[PATCH] Fix perl build problems on Stratus VOS

[p5pRT]

Migrated from rt.perl.org#78132 (status was 'resolved')

Searchable as RT78132$

[p5pRT]

From @paulg1973

Gentle Perl5 Porters,

The attached text files contain patches to correct build problems on the
Stratus VOS (recently renamed "OpenVOS") operating system. I have tested
these changes on OpenVOS Release 17.0, which is the most-current
customer release. None of these changes should affect any other OS.

Makefile.SH​: This patch removes the "miniperl" dependency of the "all"
target. On an operating system that does not require an executable
suffix, the miniperl$(EXE_EXT) dependency evaluates to "miniperl", too.
But on an operating system like VOS that does have an executable suffix,
miniperl$(EXE_EXT) evaluates to (in our case) "miniperl.pm" and the
"miniperl" target is unresolved. [ perl-5.10.x, perl-5.12.x,
perl-current ]

ext/Socket/Socket.xs​: Sadly, OpenVOS does not yet support IPv6. I edited
the code to allow for this case, while retaining IPv6 support for
operating systems that do support it. [ perl-5.12.x, perl-current ]

The attached patches apply cleanly as of today.

These patches were prepared using "diff -u"; I have not completed the
port of git to OpenVOS, so I am unable to submit them in git format. I
apologize for this; hopefully by the time I have the next patches ready,
I'll have git working.

These patches by themselves do not get perl5 to run all of the test
cases, but they at least get it to build again.

<<perl-current.patch.txt>> <<perl-5.10.x.patch.txt>>
<<perl-5.12.x.patch.txt>>
Thanks
PG
--
Paul Green, Senior Technical Consultant, Stratus Technologies.
Voice​: +1 978-461-7557; FAX​: +1 978-461-3610; Mobile​: +1 (978) 235-2451;
AIM​: PaulGreen

[p5pRT]

From @paulg1973

Inline Patch

--- perl-current/Makefile.SH~	2010-09-05 10:39:10.000000000 -0400
+++ perl-current/Makefile.SH	2010-09-14 00:49:48.000000000 -0400
@@ -555,7 +555,7 @@
 .c.s:
 	$(CCCMDSRC) -S $*.c
 
-all: $(FIRSTMAKEFILE) $(MINIPERL_EXE) miniperl $(generated_pods) $(private) $(unidatafiles) $(public) $(dynamic_ext) $(nonxs_ext) extras.make
+all: $(FIRSTMAKEFILE) $(MINIPERL_EXE) $(generated_pods) $(private) $(unidatafiles) $(public) $(dynamic_ext) $(nonxs_ext) extras.make
 	@echo " ";
 	@echo "	Everything is up to date. Type '$(MAKE) test' to run test suite."
 
--- perl-current/ext/Socket/Socket.xs~	2010-04-24 16:58:10.000000000 -0400
+++ perl-current/ext/Socket/Socket.xs	2010-09-13 17:07:15.000000000 -0400
@@ -465,22 +465,34 @@
         CODE:
 #ifdef HAS_INETNTOP
 	STRLEN addrlen, struct_size;
+#ifdef AF_INET6
 	struct in6_addr addr;
 	char str[INET6_ADDRSTRLEN];
+#else
+	struct in_addr addr;
+	char str[INET_ADDRSTRLEN];
+#endif
 	char *ip_address = SvPV(ip_address_sv, addrlen);
 
-        if(af == AF_INET) {
-            struct_size = sizeof(struct in_addr);
-        } else if(af == AF_INET6) {
-            struct_size = sizeof(struct in6_addr);
-        } else {
-           croak("Bad address family for %s, got %d, should be either AF_INET or AF_INET6",
+	struct_size = sizeof(addr);
+
+	if(af != AF_INET
+#ifdef AF_INET6
+	    && af != AF_INET6
+#endif
+	  ) {
+           croak("Bad address family for %s, got %d, should be"
+#ifdef AF_INET6
+	       " either AF_INET or AF_INET6",
+#else
+	       " AF_INET",
+#endif
                "Socket::inet_ntop",
                af);
         }
 
 	Copy( ip_address, &addr, sizeof addr, char );
-	inet_ntop(af, &addr, str, INET6_ADDRSTRLEN);
+	inet_ntop(af, &addr, str, sizeof str);
 
