3.3 PATCH: Handle mips gas 2.13 -mdebug switch

3.3 PATCH: Handle mips gas 2.13 -mdebug switch

[Rainer Orth]

As discussed in the thread starting at

	http://sources.redhat.com/ml/binutils/2002-06/msg00605.html

the introduction of the new -mdebug switch in gas 2.12.9x introduced a
considerable number of gcc testsuite failures.  I observed the same when
using gas 2.12.90 configured for mips-sgi-irix5.3 for my new
mips-sgi-irix6*o32 configuration.  The problem is that the assembler
requires the new -mdebug switch to correctly assemble input files with
stabs-in-mdebug or ecoff-in-mdebug debugging information.

The following patch fixes a large part of the problem:

* It detects if the assembler used supports the -mdebug switch at all.

* If so, it defines a new mdebug_asm_spec to handle the core of the
  problem:

** If the default debugging format is DWARF-2, only gcc invokations with
   -gstabs* or -gcoff* need to pass -mdebug on to gas.

** If the default debugging format is Stabs or ECOFF, -mdebug must be
   always passed unless -gdwarf-2* is specified.

  The problem is that the test for PREFERRED_DEBUGGING_TYPE cannot be
  performed at compile time for the same reasons leading to

	http://gcc.gnu.org/ml/gcc-patches/2002-07/msg00208.html

  Therefore, I needed a way to initialize parts of EXTRA_SPECS at compile
  time.  The new EXTRA_SPECS_INIT implemented and documented below does
  this.

This patch doesn't correctly handle ASM_DEBUG_SPEC yet.  It gets the
necessary -mdebug flag indirectly via ASM_SPEC, but the default
ASM_DEBUG_SPEC is wrong on systems that have ECOFF_DEBUGGING defined as
well:

With DWARF-2 the default, gcc -gcoff -c asm.s invokes gas with --gdwarf2
(which is clearly wrong) and -mdebug (correct).

With Stabs the default, --gstabs isn't passed without any -g switch (which
may be correct, I don't know for sure), but again -gcoff causes gas to be
invoked with --gstabs -mdebug.

This will have to wait for a follow-up patch, though.  One problem is that
a correct default for a mips ASM_DEBUG_SPEC cannot be defined in mips.h
since many target headers later redefine PREFERRED_DEBUGGING_TYPE.  This
either needs to go to gcc.c as well, or we need a new common header (to be
included after all the target headers) to correctly define ASM_DEBUG_SPEC
at this point.

Bootstrapped on mips-sgi-irix6.2o32 with gas 2.12.90 both with and without
--without-stabs (i.e. with either DWARF2_DEBUG or DBX_DEBUG), gas 2.12.1
and native as (which don't have -mdebug).

Ok for mainline?

	Rainer


Wed Jul 17 20:41:46 2002  Rainer Orth  <ro@TechFak.Uni-Bielefeld.DE>

	* config/mips/mips.h (MDEBUG_ASM_SPEC): Define.
	(ASM_SPEC): Use it.
	(EXTRA_SPECS): Add mdebug_asm_spec.
	(EXTRA_SPECS_INIT): Define.

	* gcc.c (struct spec_list_1): Remove const from ptr.
	(extra_specs_1): Remove const.
	(init_spec): Use EXTRA_SPECS_INIT if defined.
	* doc/tm.texi (EXTRA_SPECS, EXTRA_SPECS_INIT): Document.
	
	* configure.in: Test for gas -mdebug flag.
	* configure: Regenerate.
	* config.in: Likewise.
	
