No Subject

No Subject

[Jeffrey A Law]

	* Makefile.in (boostrap2-lean, bootstrap3-lean,
	bootstrap4-lean): new targets
	* gcc/Makefile.in (boostrap2-lean, bootstrap3-lean,
	bootstrap4-lean): ditto
I've installed this patch.  Thanks!

I want to take a little more time to think about the dejagnu
patches.


jeff

Re: none

[Alexandre Oliva]

[-- Attachment #1: Type: text/plain, Size: 702 bytes --]

Jeffrey A Law <law@hurl.cygnus.com> writes:

> 	* Makefile.in (boostrap2-lean, bootstrap3-lean,
> 	bootstrap4-lean): new targets
> 	* gcc/Makefile.in (boostrap2-lean, bootstrap3-lean,
> 	bootstrap4-lean): ditto
> I've installed this patch.  Thanks!

How about the attached one too?  I wrote it so long ago that I had
already forgot about it... :-(

> I want to take a little more time to think about the dejagnu
> patches.

Just have in mind that it won't change absolutely anything unless
CHECK_MULTI is set to something different from "n" or "no"...

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

[-- Attachment #2: testsuite-tmp.diff --]
[-- Type: text/x-diff, Size: 612 bytes --]

Index: gcc/Makefile.in
===================================================================
RCS file: /egcs/carton/cvsfiles/egcs/gcc/Makefile.in,v
retrieving revision 1.113
diff -u -r1.113 Makefile.in
--- gcc/Makefile.in	1998/05/23 22:45:35	1.113
+++ gcc/Makefile.in	1998/05/24 06:59:55
@@ -2558,7 +2558,7 @@
 	else true; \
 	fi
 	@if  [ $(build_canonical) != $(host_canonical) ] ; then \
-	  echo "set tmpdir /tmp" >> ./tmp0 ; \
+	  echo "set tmpdir $${TESTSUITE_TMPDIR-/tmp}" >> ./tmp0 ; \
 	else echo "set tmpdir $(objdir)/testsuite" >> ./tmp0 ; \
 	fi
 	@echo "set srcdir \"\$${srcdir}/testsuite\"" >> ./tmp0

Re: none

[Jeffrey A Law]

  In message < orbtsocg0k.fsf@zecarneiro.lsd.dcc.unicamp.br >you write:
  > --Multipart_Sun_May_24_03:57:47_1998-1
  > Content-Type: text/plain; charset=US-ASCII
  > 
  > Jeffrey A Law <law@hurl.cygnus.com> writes:
  > 
  > > 	* Makefile.in (boostrap2-lean, bootstrap3-lean,
  > > 	bootstrap4-lean): new targets
  > > 	* gcc/Makefile.in (boostrap2-lean, bootstrap3-lean,
  > > 	bootstrap4-lean): ditto
  > > I've installed this patch.  Thanks!
  > 
  > How about the attached one too?  I wrote it so long ago that I had
  > already forgot about it... :-(
That's one I wanted to look at too.  I though there was already a
way to do the tmpdir thing.


  > > I want to take a little more time to think about the dejagnu
  > > patches.
  > 
  > Just have in mind that it won't change absolutely anything unless
  > CHECK_MULTI is set to something different from "n" or "no"...
OK.  Thanks.
jeff

Re: none

[Alexandre Oliva]

Jeffrey A Law <law@hurl.cygnus.com> writes:

> That's one I wanted to look at too.  I though there was already a
> way to do the tmpdir thing.

I don't think there is any easy one, since /tmp is hard-coded in the
Makefile.  Of course one can always make site.exp, modify it, then run 
make check, but I'd rather not have to do that... :-)

BTW, contrib/test_summary is not executable, could you please chmod it 
in the CVS main tree?