 	ST(0) = newSVpvn_flags(str, strlen(str), SVs_TEMP);
 #else
@@ -494,9 +506,23 @@
         CODE:
 #ifdef HAS_INETPTON
         int ok;
-        struct in6_addr ip_address;
-        if(af != AF_INET && af != AF_INET6) {
-           croak("Bad address family for %s, got %d, should be either AF_INET or AF_INET6",
+#ifdef AF_INET6
+	struct in6_addr ip_address;
+#else
+	struct in_addr ip_address;
+#endif
+
+	if(af != AF_INET
+#ifdef AF_INET6
+		&& af != AF_INET6
+#endif
+	  ) {
+		croak("Bad address family for %s, got %d, should be"
+#ifdef AF_INET6
+			" either AF_INET or AF_INET6",
+#else
+			" AF_INET",
+#endif
                         "Socket::inet_pton",
                         af);
         }
@@ -504,8 +530,7 @@
 
         ST(0) = sv_newmortal();
         if (ok) {
-                sv_setpvn( ST(0), (char *)&ip_address,
-                           af == AF_INET6 ? sizeof(ip_address) : sizeof(struct in_addr) );
+                sv_setpvn( ST(0), (char *)&ip_address, sizeof(ip_address) );
         }
 #else
         ST(0) = (SV *)not_here("inet_pton");

[p5pRT]

From @paulg1973

Inline Patch

--- perl-5.10.x/Makefile.SH~	2009-08-23 12:24:05.000000000 -0400
+++ perl-5.10.x/Makefile.SH	2010-09-14 11:53:19.000000000 -0400
@@ -537,7 +537,7 @@
 .c.s:
 	$(CCCMDSRC) -S $*.c
 
-all: $(FIRSTMAKEFILE) miniperl$(EXE_EXT) miniperl $(generated_pods) $(private) $(unidatafiles) $(public) $(dynamic_ext) $(nonxs_ext) extras.make
+all: $(FIRSTMAKEFILE) miniperl$(EXE_EXT) $(generated_pods) $(private) $(unidatafiles) $(public) $(dynamic_ext) $(nonxs_ext) extras.make
 	@echo " ";
 	@echo "	Everything is up to date. Type '$(MAKE) test' to run test suite."
 

[p5pRT]

From @paulg1973

Inline Patch

--- perl-5.12.x/Makefile.SH~	2010-08-21 01:01:06.000000000 -0400
+++ perl-5.12.x/Makefile.SH	2010-09-14 11:57:50.000000000 -0400
@@ -553,7 +553,7 @@
 .c.s:
 	$(CCCMDSRC) -S $*.c
 
-all: $(FIRSTMAKEFILE) $(MINIPERL_EXE) miniperl $(generated_pods) $(private) $(unidatafiles) $(public) $(dynamic_ext) $(nonxs_ext) extras.make
+all: $(FIRSTMAKEFILE) $(MINIPERL_EXE) $(generated_pods) $(private) $(unidatafiles) $(public) $(dynamic_ext) $(nonxs_ext) extras.make
 	@echo " ";
 	@echo "	Everything is up to date. Type '$(MAKE) test' to run test suite."
 
--- perl-current/ext/Socket/Socket.xs~	2010-04-24 16:58:10.000000000 -0400
+++ perl-current/ext/Socket/Socket.xs	2010-09-13 17:07:15.000000000 -0400
@@ -465,21 +465,34 @@
         CODE:
 #ifdef HAS_INETNTOP
 	STRLEN addrlen, struct_size;
+#ifdef AF_INET6
 	struct in6_addr addr;
 	char str[INET6_ADDRSTRLEN];
+#else
+	struct in_addr addr;
+	char str[INET_ADDRSTRLEN];
+#endif
 	char *ip_address = SvPV(ip_address_sv, addrlen);
 
-        if(af == AF_INET) {
-            struct_size = sizeof(struct in_addr);
-        } else if(af == AF_INET6) {
-            struct_size = sizeof(struct in6_addr);
-        } else {
-           croak("Bad address family for Socket::inet_ntop, got %d, should be either AF_INET or AF_INET6",
+	struct_size = sizeof(addr);
+
+	if(af != AF_INET
+#ifdef AF_INET6
+	    && af != AF_INET6
+#endif
+	  ) {
+           croak("Bad address family for %s, got %d, should be"
+#ifdef AF_INET6
+	       " either AF_INET or AF_INET6",
+#else
+	       " AF_INET",
+#endif
+	       "Socket::inet_ntop",
                af);
         }
 
 	Copy( ip_address, &addr, sizeof addr, char );
-	inet_ntop(af, &addr, str, INET6_ADDRSTRLEN);
+	inet_ntop(af, &addr, str, sizeof str);
 