Index: gcc/gcc.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/gcc.c,v
retrieving revision 1.327
diff -u -p -r1.327 gcc.c
--- gcc/gcc.c	5 Jul 2002 12:33:52 -0000	1.327
+++ gcc/gcc.c	19 Jul 2002 17:03:43 -0000
@@ -1410,10 +1410,10 @@ static struct spec_list static_specs[] =
 struct spec_list_1
 {
   const char *const name;
-  const char *const ptr;
+  const char *ptr;
 };
 
-static const struct spec_list_1 extra_specs_1[] = { EXTRA_SPECS };
+static struct spec_list_1 extra_specs_1[] = { EXTRA_SPECS };
 static struct spec_list *extra_specs = (struct spec_list *) 0;
 #endif
 
@@ -1466,6 +1466,12 @@ init_spec ()
 
   if (verbose_flag)
     notice ("Using built-in specs.\n");
+
+#ifdef EXTRA_SPECS_INIT
+  for (i = ARRAY_SIZE (extra_specs_1) - 1; i >= 0; i--)
+    if (extra_specs_1[i].ptr == NULL)
+      EXTRA_SPECS_INIT (extra_specs_1[i].name, extra_specs_1[i].ptr);
+#endif
 
 #ifdef EXTRA_SPECS
   extra_specs = (struct spec_list *)
Index: gcc/config/mips/mips.h
===================================================================
RCS file: /cvs/gcc/gcc/gcc/config/mips/mips.h,v
retrieving revision 1.199
diff -u -p -r1.199 mips.h
--- gcc/config/mips/mips.h	17 Jul 2002 09:24:08 -0000	1.199
+++ gcc/config/mips/mips.h	19 Jul 2002 17:03:47 -0000
@@ -969,6 +969,17 @@ extern int mips_abi;
 #define SUBTARGET_ASM_SPEC ""
 #endif
 
+/* Beginning with gas 2.13, -mdebug must be passed to correctly handle COFF
+   and stabs debugging info.  */
+#ifdef HAVE_AS_MDEBUG_DEBUG_FLAG
+#define MDEBUG_ASM_SPEC					\
+	(PREFERRED_DEBUGGING_TYPE == DWARF2_DEBUG)	\
+	 ? "%{gcoff*|gstabs*:-mdebug}"			\
+	 : "%{!gdwarf-2*:-mdebug}")
+#else
+#define MDEBUG_ASM_SPEC ""
+#endif
+
 /* ASM_SPEC is the set of arguments to pass to the assembler.  */
 
 #undef ASM_SPEC
@@ -977,6 +988,7 @@ extern int mips_abi;
 %{mips16:%{!mno-mips16:-mips16}} %{mno-mips16:-no-mips16} \
 %(subtarget_asm_optimizing_spec) \
 %(subtarget_asm_debugging_spec) \
+%(mdebug_asm_spec) \
 %{membedded-pic} \
 %{mabi=32:-32}%{mabi=o32:-32}%{mabi=n32:-n32}%{mabi=64:-64}%{mabi=n64:-64} \
 %(target_asm_spec) \
@@ -1093,10 +1105,19 @@ extern int mips_abi;
   { "subtarget_mips_as_asm_spec", SUBTARGET_MIPS_AS_ASM_SPEC }, 	\
   { "subtarget_asm_optimizing_spec", SUBTARGET_ASM_OPTIMIZING_SPEC },	\
   { "subtarget_asm_debugging_spec", SUBTARGET_ASM_DEBUGGING_SPEC },	\
+  { "mdebug_asm_spec", NULL },						\
   { "subtarget_asm_spec", SUBTARGET_ASM_SPEC },				\
   { "endian_spec", ENDIAN_SPEC },					\
   SUBTARGET_EXTRA_SPECS
 
+/* mdebug_asm_spec needs to be initialized at run-time.  */
+
+#define EXTRA_SPECS_INIT(NAME, PTR)					\
+do {									\
+  if (strcmp (NAME, "mdebug_asm_spec") == 0)				\
+    PTR = MDEBUG_ASM_SPEC;						\
+} while (0);
+
 #ifndef SUBTARGET_EXTRA_SPECS
 #define SUBTARGET_EXTRA_SPECS
 #endif
Index: gcc/doc/tm.texi
===================================================================
RCS file: /cvs/gcc/gcc/gcc/doc/tm.texi,v
retrieving revision 1.145
diff -u -p -r1.145 tm.texi
--- gcc/doc/tm.texi	16 Jul 2002 02:16:45 -0000	1.145
+++ gcc/doc/tm.texi	19 Jul 2002 17:03:58 -0000
@@ -278,7 +278,10 @@ Define this macro to provide additional 
 
 The definition should be an initializer for an array of structures,
 containing a string constant, that defines the specification name, and a
-string constant that provides the specification.
+string constant that provides the specification.  If the specification
+needs to be initialized at run-time, use a @code{NULL} specification in
+@code{EXTRA_SPECS} to signal this and perform the real initialization in
+@code{EXTRA_SPECS_INIT} below.
 