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

Re: none

[Jeffrey A Law]

  > BTW, contrib/test_summary is not executable, could you please chmod it 
  > in the CVS main tree?
Fixed.
jeff

Re: none

[Jeffrey A Law]

  In message < orbtsocg0k.fsf@zecarneiro.lsd.dcc.unicamp.br >you write:
  > Index: gcc/Makefile.in
  > ===================================================================
  > RCS file: /egcs/carton/cvsfiles/egcs/gcc/Makefile.in,v
  > retrieving revision 1.113
  > diff -u -r1.113 Makefile.in
  > --- gcc/Makefile.in	1998/05/23 22:45:35	1.113
  > +++ gcc/Makefile.in	1998/05/24 06:59:55
  > @@ -2558,7 +2558,7 @@
  >  	else true; \
  >  	fi
  >  	@if  [ $(build_canonical) != $(host_canonical) ] ; then \
  > -	  echo "set tmpdir /tmp" >> ./tmp0 ; \
  > +	  echo "set tmpdir $${TESTSUITE_TMPDIR-/tmp}" >> ./tmp0 ; \
  >  	else echo "set tmpdir $(objdir)/testsuite" >> ./tmp0 ; \
  >  	fi
  >  	@echo "set srcdir \"\$${srcdir}/testsuite\"" >> ./tmp0
Is there some reason to use TESTSUITE_TMPDIR over TMPDIR?

TMPDIR is a relatively standard way of specifying temporary
directories.

I'd consider the fact that using TMPDIR will force all the compiler
intermediate into the specified temporary directory desirable :-)

jeff

Re: none

[Alexandre Oliva]

Jeffrey A Law <law@hurl.cygnus.com> writes:

>   In message < orbtsocg0k.fsf@zecarneiro.lsd.dcc.unicamp.br >you write:
>> -	  echo "set tmpdir /tmp" >> ./tmp0 ; \
>> +	  echo "set tmpdir $${TESTSUITE_TMPDIR-/tmp}" >> ./tmp0 ; \

> Is there some reason to use TESTSUITE_TMPDIR over TMPDIR?

Yes.  I already have TMPDIR set to /tmp, which is usually mounted on
virtual memory (tmpfs); this is very fast for compiling files.
However, I wanted to be able to run simultaneous tests of egcs, with
different configuration flags (--{dis,en}able-haifa, for example), but
using a single TMPDIR caused name clashes in files created by
dejagnu.  Furthermore, I'd like failed binaries to be kept for further 
analysis, so I didn't want them to be stored in /tmp.

So I came up with a solution that would keep gcc using TMPDIR for its
temporaries, for speed, and would allow me to specify different
directories for each run of dejagnu.  Perhaps the patch should use
TMPDIR if TESTSUITE_TMPDIR is not set:

-	  echo "set tmpdir /tmp" >> ./tmp0 ; \
+	  echo "set tmpdir $${TESTSUITE_TMPDIR-$${TMPDIR-/tmp}}" >> ./tmp0 ; \

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

Re: none

[Jeffrey A Law]

  In message < oru36c2pju.fsf@zecarneiro.lsd.dcc.unicamp.br >you write:
  > > Is there some reason to use TESTSUITE_TMPDIR over TMPDIR?
  > 
  > Yes.  I already have TMPDIR set to /tmp, which is usually mounted on
  > virtual memory (tmpfs); this is very fast for compiling files.
  > However, I wanted to be able to run simultaneous tests of egcs, with
  > different configuration flags (--{dis,en}able-haifa, for example), but
  > using a single TMPDIR caused name clashes in files created by
  > dejagnu.  Furthermore, I'd like failed binaries to be kept for further 
  > analysis, so I didn't want them to be stored in /tmp.
  > 
  > So I came up with a solution that would keep gcc using TMPDIR for its
  > temporaries, for speed, and would allow me to specify different
  > directories for each run of dejagnu.  Perhaps the patch should use
  > TMPDIR if TESTSUITE_TMPDIR is not set:
Ah.  In that case, maybe the right way to go is to put temporary
files in the testsuite object directory itself since you'd have
different testsuite object directories for the different configs
right?

jeff

Re: none

[Alexandre Oliva]

Jeffrey A Law <law@hurl.cygnus.com> writes:

> Ah.  In that case, maybe the right way to go is to put temporary
> files in the testsuite object directory itself since you'd have
> different testsuite object directories for the different configs
> right?