 	ST(0) = newSVpvn_flags(str, strlen(str), SVs_TEMP);
 #else
@@ -493,9 +506,23 @@
         CODE:
 #ifdef HAS_INETPTON
         int ok;
-        struct in6_addr ip_address;
-        if(af != AF_INET && af != AF_INET6) {
-           croak("Bad address family for %s, got %d, should be either AF_INET or AF_INET6",
+#ifdef AF_INET6
+	struct in6_addr ip_address;
+#else
+	struct in_addr ip_address;
+#endif
+
+	if(af != AF_INET
+#ifdef AF_INET6
+		&& af != AF_INET6
+#endif
+	  ) {
+		croak("Bad address family for %s, got %d, should be"
+#ifdef AF_INET6
+			" either AF_INET or AF_INET6",
+#else
+			" AF_INET",
+#endif
                         "Socket::inet_pton",
                         af);
         }
@@ -503,8 +530,7 @@
 
         ST(0) = sv_newmortal();
         if (ok) {
-                sv_setpvn( ST(0), (char *)&ip_address,
-                           af == AF_INET6 ? sizeof(ip_address) : sizeof(struct in_addr) );
+                sv_setpvn( ST(0), (char *)&ip_address, sizeof(ip_address) );
         }
 #else
         ST(0) = (SV *)not_here("inet_pton");

[p5pRT]

From @paulg1973

I don't know what happened to mess up this letter. There were 3 enclosures. One of the enclosures seems to have become the body of the letter. The body of the letter got turned into an enclosure. The other two enclosures seem to be ok. Here is the original letter again. If anybody has any questions about these proposed changes, please let me know.

Gentle Perl5 Porters,

The attached text files contain patches to correct build problems on the
Stratus VOS (recently renamed "OpenVOS") operating system. I have tested
these changes on OpenVOS Release 17.0, which is the most-current
customer release. None of these changes should affect any other OS.

Makefile.SH​: This patch removes the "miniperl" dependency of the "all"
target. On an operating system that does not require an executable
suffix, the miniperl$(EXE_EXT) dependency evaluates to "miniperl", too.
But on an operating system like VOS that does have an executable suffix,
miniperl$(EXE_EXT) evaluates to (in our case) "miniperl.pm" and the
"miniperl" target is unresolved. [ perl-5.10.x, perl-5.12.x,
perl-current ]

ext/Socket/Socket.xs​: Sadly, OpenVOS does not yet support IPv6. I edited
the code to allow for this case, while retaining IPv6 support for
operating systems that do support it. [ perl-5.12.x, perl-current ]

The attached patches apply cleanly as of today.

These patches were prepared using "diff -u"; I have not completed the
port of git to OpenVOS, so I am unable to submit them in git format. I
apologize for this; hopefully by the time I have the next patches ready,
I'll have git working.

These patches by themselves do not get perl5 to run all of the test
cases, but they at least get it to build again.

<<perl-current.patch.txt>> <<perl-5.10.x.patch.txt>>
<<perl-5.12.x.patch.txt>>

Thanks
PG
--
Paul Green, Senior Technical Consultant, Stratus Technologies.
Voice​: +1 978-461-7557; FAX​: +1 978-461-3610; Mobile​: +1 (978) 235-2451;
AIM​: PaulGreen

[p5pRT]

From @cpansprout

On Fri Oct 01 13​:48​:32 2010, paulg wrote​:

I don't know what happened to mess up this letter. There were 3
enclosures. One of the enclosures seems to have become the body of
the letter. The body of the letter got turned into an enclosure.
The other two enclosures seem to be ok. Here is the original
letter again. If anybody has any questions about these proposed
changes, please let me know.

Gentle Perl5 Porters,

The
attached text files contain patches to correct build problems on
the
Stratus VOS (recently renamed "OpenVOS") operating system. I
have tested
these changes on OpenVOS Release 17.0, which is the
most-current
customer release. None of these changes should affect
any other OS.

Makefile.SH​: This patch removes the "miniperl"
dependency of the "all"
target. On an operating system that does
not require an executable
suffix, the miniperl$(EXE_EXT)
dependency evaluates to "miniperl", too.
But on an operating
system like VOS that does have an executable suffix,
miniperl$(EXE_EXT) evaluates to (in our case) "miniperl.pm" and the
"miniperl" target is unresolved. [ perl-5.10.x, perl-5.12.x,
perl-
current ]

ext/Socket/Socket.xs​: Sadly, OpenVOS does not yet
support IPv6. I edited
the code to allow for this case, while
retaining IPv6 support for
operating systems that do support it. [
perl-5.12.x, perl-current ]

The attached patches apply cleanly
as of today.