 Do not define this macro if it does not need to do anything.
 
@@ -321,6 +324,14 @@ while the @file{config/rs6000/eabiaix.h}
 #undef CPP_SYSV_DEFAULT
 #define CPP_SYSV_DEFAULT "-D_CALL_AIX"
 @end smallexample
+
+@findex EXTRA_SPECS_INIT
+@item EXTRA_SPECS_INIT (@var{name}, @var{ptr})
+If any specification in @code{EXTRA_SPECS} needs to be initialized at
+run-time, @code{EXTRA_SPECS_INIT} is invoked for each @code{NULL}
+element in @code{EXTRA_SPECS} with @code{@var{name}} set to the
+specification name to initialize and should assign the proper
+specification to @code{@var{ptr}}.
 
 @findex LINK_LIBGCC_SPECIAL
 @item LINK_LIBGCC_SPECIAL

Re: 3.3 PATCH: Handle mips gas 2.13 -mdebug switch

[Eric Christopher]

> mips-sgi-irix6*o32 configuration.  The problem is that the assembler
> requires the new -mdebug switch to correctly assemble input files with
> stabs-in-mdebug or ecoff-in-mdebug debugging information.
> 
> The following patch fixes a large part of the problem:
> 
> * It detects if the assembler used supports the -mdebug switch at all.
> 
> * If so, it defines a new mdebug_asm_spec to handle the core of the
>   problem:
> 
> ** If the default debugging format is DWARF-2, only gcc invokations with
>    -gstabs* or -gcoff* need to pass -mdebug on to gas.
> 
> ** If the default debugging format is Stabs or ECOFF, -mdebug must be
>    always passed unless -gdwarf-2* is specified.
> 

Couple of things:

0) I was just working on this myself. If you require #1 it's much easier
;)

1) I wish there was a way to require certain versions of gas with
certain versions of gcc.

2) The configure part of this patch is missing :)

3) I'm trying to get rid of SDB_DEBUGGING, in a patch I'll be committing
shortly I'm moving everything but mips-ecoff, mips-sni, and irix5. I'm
not sure about the o32 irix6.

4) The patch in general looks ok to me after these comments. You'll need
more approval than me though.

-eric

-- 
I will not grease the monkey bars

Re: 3.3 PATCH: Handle mips gas 2.13 -mdebug switch

[Rainer Orth]

Eric Christopher writes:

> 0) I was just working on this myself. If you require #1 it's much easier
> ;)

Indeed: unfortunately, I'm currently hampered by a bug (yet to be found) in
gas 2.12.90 which causes many additional testsuite failures for my irix6o32
configuration with stabs, compared to 2.12.1.  So I need to support the
latter for some time.  Testresults for gas 2.12.1 with stabs are idential
to gas 2.12.90 with DWARF-2 :-).

> 1) I wish there was a way to require certain versions of gas with
> certain versions of gcc.

This is currently done both by feature testing and mentioning the
requirements in install.texi (which too few people read, unfortunately).

> 2) The configure part of this patch is missing :)

Right: included below.  This part was mostly cut-and-paste, but anyway
tested on the various assembler versions indicated in the patch submission.

> 3) I'm trying to get rid of SDB_DEBUGGING, in a patch I'll be committing
> shortly I'm moving everything but mips-ecoff, mips-sni, and irix5. I'm
> not sure about the o32 irix6.

