Re: FIXINCL goes into an infinite loop on alpha-osf4

Re: FIXINCL goes into an infinite loop on alpha-osf4

[Bruce Korb]

"Kaveh R. Ghazi" wrote:
> 
> Today's CVS now completes running fixincl on my alpha-osf4 box.  BTW,
> it seems to run a *lot* faster now.  And only about half as many files
> are transformed.  Good work!  (Next I'll try sunos4 and see how that
> does.)

Yes, Zack, GOOD WORK!  I had been wanting to get to it
for almost a year now.  Round tuits can be hard to come by :-(

Given that you (Kaveh) have access to Sun OS4 and I don't,
please try defining NO_BOGOSITY and see if it works.
We have fixed some troublesome issues since I deemed
it necessary, long, long ago.

If it does work, it will save a lot of fork-ing around.
The other troublesome box is HP/UX.  If we can dump
the BOGOSITY code for those boxes, we ought to
be able to dump it for all.

Regards,
	Bruce

Re: FIXINCL goes into an infinite loop on alpha-osf4

[Zack Weinberg]

I committed the patch I sent previously, I believe it qualifies as an
obvious bugfix.

ChangeLog:

	* fixinc/Makefile.in (machname.h): SunOS 4 sed can't handle a
	file with one line and no trailing newline.
	Patch by Kaveh Ghazi <ghazi@caip.rutgers.edu>.
	* fixinc/fixtests.c (machine_name_test): Fix fencepost error
	checking if the match is on the line.
	* fixinc/gnu-regex.c: Provide regerror not __regerror.

zw

Re: FIXINCL goes into an infinite loop on alpha-osf4

[Zack Weinberg]

Ok.  Here is my patch for the problems you saw and for a silly fencepost
error in machine_name_test.  With this applied, I am confident that
fixincludes doesn't munge anything it shouldn't in the SunOS 4 headers.
I am less confident that it does munge everything that it should, but an
eyeball of the diffs looks okay.

An interesting statistic:  If you turn off else_endif_label, the number
of headers touched on SunOS drops by 313 files.

zw

===================================================================
Index: Makefile.in
--- Makefile.in	2000/01/20 18:25:12	1.13
+++ Makefile.in	2000/01/26 02:16:25
@@ -145,7 +145,7 @@ machname.h: ../specs
 	    sed -n 's/^.*-D\([a-zA-Z_][a-zA-Z0-9_]*\).*$$/\1/p' | sort -u | \
 	    grep -v '^_[_A-Z]' > mn.T
 	@echo "Forbidden identifiers: `tr '\012' ' ' <mn.T`"
-	@sed 's/^/\\\\</; s/$$/\\\\>/' <mn.T | tr '\012' '|' | \
-	     sed 's/^/#define MN_NAME_PAT "/; s/|$$/"/' > machname.h
-	@echo >> machname.h
-	@-rm -f mn.T
+	@sed 's/^/\\\\</; s/$$/\\\\>/' <mn.T | tr '\012' '|' > mn.T2
+	@echo '' >>mn.T2
+	@sed 's/^/#define MN_NAME_PAT "/; s/|$$/"/' < mn.T2 > machname.h
+	@-rm -f mn.T mn.T2
===================================================================
Index: fixtests.c
--- fixtests.c	2000/01/20 18:25:12	1.10
+++ fixtests.c	2000/01/26 02:16:25
@@ -307,7 +307,7 @@ TEST_FOR_FIX_PROC_HEAD( machine_name_tes
 	return SKIP_FIX;  /* No match in file - no fix needed */
 
       /* Match; is it on the line?  */
-      if (match[0].rm_eo < limit - base)
+      if (match[0].rm_eo <= limit - base)
 	return APPLY_FIX;  /* Yup */
 
       /* Otherwise, keep looking... */
===================================================================
Index: gnu-regex.c
--- gnu-regex.c	2000/01/20 18:25:12	1.3
+++ gnu-regex.c	2000/01/26 02:16:27
@@ -5745,7 +5745,7 @@ weak_alias (__regexec, regexec)
    from either regcomp or regexec.   We don't use PREG here.  */
 
 size_t
-__regerror (errcode, preg, errbuf, errbuf_size)
+regerror (errcode, preg, errbuf, errbuf_size)
     int errcode;
     const regex_t *preg;
     char *errbuf;