That would be fine.

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

Re: none

[Joern Rennecke]

> 
>   In message < oru36c2pju.fsf@zecarneiro.lsd.dcc.unicamp.br >you write:
>   > > Is there some reason to use TESTSUITE_TMPDIR over TMPDIR?
>   > 
>   > Yes.  I already have TMPDIR set to /tmp, which is usually mounted on
>   > virtual memory (tmpfs); this is very fast for compiling files.
>   > However, I wanted to be able to run simultaneous tests of egcs, with
>   > different configuration flags (--{dis,en}able-haifa, for example), but
>   > using a single TMPDIR caused name clashes in files created by
>   > dejagnu.  Furthermore, I'd like failed binaries to be kept for further 
>   > analysis, so I didn't want them to be stored in /tmp.
>   > 
>   > So I came up with a solution that would keep gcc using TMPDIR for its
>   > temporaries, for speed, and would allow me to specify different
>   > directories for each run of dejagnu.  Perhaps the patch should use
>   > TMPDIR if TESTSUITE_TMPDIR is not set:
> Ah.  In that case, maybe the right way to go is to put temporary
> files in the testsuite object directory itself since you'd have
> different testsuite object directories for the different configs
> right?

No, this doesn't work when you want to run different tests from the
same (e.g. multilibbed) directory.

Re: none

[Jeffrey A Law]

  In message < 199805272005.VAA20442@phal.cygnus.co.uk >you write:
  > >   > So I came up with a solution that would keep gcc using TMPDIR for its
  > >   > temporaries, for speed, and would allow me to specify different
  > >   > directories for each run of dejagnu.  Perhaps the patch should use
  > >   > TMPDIR if TESTSUITE_TMPDIR is not set:
  > > Ah.  In that case, maybe the right way to go is to put temporary
  > > files in the testsuite object directory itself since you'd have
  > > different testsuite object directories for the different configs
  > > right?
  > 
  > No, this doesn't work when you want to run different tests from the
  > same (e.g. multilibbed) directory.
That's only an issue for folks doing parallel "make check" for multilibbed
configs.  Granted it'd be nice to handle that, but I don't think that's
the most important issue at hand, and it certainly doesn't work right
now!

If we want to handle that case, then I might suggest we have the
tmpdir be a subdir of the testsuite object directory.  The name of
the subdir would be derived from the multilib options.

So, does someone want to write a patch for that?  Or do we go with
a partial solution?

jeff

A patch for dejagnu

[H.J. Lu]

> That's only an issue for folks doing parallel "make check" for multilibbed
> configs.  Granted it'd be nice to handle that, but I don't think that's

While we are on parallel "make check", can you consider this patch
for dejagnu? Several calls to config.guess may be made at the same
time from the same directory during parallel "make check" on a SMP
machine. The current config.guess assumes only one call per directory
at any time. It is not always true.

Thanks.


H.J.
---
--- config.guess.orig	Fri Oct 31 13:26:24 1997
+++ config.guess	Fri Oct 31 13:28:31 1997
@@ -46,7 +46,9 @@
 UNAME_SYSTEM=`(uname -s) 2>/dev/null` || UNAME_SYSTEM=unknown
 UNAME_VERSION=`(uname -v) 2>/dev/null` || UNAME_VERSION=unknown
 
-trap 'rm -f dummy.c dummy.o dummy; exit 1' 1 2 15
+dummy=$$
+
+trap 'rm -f $dummy.c $dummy.o $dummy; exit 1' 1 2 15
 