I cannot really parse this: irix6o32 only inherits SDB_DEBUGGING from
iris5gas.h, so if you remove it there, irix6o32 will loose it as well.  I
certainly agree that this is useful: gdb support is probably even more
bit-rotten than stabs-in-mdebug.

> 4) The patch in general looks ok to me after these comments. You'll need
> more approval than me though.

Ok, I fear I'll need a global-write-privs maintainer for the gcc.c (and
configure.in) parts.  Any takers?  The configure.in part is below, the rest
of the patch at

	http://gcc.gnu.org/ml/gcc-patches/2002-07/msg01070.html

Thanks.

	Rainer


Index: gcc/configure.in
===================================================================
RCS file: /cvs/gcc/gcc/gcc/configure.in,v
retrieving revision 1.610
diff -u -p -r1.610 configure.in
--- gcc/configure.in	14 Jul 2002 01:59:13 -0000	1.610
+++ gcc/configure.in	19 Jul 2002 17:03:38 -0000
@@ -2154,6 +2156,35 @@ if test x"$gcc_cv_as_gstabs_flag" = xyes
 fi
 AC_MSG_RESULT($gcc_cv_as_gstabs_flag)
 
+case "$target" in
+  mips*-*-*)
+    AC_MSG_CHECKING(assembler -mdebug support)
+    gcc_cv_as_mdebug_flag=no
+    if test x$gcc_cv_gas_major_version != x -a x$gcc_cv_gas_minor_version != x;
+    then
+      if test "$gcc_cv_gas_major_version" -eq 2 \
+	  -a "$gcc_cv_gas_minor_version" -ge 13 \
+	  -o "$gcc_cv_gas_major_version" -gt 2 \
+         && grep 'obj_format = elf' ../gas/Makefile > /dev/null \
+	 && test x"$insn" != x ; then
+        gcc_cv_as_mdebug_flag="yes"
+      fi
+    elif test x$gcc_cv_as != x -a x"$insn" != x ; then
+	echo '' > conftest.s
+	# ??? This fails with non-gnu grep.
+	if $gcc_cv_as -mdebug -o conftest.o conftest.s > /dev/null 2>&1 ; then
+	  gcc_cv_as_mdebug_flag="yes"
+	fi
+	rm -f conftest.s conftest.o
+    fi
+    if test x"$gcc_cv_as_mdebug_flag" = xyes; then
+	AC_DEFINE(HAVE_AS_MDEBUG_DEBUG_FLAG, 1,
+    [Define if your assembler supports the -mdebug option.])
+    fi
+    AC_MSG_RESULT($gcc_cv_as_mdebug_flag)
+  ;;
+esac
+
 AC_MSG_CHECKING(linker PT_GNU_EH_FRAME support)
 gcc_cv_ld_eh_frame_hdr=no
 if test x$gcc_cv_gld_major_version != x -a x$gcc_cv_gld_minor_version != x; then

Re: 3.3 PATCH: Handle mips gas 2.13 -mdebug switch

[Daniel Jacobowitz]

On Fri, Jul 19, 2002 at 11:22:40PM +0200, Rainer Orth wrote:
> Eric Christopher writes:
> 
> > 0) I was just working on this myself. If you require #1 it's much easier
> > ;)
> 
> Indeed: unfortunately, I'm currently hampered by a bug (yet to be found) in
> gas 2.12.90 which causes many additional testsuite failures for my irix6o32
> configuration with stabs, compared to 2.12.1.  So I need to support the
> latter for some time.  Testresults for gas 2.12.1 with stabs are idential
> to gas 2.12.90 with DWARF-2 :-).
> 
> > 1) I wish there was a way to require certain versions of gas with
> > certain versions of gcc.
> 
> This is currently done both by feature testing and mentioning the
> requirements in install.texi (which too few people read, unfortunately).
> 
> > 2) The configure part of this patch is missing :)
> 
> Right: included below.  This part was mostly cut-and-paste, but anyway
> tested on the various assembler versions indicated in the patch submission.
> 
> > 3) I'm trying to get rid of SDB_DEBUGGING, in a patch I'll be committing
> > shortly I'm moving everything but mips-ecoff, mips-sni, and irix5. I'm
> > not sure about the o32 irix6.
> 
> I cannot really parse this: irix6o32 only inherits SDB_DEBUGGING from
> iris5gas.h, so if you remove it there, irix6o32 will loose it as well.  I
> certainly agree that this is useful: gdb support is probably even more
> bit-rotten than stabs-in-mdebug.