Re: FIXINCL goes into an infinite loop on alpha-osf4

[Zack Weinberg]

On Tue, Jan 25, 2000 at 02:59:16PM -0500, Kaveh R. Ghazi wrote:
> --- orig/egcs-CVS20000124/gcc/fixinc/Makefile.in	Fri Jan 21 10:43:14 2000
> +++ egcs-CVS20000124/gcc/fixinc/Makefile.in	Tue Jan 25 12:03:05 2000
> @@ -145,7 +145,7 @@ machname.h: ../specs
>  	    sed -n 's/^.*-D\([a-zA-Z_][a-zA-Z0-9_]*\).*$$/\1/p' | sort -u | \
>  	    grep -v '^_[_A-Z]' > mn.T
>  	@echo "Forbidden identifiers: `tr '\012' ' ' <mn.T`"
> -	@sed 's/^/\\\\</; s/$$/\\\\>/' <mn.T | tr '\012' '|' | \
> -	     sed 's/^/#define MN_NAME_PAT "/; s/|$$/"/' > machname.h
> -	@echo >> machname.h
> -	@-rm -f mn.T
> +	@sed 's/^/\\\\</; s/$$/\\\\>/' <mn.T | tr '\012' '|' > mn.T2
> +	@echo '' >> mn.T2
> +	@sed 's/^/#define MN_NAME_PAT "/; s/|$$/"/' < mn.T2 > machname.h
> +	@-rm -f mn.T mn.T2

This part looks fine to me.

> --- orig/egcs-CVS20000124/gcc/fixinc/fixlib.c	Fri Jan 21 10:43:15 2000
> +++ egcs-CVS20000124/gcc/fixinc/fixlib.c	Tue Jan 25 13:34:33 2000
> @@ -166,7 +166,7 @@ compile_re( pat, re, match, e1, e2 )
>    if (err)
>      {
>        char rerrbuf[1024];
> -      regerror (err, re, rerrbuf, 1024);
> +      __regerror (err, re, rerrbuf, 1024);
>        fprintf (stderr, z_bad_comp, e1, e2, pat, rerrbuf);
>        exit (EXIT_FAILURE);
>      }

Better you should do this:

--- gnu-regex.c 2000/01/20 18:25:12     1.3
+++ gnu-regex.c 2000/01/26 01:31:08
@@ -5745,7 +5745,7 @@
    from either regcomp or regexec.   We don't use PREG here.  */
 
 size_t
-__regerror (errcode, preg, errbuf, errbuf_size)
+regerror (errcode, preg, errbuf, errbuf_size)
     int errcode;
     const regex_t *preg;
     char *errbuf;

This is the way it is in libc's copy.  (Note, the patch was pasted from
another xterm and may have tab damage.)

With those changes fixincludes runs to completion on sunos 4.1.4, but
there appears to be something wrong with the machine_name fix - it doesn't
touch "#ifdef sparc", but does touch "#if defined(sparc)".  I'm looking
into it now.

zw

Re: FIXINCL goes into an infinite loop on alpha-osf4

[Zack Weinberg]

On Tue, Jan 25, 2000 at 04:37:17PM -0800, Richard Henderson wrote:
> On Tue, Jan 25, 2000 at 02:59:16PM -0500, Kaveh R. Ghazi wrote:
> > It looks like gnu-regex.c only provides a weak-alias for regerror if
> > we define _LIBC.  This is sunos4 land, so I changed the call in
> > fixlib.c from regerror to __regerror and it worked.
> 
> I'd prefer we hacked sources the other direction.

Agree.

