 2.1.
	* libioP.h (_IO_seekoff_t, _IO_seekpos_t, _IO_seek_t,
	_IO_seekoff, _IO_seekpos, _IO_default_seekoff,
	_IO_default_seekpos, _IO_default_seek, _IO_file_seekoff,
	_IO_file_seek, _IO_str_seekoff, _IO_pos_BAD, _IO_pos_as_off,
	_IO_pos_0): Ditto.
	* streambuf.h (streamoff, streampos): Ditto.

	* gen-params (__extension__): Use only if gcc version >= 2.8.

Sun Feb  1 13:08:18 1998  Krister Walfridsson (cato@df.lth.se)

	* dbz/dbz.c (putconf): Handle systems which use "long long" as type
	for "off_t".
	* dbz/dbzmain.c (mkfiles): Likewise.

Tue Jan 27 10:35:22 1998  H.J. Lu  (hjl@gnu.org)

	* configure.in (topsrcdir): New.
	(CHECK_SUBDIRS, configdirs): Check ${topsrcdir}/gcc instead.
	(config-ml.in): Use ${topsrcdir}/config-ml.in.

	* tests/configure.in (topsrcdir): New.
	(CHECK): Check ${topsrcdir}/gcc instead.

Tue Jan 13 21:30:33 1998  H.J. Lu  (hjl@gnu.org)

	* configure.in (CHECK_SUBDIRS): Set to testsuite only if
	${srcdir}/../gcc exists.
	(configdirs): Include testsuite only if ${srcdir}/../gcc exists.

	* tests/Makefile.in (check): Depend on $(CHECK).

	* tests/configure.in (CHECK): Set to "check-iostream
	check-stdio" if ${srcdir}/../../gcc doesn't exists.

Re: egcs-1.0.2 stuff

[Jeffrey A Law]

  In message < 9802101822.AA19464@moene.indiv.nluug.nl >you write:
  > BTW, perhaps the comment in stor-layout.c can be repeated in  
  > f/com.c ?  Otherwise the link (that's now there via the ChangeLog  
  > entry) might get lost over time.
This is a good idea.  Richard, can you take care of it?

jeff

Re: egcs-1.0.2 stuff

[Jeffrey A Law]

  In message < 19980210032616.07788@dot.cygnus.com >you write:
  > At issue is that a negative SImode lower array bound is being
  > converted to UDImode, makes for a program that doesn't work
  > quite so well as one might hope.
  > 
  > How does this look?  
  > 
  > 
  > r~
  > 
  > --pf9I7BMVVzbSWLtt
  > Content-Type: text/plain; charset=us-ascii
  > Content-Disposition: attachment; filename=d-g77-arrsize
  > 
  > Tue Feb 10 03:13:48 1998  Richard Henderson  <rth@cygnus.com>
  > 
  > 	* stor-layout.c (layout_type): Do upper - lower in the native type,
  > 	so as to properly handle negative indices.
  > 
  > 	* f/com.c (ffecom_ptr_to_expr) [FFEBLD_opARRAYREF]: Likewise.
Looks good to me.  I installed it on the mainline & release branches.

jeff

Re: egcs-1.0.2 stuff

[Jeffrey A Law]

  In message < 34E07F82.58E@blueskystudios.com >you write:
  > [egcs-1.0.1 can't find headers in -I./]
  > 
  > Jeffrey A Law wrote:
  > > 
  > > Please send this as a context diff -- using line numbers like this isn't
  > > reliable.  I'd also be interested to know what version of the compiler
  > > you made that diff from -- I think all explicit uses of '/' have been
  > > changed to DIR_SEPARATOR.
Thanks.  I've installed this on both the release branch and the mainline
sources.

jeff

Re: egcs-1.0.2 stuff

[Krister Walfridsson]

I'd like to see the following changes from the development branch
go into 1.0.2:

Make it possible to use egcs on NetBSD's a.out ports:


   Tue Jan 13 21:55:06 1998  Jeffrey A Law  (law@cygnus.com)

           * arm/netbsd.h (DWARF2_UNWIND_INFO): Define as zero for now.
           * i386/netbsd.h, m68k/netbsd.h, ns32k/netbsd.h: Likewise.
           * sparc/netbsd.h, vax/netbsd.h: Likewise.


Prevent random lossage with libio (prototype for lseek since it returns
off_t which is long long on some NetBSD ports)


   Sun Feb  8 20:58:42 1998  Krister Walfridsson (cato@df.lth.se)
   
           * fileops.c: #include <unistd.h>.
           * ioprims.c: Likewise.


This isn't really critical, but I want it in 1.0.2 since it breaks
'make check' in a way that look really bad. That gives the impression
that egcs isn't stable enough to be worth its disk space...


   Sun Feb  1 13:08:18 1998  Krister Walfridsson (cato@df.lth.se)

           * dbz/dbz.c (putconf): Handle systems which use "long long" as type
           for "off_t".
           * dbz/dbzmain.c (mkfiles): Likewise.