Absolutely.  It was only under pressure that I left the support in GAS
at all, as I'm sure Eric remembers :)  GDB behaves much, much better
with DWARF-2 or even elf stabs than it does with the mdebug reader.

-- 
Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer

Re: 3.3 PATCH: Handle mips gas 2.13 -mdebug switch

[Eric Christopher]

> Indeed: unfortunately, I'm currently hampered by a bug (yet to be found) in
> gas 2.12.90 which causes many additional testsuite failures for my irix6o32
> configuration with stabs, compared to 2.12.1.  So I need to support the
> latter for some time.  Testresults for gas 2.12.1 with stabs are idential
> to gas 2.12.90 with DWARF-2 :-).
> 

Hrm. If you can show what it is I'll help you track it down.

> > 1) I wish there was a way to require certain versions of gas with
> > certain versions of gcc.
> 
> This is currently done both by feature testing and mentioning the
> requirements in install.texi (which too few people read, unfortunately).
> 

Yeah, I'd really like to require it, but...

> > 3) I'm trying to get rid of SDB_DEBUGGING, in a patch I'll be committing
> > shortly I'm moving everything but mips-ecoff, mips-sni, and irix5. I'm
> > not sure about the o32 irix6.
> 
> I cannot really parse this: irix6o32 only inherits SDB_DEBUGGING from
> iris5gas.h, so if you remove it there, irix6o32 will loose it as well.  I
> certainly agree that this is useful: gdb support is probably even more
> bit-rotten than stabs-in-mdebug.
> 

Hrm. OK, well, I'll happily accept a patch to remove it from that. :)

> > 4) The patch in general looks ok to me after these comments. You'll need
> > more approval than me though.
> 
> Ok, I fear I'll need a global-write-privs maintainer for the gcc.c (and
> configure.in) parts.  Any takers?  The configure.in part is below, the rest
> of the patch at
> 
> 	http://gcc.gnu.org/ml/gcc-patches/2002-07/msg01070.html
> 

You will need someone else to approve, I don't see a problem with it
though.

-eric

-- 
I will not grease the monkey bars

Re: 3.3 PATCH: Handle mips gas 2.13 -mdebug switch

[Eric Christopher]

> Absolutely.  It was only under pressure that I left the support in GAS
> at all, as I'm sure Eric remembers :)  GDB behaves much, much better
> with DWARF-2 or even elf stabs than it does with the mdebug reader.
> 

Yup. I'd like to remove all of that, but I (or someone) will need to
look at the supported targets and make sure that all of them can be
moved over.

-eric

-- 
I will not grease the monkey bars

Re: 3.3 PATCH: Handle mips gas 2.13 -mdebug switch

[Eric Christopher]

> Absolutely.  It was only under pressure that I left the support in GAS
> at all, as I'm sure Eric remembers :)  GDB behaves much, much better
> with DWARF-2 or even elf stabs than it does with the mdebug reader.

Just as a note, here's what I'd _prefer_ to be able to do with all of
the elf targets (at least).