These patches were prepared using "diff -u"; I have
not completed the
port of git to OpenVOS, so I am unable to submit
them in git format. I
apologize for this; hopefully by the time I
have the next patches ready,
I'll have git working.

These
patches by themselves do not get perl5 to run all of the test
cases, but they at least get it to build again.

<<perl-
current.patch.txt>> <<perl-5.10.x.patch.txt>>
<<perl-
5.12.x.patch.txt>>

Thank you. I’ve just applied perl-current.patch.txt as a5addb1.

I’m unsure as to the current policy with regard to maintenance patches.
Usually they are cherry-picked from blead. (I don’t know how to do that,
anyway. :-)

I’ll let someone else take over.

[p5pRT]

The RT System itself - Status changed from 'new' to 'open'

[p5pRT]

From @cpansprout

On Fri Oct 01 13​:48​:32 2010, paulg wrote​:

The
attached text files contain patches to correct build problems on
the
Stratus VOS (recently renamed "OpenVOS") operating system. I
have tested
these changes on OpenVOS Release 17.0, which is the
most-current
customer release. None of these changes should affect
any other OS.

Is this deserving of a mention in perldelta?

[p5pRT]

From @rafl

"Father Chrysostomos via RT" <perlbug-followup@​perl.org> writes​:

On Fri Oct 01 13​:48​:32 2010, paulg wrote​:

The attached text files contain patches to correct build problems on
the Stratus VOS (recently renamed "OpenVOS") operating system. I have
tested these changes on OpenVOS Release 17.0, which is the
most-current customer release. None of these changes should affect
any other OS.

Is this deserving of a mention in perldelta?

Yes, I think so. In the section for platform-specific notes.

[p5pRT]

From @cpansprout

On Sun Oct 03 14​:28​:06 2010, rafl@​debian.org wrote​:

"Father Chrysostomos via RT" <perlbug-followup@​perl.org> writes​:

On Fri Oct 01 13​:48​:32 2010, paulg wrote​:

The attached text files contain patches to correct build problems on
the Stratus VOS (recently renamed "OpenVOS") operating system. I have
tested these changes on OpenVOS Release 17.0, which is the
most-current customer release. None of these changes should affect
any other OS.

Is this deserving of a mention in perldelta?

Yes, I think so. In the section for platform-specific notes.

Added in 810f3b7.

[p5pRT]

From [Unknown Contact. See original ticket]

On Sun Oct 03 14​:28​:06 2010, rafl@​debian.org wrote​:

"Father Chrysostomos via RT" <perlbug-followup@​perl.org> writes​:

On Fri Oct 01 13​:48​:32 2010, paulg wrote​:

The attached text files contain patches to correct build problems on
the Stratus VOS (recently renamed "OpenVOS") operating system. I have
tested these changes on OpenVOS Release 17.0, which is the
most-current customer release. None of these changes should affect
any other OS.

Is this deserving of a mention in perldelta?

Yes, I think so. In the section for platform-specific notes.

Added in 810f3b7.

[p5pRT]

From @obra

On Sun 3.Oct'10 at 14​:22​:19 -0700, Father Chrysostomos via RT wrote​:

I’m unsure as to the current policy with regard to maintenance patches.
Usually they are cherry-picked from blead. (I don’t know how to do that,
anyway. :-)

It's docced in perlpolicy.pod - this looks like a big enough chnage that
it probably doesn't qualify for backport to 5.12.

-J

I’ll let someone else take over.

[p5pRT]

From @craigberry

On Sun, Oct 3, 2010 at 8​:02 PM, Jesse Vincent <jesse@​fsck.com> wrote​:

On Sun  3.Oct'10 at 14​:22​:19 -0700, Father Chrysostomos via RT wrote​:

I’m unsure as to the current policy with regard to maintenance patches.
Usually they are cherry-picked from blead. (I don’t know how to do that,
anyway. :-)

It's docced in perlpolicy.pod - this looks like a big enough chnage that
it probably doesn't qualify for backport to 5.12.

It looked like two fairly small changes to me, both in the category of
fixing portability/build breakage and thus nominally within the
policy. Of course the pumpking's ruling is also part of the policy
and 5.14 now glimmers on the horizon, but since there was no actual
feedback on Paul's changes, I thought I'd belatedly provide some.

The one change to Socket.xs restores buildability anywhere AF_INET6 is
not defined and looks like it will leave things unchanged anywhere
that AF_INET6 is defined.