   /Krister

Re: egcs-1.0.2 stuff

[Jeffrey A Law]

  In message < Pine.LNX.3.96.980214182235.31326B-100000@bartlet.df.lth.se >you wr
ite:
  > 
  > I'd like to see the following changes from the development branch
  > go into 1.0.2:
  > 
  >    Tue Jan 13 21:55:06 1998  Jeffrey A Law  (law@cygnus.com)
  > 
  >            * arm/netbsd.h (DWARF2_UNWIND_INFO): Define as zero for now.
  >            * i386/netbsd.h, m68k/netbsd.h, ns32k/netbsd.h: Likewise.
  >            * sparc/netbsd.h, vax/netbsd.h: Likewise.
Done.



  > Prevent random lossage with libio (prototype for lseek since it returns
  > off_t which is long long on some NetBSD ports)
  > 
  > 
  >    Sun Feb  8 20:58:42 1998  Krister Walfridsson (cato@df.lth.se)
  >    
  >            * fileops.c: #include <unistd.h>.
  >            * ioprims.c: Likewise.
Done.

  > This isn't really critical, but I want it in 1.0.2 since it breaks
  > 'make check' in a way that look really bad. That gives the impression
  > that egcs isn't stable enough to be worth its disk space...
  > 
  > 
  >    Sun Feb  1 13:08:18 1998  Krister Walfridsson (cato@df.lth.se)
  > 
  >            * dbz/dbz.c (putconf): Handle systems which use "long long" as type
  >            for "off_t".
  >            * dbz/dbzmain.c (mkfiles): Likewise.
Done.
jeff

Re: egcs-1.0.2 stuff

[Jeffrey A Law]

  In message < orvhumn4ff.fsf@araguaia.dcc.unicamp.br >you write:
  > Mike Stump writes:
  > 
  > >         * rtlanal.c (dead_or_set_regno_p): Ignore REG_DEAD notes after
  > >         reload completes.
  > >         * genattrtab.c (reload_completed): Define.
  >  
  > Will this fix a bug I've just found in gcc-2.8.1-980206 that causes an
  > internal compiler error while compiling libg++'s RNG.cc on
  > sparc-sun-sunos4.1.3?
Unlikely.

jeff

Re: egcs-1.0.2 stuff

[Alexandre Oliva]

Mike Stump writes:

>         * rtlanal.c (dead_or_set_regno_p): Ignore REG_DEAD notes after
>         reload completes.
>         * genattrtab.c (reload_completed): Define.
 
Will this fix a bug I've just found in gcc-2.8.1-980206 that causes an
internal compiler error while compiling libg++'s RNG.cc on
sparc-sun-sunos4.1.3?

It aborts in schedule_block, just as it finds dead_notes != 0:

4001      /* There should be no REG_DEAD notes leftover at the end.
4002         In practice, this can occur as the result of bugs in flow, combine.c,
4003         and/or sched.c.  The values of the REG_DEAD notes remaining are
4004         meaningless, because dead_notes is just used as a free list.  */
4005    #if 1
4006      if (dead_notes != 0)
(gdb) list 
4007        abort ();
4008    #endif

-- 
Alexandre Oliva
mailto:oliva@dcc.unicamp.br mailto:aoliva@acm.org
http://www.dcc.unicamp.br/~oliva
Universidade Estadual de Campinas, SP, Brasil

Re: egcs-1.0.2 stuff

[Jeffrey A Law]

  In message < 199802092133.NAA13204@kankakee.wrs.com >you write:
  > 1998-01-28  Mike Stump  <mrs@wrs.com>
  >  
  >         * rtlanal.c (dead_or_set_regno_p): Ignore REG_DEAD notes after
  >         reload completes.
  >         * genattrtab.c (reload_completed): Define.
  >  
  > Seems like a win for stability and robustness v bugginess, though it
  > should be very minor win (many not even important enough for the
  > release).  Kenner thought it was fairly safe also.
Added.

jeff

Re: egcs-1.0.2 stuff

[Mike Stump]

> To: egcs@cygnus.com
> Date: Mon, 09 Feb 1998 01:48:55 -0700
> From: Jeffrey A Law <law@cygnus.com>

> I've started putting together some patches for the egcs-1.0.2 release. 
> Like egcs-1.0.1 we want to fix critical bugs.

> We don't have any kind of a schedule yet, but I figured it was worth
> letting folks go ahead and poke at the changes.

> Right now these are only available on the cvs server; I'll bundle
> them up for ftp after I take care of a few more critical fixes.

1998-01-28  Mike Stump  <mrs@wrs.com>
 
        * rtlanal.c (dead_or_set_regno_p): Ignore REG_DEAD notes after
        reload completes.
        * genattrtab.c (reload_completed): Define.
 
Seems like a win for stability and robustness v bugginess, though it
should be very minor win (many not even important enough for the
release).  Kenner thought it was fairly safe also.