/* ??? Move all SDB stuff into separate header file.  */
#undef  SDB_DEBUGGING_INFO

#define DBX_DEBUGGING_INFO
#define DWARF2_DEBUGGING_INFO

#undef  PREFERRED_DEBUGGING_TYPE
#define PREFERRED_DEBUGGING_TYPE DWARF2_DEBUG

#undef  SUBTARGET_ASM_DEBUGGING_SPEC
#define SUBTARGET_ASM_DEBUGGING_SPEC "-g0 \
%{!gstabs*:-no-mdebug}%{gstabs*:-mdebug}"

Which is what I have in my current working sources. I'll raise the call
for objections right now :)

-eric

-- 
I will not grease the monkey bars

Re: 3.3 PATCH: Handle mips gas 2.13 -mdebug switch

[Daniel Jacobowitz]

On Fri, Jul 19, 2002 at 02:40:31PM -0700, Eric Christopher wrote:
> 
> > Absolutely.  It was only under pressure that I left the support in GAS
> > at all, as I'm sure Eric remembers :)  GDB behaves much, much better
> > with DWARF-2 or even elf stabs than it does with the mdebug reader.
> 
> Just as a note, here's what I'd _prefer_ to be able to do with all of
> the elf targets (at least).
> 
> /* ??? Move all SDB stuff into separate header file.  */
> #undef  SDB_DEBUGGING_INFO
> 
> #define DBX_DEBUGGING_INFO
> #define DWARF2_DEBUGGING_INFO
> 
> #undef  PREFERRED_DEBUGGING_TYPE
> #define PREFERRED_DEBUGGING_TYPE DWARF2_DEBUG
> 
> #undef  SUBTARGET_ASM_DEBUGGING_SPEC
> #define SUBTARGET_ASM_DEBUGGING_SPEC "-g0 \
> %{!gstabs*:-no-mdebug}%{gstabs*:-mdebug}"
> 
> Which is what I have in my current working sources. I'll raise the call
> for objections right now :)

Eh?  No!  I went to too much trouble getting MIPS/Linux moved away from
-mdebug.  And -no-mdebug doesn't mean terribly much if you aren't
generating stabs data, anyway.

-- 
Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer

Re: 3.3 PATCH: Handle mips gas 2.13 -mdebug switch

[Richard Henderson]

On Fri, Jul 19, 2002 at 09:16:41PM +0200, Rainer Orth wrote:
> 	* gcc.c (struct spec_list_1): Remove const from ptr.
> 	(extra_specs_1): Remove const.
> 	(init_spec): Use EXTRA_SPECS_INIT if defined.
> 	* doc/tm.texi (EXTRA_SPECS, EXTRA_SPECS_INIT): Document.

I am unthrilled by this.  Can we work around this stuff
by defining some mips-local tag when preferred debugging
is dwarf2, and then just using ifdefs instead of the
supposed-to-be-compile-time eval?


r~

Re: 3.3 PATCH: Handle mips gas 2.13 -mdebug switch

[echristo]

> Eh?  No!  I went to too much trouble getting MIPS/Linux moved away from
> -mdebug.  And -no-mdebug doesn't mean terribly much if you aren't
> generating stabs data, anyway.

OK. Barring linux.

The real question is if anyone sees a need for SDB debugging anymore...

-eric

Re: 3.3 PATCH: Handle mips gas 2.13 -mdebug switch

[Daniel Jacobowitz]

On Sat, Jul 20, 2002 at 03:41:19PM -0700, echristo@potter.sfbay.redhat.com wrote:
> > Eh?  No!  I went to too much trouble getting MIPS/Linux moved away from
> > -mdebug.  And -no-mdebug doesn't mean terribly much if you aren't
> > generating stabs data, anyway.
> 
> OK. Barring linux.
> 
> The real question is if anyone sees a need for SDB debugging anymore...

I don't see why you should default any ELF target to -mdebug; I think
that only COFF targets and people with very particular needs should be
doing this.  This could depend on what SGI's debuggers like, I
suppose...

-- 
Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer

Re: 3.3 PATCH: Handle mips gas 2.13 -mdebug switch

[Eric Christopher]

> I don't see why you should default any ELF target to -mdebug; I think
> that only COFF targets and people with very particular needs should be
> doing this.  This could depend on what SGI's debuggers like, I
> suppose...

IRIX 6 uses dwarf2, but the idea was mainly for backwards compatability.
Though, I don't suppose it matters in the long run as we move to dwarf2.
Anyone else have any thoughts?

-eric

-- 
I will not grease the monkey bars