 # Note: order is significant - the case branches are not exclusive.
 
@@ -59,7 +61,7 @@
 	# A Tn.n version is a released field test version.
 	# A Xn.n version is an unreleased experimental baselevel.
 	# 1.2 uses "1.2" for uname -r.
-	cat <<EOF >dummy.s
+	cat <<EOF >$dummy.s
 	.globl main
 	.ent main
 main:
@@ -71,9 +73,9 @@
 	ret \$31,(\$26),1
 	.end main
 EOF
-	${CC-cc} dummy.s -o dummy 2>/dev/null
+	${CC-cc} $dummy.s -o $dummy 2>/dev/null
 	if test "$?" = 0 ; then
-		./dummy
+		./$dummy
 		case "$?" in
 			1)
 				UNAME_MACHINE="alphaev5"
@@ -83,7 +85,7 @@
 				;;
 		esac
 	fi
-	rm -f dummy.s dummy
+	rm -f $dummy.s $dummy
 	echo ${UNAME_MACHINE}-dec-osf`echo ${UNAME_RELEASE} | sed -e 's/^[VTX]//' | tr [[A-Z]] [[a-z]]`
 	exit 0 ;;
     21064:Windows_NT:50:3)
@@ -172,6 +174,18 @@
 		;;
 	esac
 	exit 0 ;;
+    sun*:*:4.2BSD:*)
+	UNAME_RELEASE=`(head -1 /etc/motd | awk '{print substr($5,1,3)}') 2>/dev/null`
+	test "x${UNAME_RELEASE}" = "x" && UNAME_RELEASE=3
+	case "`/bin/arch`" in
+	    sun3)
+		echo m68k-sun-sunos${UNAME_RELEASE}
+		;;
+	    sun4)
+		echo sparc-sun-sunos${UNAME_RELEASE}
+		;;
+	esac
+	exit 0 ;;
     aushp:SunOS:*:*)
 	echo sparc-auspex-sunos${UNAME_RELEASE}
 	exit 0 ;;
@@ -215,7 +229,7 @@
 	echo clipper-intergraph-clix${UNAME_RELEASE}
 	exit 0 ;;
     mips:*:*:UMIPS | mips:*:*:RISCos)
-	sed 's/^	//' << EOF >dummy.c
+	sed 's/^	//' << EOF >$dummy.c
 	int main (argc, argv) int argc; char **argv; {
 	#if defined (host_mips) && defined (MIPSEB)
 	#if defined (SYSTYPE_SYSV)
@@ -231,10 +245,10 @@
 	  exit (-1);
 	}
 EOF
-	${CC-cc} dummy.c -o dummy \
-	  && ./dummy `echo "${UNAME_RELEASE}" | sed -n 's/\([0-9]*\).*/\1/p'` \
-	  && rm dummy.c dummy && exit 0
-	rm -f dummy.c dummy
+	${CC-cc} $dummy.c -o $dummy \
+	  && ./$dummy `echo "${UNAME_RELEASE}" | sed -n 's/\([0-9]*\).*/\1/p'` \
+	  && rm $dummy.c $dummy && exit 0
+	rm -f $dummy.c $dummy
 	echo mips-mips-riscos${UNAME_RELEASE}
 	exit 0 ;;
     Night_Hawk:Power_UNIX:*:*)
@@ -286,7 +300,7 @@
 	exit 0 ;;
     *:AIX:2:3)
 	if grep bos325 /usr/include/stdio.h >/dev/null 2>&1; then
-		sed 's/^		//' << EOF >dummy.c
+		sed 's/^		//' << EOF >$dummy.c
 		#include <sys/systemcfg.h>
 