I'm testing fixinc on SunOS 4 right now, and will look at this.

zw

Re: FIXINCL goes into an infinite loop on alpha-osf4

[Richard Henderson]

On Tue, Jan 25, 2000 at 02:59:16PM -0500, Kaveh R. Ghazi wrote:
> It looks like gnu-regex.c only provides a weak-alias for regerror if
> we define _LIBC.  This is sunos4 land, so I changed the call in
> fixlib.c from regerror to __regerror and it worked.

I'd prefer we hacked sources the other direction.


r~

Re: FIXINCL goes into an infinite loop on alpha-osf4

[Kaveh R. Ghazi]

 > From: Bruce Korb <bkorb@sco.COM>
 > 
 > Given that you (Kaveh) have access to Sun OS4 and I don't, please try
 > defining NO_BOGOSITY and see if it works.  We have fixed some
 > troublesome issues since I deemed it necessary, long, long ago.

Well, before trying NO_BOGOSITY, I needed to have it work at all on
sunos4. :-)  There were two problems.  

1.  The first was that it couldn't compile fixlib.c:
 > fixlib.c", line 185: warning: undeclared initializer name MN_NAME_PAT

This happened because machname.h was empty.  I tracked this down to
the fact that sed on sunos4 can't handle a file which doesn't end in a
newline.  The Makefile rules for creating machname.h did create this
circumstance.  So I rewrote it to make sure that there is always a
newline by reordering when the "echo" statement is run.  I also had it
say echo '' instead of just echo, because echo alone doesn't do
anything.

2.  I then had a problem linking fixincl and fixfixes:
 > ld: Undefined symbol
 >    _regerror

It looks like gnu-regex.c only provides a weak-alias for regerror if
we define _LIBC.  This is sunos4 land, so I changed the call in
fixlib.c from regerror to __regerror and it worked.


Okay to install?



2000-01-25  Kaveh R. Ghazi  <ghazi@caip.rutgers.edu>

	* fixinc/Makefile.in (machname.h): Make sure data sent to `sed'
	has a terminating newline.

	* fixlib.c (compile_re): Call `__regerror', not `regerror'.

diff -rup orig/egcs-CVS20000124/gcc/fixinc/Makefile.in egcs-CVS20000124/gcc/fixinc/Makefile.in
--- orig/egcs-CVS20000124/gcc/fixinc/Makefile.in	Fri Jan 21 10:43:14 2000
+++ egcs-CVS20000124/gcc/fixinc/Makefile.in	Tue Jan 25 12:03:05 2000
@@ -145,7 +145,7 @@ machname.h: ../specs
 	    sed -n 's/^.*-D\([a-zA-Z_][a-zA-Z0-9_]*\).*$$/\1/p' | sort -u | \
 	    grep -v '^_[_A-Z]' > mn.T
 	@echo "Forbidden identifiers: `tr '\012' ' ' <mn.T`"
-	@sed 's/^/\\\\</; s/$$/\\\\>/' <mn.T | tr '\012' '|' | \
-	     sed 's/^/#define MN_NAME_PAT "/; s/|$$/"/' > machname.h
-	@echo >> machname.h
-	@-rm -f mn.T
+	@sed 's/^/\\\\</; s/$$/\\\\>/' <mn.T | tr '\012' '|' > mn.T2
+	@echo '' >> mn.T2
+	@sed 's/^/#define MN_NAME_PAT "/; s/|$$/"/' < mn.T2 > machname.h
+	@-rm -f mn.T mn.T2
diff -rup orig/egcs-CVS20000124/gcc/fixinc/fixlib.c egcs-CVS20000124/gcc/fixinc/fixlib.c
--- orig/egcs-CVS20000124/gcc/fixinc/fixlib.c	Fri Jan 21 10:43:15 2000
+++ egcs-CVS20000124/gcc/fixinc/fixlib.c	Tue Jan 25 13:34:33 2000
@@ -166,7 +166,7 @@ compile_re( pat, re, match, e1, e2 )
   if (err)
     {
       char rerrbuf[1024];
-      regerror (err, re, rerrbuf, 1024);
+      __regerror (err, re, rerrbuf, 1024);
       fprintf (stderr, z_bad_comp, e1, e2, pat, rerrbuf);
       exit (EXIT_FAILURE);
     }