 		main()
@@ -297,8 +311,8 @@
 			exit(0);
 			}
 EOF
-		${CC-cc} dummy.c -o dummy && ./dummy && rm dummy.c dummy && exit 0
-		rm -f dummy.c dummy
+		${CC-cc} $dummy.c -o $dummy && ./$dummy && rm $dummy.c $dummy && exit 0
+		rm -f $dummy.c $dummy
 		echo rs6000-ibm-aix3.2.5
 	elif grep bos324 /usr/include/stdio.h >/dev/null 2>&1; then
 		echo rs6000-ibm-aix3.2.4
@@ -351,7 +365,7 @@
 	echo ${HP_ARCH}-hp-hpux${HPUX_REV}
 	exit 0 ;;
     3050*:HI-UX:*:*)
-	sed 's/^	//' << EOF >dummy.c
+	sed 's/^	//' << EOF >$dummy.c
 	#include <unistd.h>
 	int
 	main ()
@@ -376,8 +390,8 @@
 	  exit (0);
 	}
 EOF
-	${CC-cc} dummy.c -o dummy && ./dummy && rm dummy.c dummy && exit 0
-	rm -f dummy.c dummy
+	${CC-cc} $dummy.c -o $dummy && ./$dummy && rm $dummy.c $dummy && exit 0
+	rm -f $dummy.c $dummy
 	echo unknown-hitachi-hiuxwe2
 	exit 0 ;;
     9000/7??:4.3bsd:*:* | 9000/8?[79]:4.3bsd:*:* )
@@ -497,7 +511,7 @@
 	esac
 
 	if test "${UNAME_MACHINE}" = "alpha" ; then
-		sed 's/^	//'  <<EOF >dummy.s
+		sed 's/^	//'  <<EOF >$dummy.s
 		.globl main
 		.ent main
 	main:
@@ -510,9 +524,9 @@
 		.end main
 EOF
 		LIBC=""
-		${CC-cc} dummy.s -o dummy 2>/dev/null
+		${CC-cc} $dummy.s -o $dummy 2>/dev/null
 		if test "$?" = 0 ; then
-			./dummy
+			./$dummy
 			case "$?" in
 				1)
 					UNAME_MACHINE="alphaev5"
@@ -522,16 +536,16 @@
 				;;	
 			esac	
 
-			objdump --private-headers dummy | \
+			objdump --private-headers $dummy | \
 			  grep ld.so.1 > /dev/null
 			if test "$?" = 0 ; then
 				LIBC="libc1"
 			fi
 		fi	
-		rm -f dummy.s dummy
+		rm -f $dummy.s $dummy
 		echo ${UNAME_MACHINE}-unknown-linux-gnu${LIBC} ; exit 0
 	elif test "${UNAME_MACHINE}" = "mips" ; then
-	  cat >dummy.c <<EOF
+	  cat >$dummy.c <<EOF
 main(argc, argv)
      int argc;
      char *argv[];
@@ -545,8 +559,8 @@
   return 0;
 }
 EOF
-	  ${CC-cc} dummy.c -o dummy 2>/dev/null && ./dummy "${UNAME_MACHINE}" && rm dummy.c dummy && exit 0
-	  rm -f dummy.c dummy
+	  ${CC-cc} $dummy.c -o $dummy 2>/dev/null && ./$dummy "${UNAME_MACHINE}" && rm $dummy.c $dummy && exit 0
+	  rm -f $dummy.c $dummy
 	else
 	  # Either a pre-BFD a.out linker (linux-gnuoldld)
 	  # or one that does not give us useful --help.
@@ -565,7 +579,7 @@
 	    ;;
 	  esac
 	  # Determine whether the default compiler is a.out or elf
-	  cat >dummy.c <<EOF
+	  cat >$dummy.c <<EOF
 #include <features.h>
 main(argc, argv)
      int argc;
@@ -587,8 +601,8 @@
   return 0;
 }
 EOF
-	  ${CC-cc} dummy.c -o dummy 2>/dev/null && ./dummy "${UNAME_MACHINE}" && rm dummy.c dummy && exit 0
-	  rm -f dummy.c dummy
+	  ${CC-cc} $dummy.c -o $dummy 2>/dev/null && ./$dummy "${UNAME_MACHINE}" && rm $dummy.c $dummy && exit 0
+	  rm -f $dummy.c $dummy
 	fi ;;
 # ptx 4.0 does uname -s correctly, with DYNIX/ptx in there.  earlier versions
 # are messed up and put the nodename in both sysname and nodename.
@@ -719,7 +733,7 @@
 #echo '(No uname command or uname output not recognized.)' 1>&2
 #echo "${UNAME_MACHINE}:${UNAME_SYSTEM}:${UNAME_RELEASE}:${UNAME_VERSION}" 1>&2
 
-cat >dummy.c <<EOF
+cat >$dummy.c <<EOF
 #ifdef _SEQUENT_
 # include <sys/types.h>
 # include <sys/utsname.h>
@@ -817,8 +831,8 @@
 }
 EOF
 
-${CC-cc} dummy.c -o dummy 2>/dev/null && ./dummy && rm dummy.c dummy && exit 0
-rm -f dummy.c dummy
+${CC-cc} $dummy.c -o $dummy 2>/dev/null && ./$dummy && rm $dummy.c $dummy && exit 0
+rm -f $dummy.c $dummy
 
 # Apollos put the system type in the environment.
 

A patch for dejagnu

[H.J. Lu]

Hi,

Some people have reported problems with the latest dejagnu and
egcs that dejagnu cannot find the correct libstdc++ and libio.
Here is a patch which seems to work for me.


-- 
H.J. Lu (hjl@gnu.org)
---
--- libgloss.exp.orig	Wed Mar 11 08:26:07 1998
+++ libgloss.exp	Wed Mar 11 09:33:34 1998
@@ -425,14 +425,17 @@
 
 
     # search for the top level multilib directory
-    set multitop [lookfor_file "${comp_base_dir}" "${target_alias}"]
+    set multitop [lookfor_file "${comp_base_dir}" "libraries"]
     if { $multitop == "" } {
-	set multitop [lookfor_file "${comp_base_dir}" "libraries"]
-	if { $multitop == "" } {
-	    set multitop "[lookfor_file ${comp_base_dir} gcc/xgcc]"
-	    if { $multitop != "" } {
-		set multitop [file dirname [file dirname $multitop]];
-	    } else {
+	set multitop "[lookfor_file ${comp_base_dir} gcc/xgcc]"
+	if { $multitop != "" } {
+	    set multitop [file dirname [file dirname $multitop]];
+	} else {
+	    # We search ${target_alias} last since it is only valid
+	    # when we are running dejagnu on an installed gcc. We
+	    # don't want to get it by accident.
+	    set multitop [lookfor_file "${comp_base_dir}" "${target_alias}"]
+	    if { $multitop == "" } {
 		return ""
 	    }
 	}

Re: A patch for dejagnu

[Jim Wilson]

	+	    # We search ${target_alias} last since it is only valid
	+	    # when we are running dejagnu on an installed gcc. We
	+	    # don't want to get it by accident.

This is due to an unfortunate difference between the Cygnus source tree and
the EGCS source tree.  In EGCS, we build libraries in a subdirectory called
`libraries'.  In the Cygnus source tree, we build libraries in a subdirectory
called `${target_alias}'.  Hence searching target_alias first is correct
for the Cygnus sources, but not correct for EGCS.

Historical info: The Cygnus source tree creates the subdirectory only if
we are doing a cross, in which case --target=foo must have been specified,
and using foo for the subdirectory name was reasonable.  EGCS always creates
the subdirectory, so as to make multilibbed native builds work.  Since there
is usually no --target=foo option, using target_alias is unsafe, as it might
not be a valid filename.  I changed EGCS to use libraries instead, but this
change never got propagated back into the Cygnus sources.

I probably should make an attempt to get this `libraries' change into the
Cygnus sources, so that the two source trees will be consistent, and then
we can make a dejagnu that will work for both of them.

Jim

Re: A patch for dejagnu

[Jeffrey A Law]

  In message < 199803130100.RAA22598@rtl.cygnus.com >you write:
  > I probably should make an attempt to get this `libraries' change into the
  > Cygnus sources, so that the two source trees will be consistent, and then
  > we can make a dejagnu that will work for both of them.
This is probably best.

The core issue which caused us to create "libraries" is also an issue
we need to solve for Cygnus releases.

Jeff