The change to Makefile.SH, while smaller, seems a bit more dangerous.
If I were to champion this change, I'd want to know why
$(MINIPERL_EXE) and miniperl both were listed as dependencies in the
all target in the first place, which should be discernible from git's
blame feature.

[p5pRT]

From @paulg1973

Craig A. Berry [mailto​:craig.a.berry@​gmail.com] wrote​:

On Sun, Oct 3, 2010 at 8​:02 PM, Jesse Vincent <jesse@​fsck.com> wrote​:

On Sun  3.Oct'10 at 14​:22​:19 -0700, Father Chrysostomos via RT wrote​:

I'm unsure as to the current policy with regard to maintenance
patches. Usually they are cherry-picked from blead.
(I don't know how to do that, anyway. :-)

Feel free to ignore the files that gave the exact patch for the back-releases, but you may have to fiddle the edits a little if you back-port from blead. I submitted patches that need no fiddling.

It's docced in perlpolicy.pod - this looks like a big enough change
that it probably doesn't qualify for backport to 5.12.

I would hope they can be back-ported. They are pretty simple changes (but see below).

It looked like two fairly small changes to me, both in the category of
fixing portability/build breakage and thus nominally within the
policy. Of course the pumpking's ruling is also part of the policy
and 5.14 now glimmers on the horizon, but since there was no actual
feedback on Paul's changes, I thought I'd belatedly provide some.

Thanks, Craig.

The one change to Socket.xs restores buildability anywhere AF_INET6 is
not defined and looks like it will leave things unchanged anywhere
that AF_INET6 is defined.

Correct.

The change to Makefile.SH, while smaller, seems a bit more dangerous.
If I were to champion this change, I'd want to know why
$(MINIPERL_EXE) and miniperl both were listed as dependencies in the
all target in the first place, which should be discernible from git's
blame feature.

Agreed.

Comment​: According to Configure, the only platforms that require an executable suffix are Stratus VOS, DJGPP, OS/2, and Cygwin. On every other platform (clearly the majority of them), this change is a no-op. I was not paying attention when this change went in, or I would have noticed that it broke our build way back when. On the 4 platforms that require an executable suffix, the effect of my patch is to retain "miniperl.pm" (VOS) or "miniperl.exe" (DJGPP, OS/2, Cygwin) as a target, while removing "miniperl" as a target. I only see one reference to a bare "miniperl" that remains in Makefile.SH, in the os/2 section (~line 694). It doesn't use an executable suffix, despite the setting of the _exe variable for OS/2 in configure. This seems inconsistent to me, and I suspect that it is either evidence that the OS/2 port has not been built in a while, or that OS/2 doesn't really care whether the suffix is used or not. I would have expected line 694 to also reference $(MINIPERL_EXE) not miniperl.

I'd be grateful if the maintainer(s) of the DJGPP, OS/2 and Cygwin ports would build perl with this patch in place and report on its effect. I do not think it will have any effect on them, but life is complicated and a simple Configure/Build/Test cycle would prove or disprove my proposed change. At the very least, it would indicate whether perl still builds on those platforms, with or without these changes.

If these ports are no longer maintained, then the question is moot.

Thanks
PG

[p5pRT]

From @doughera88

On Mon, 4 Oct 2010, Craig A. Berry wrote​:

On Sun, Oct 3, 2010 at 8​:02 PM, Jesse Vincent <jesse@​fsck.com> wrote​:

On Sun  3.Oct'10 at 14​:22​:19 -0700, Father Chrysostomos via RT wrote​:

I?m unsure as to the current policy with regard to maintenance patches.
Usually they are cherry-picked from blead. (I don?t know how to do that,
anyway. :-)

It's docced in perlpolicy.pod - this looks like a big enough chnage that
it probably doesn't qualify for backport to 5.12.

That's a good, sensible, and safe first pass assessment.

But, as Craig sensibly points out, upon more careful examination, this is
the sort of thing that perhaps should be applied. It's on my To-Do list
to take a look.

It looked like two fairly small changes to me, both in the category of
fixing portability/build breakage and thus nominally within the
policy. Of course the pumpking's ruling is also part of the policy
and 5.14 now glimmers on the horizon, but since there was no actual
feedback on Paul's changes, I thought I'd belatedly provide some.

The one change to Socket.xs restores buildability anywhere AF_INET6 is
not defined and looks like it will leave things unchanged anywhere
that AF_INET6 is defined.

The change to Makefile.SH, while smaller, seems a bit more dangerous.
If I were to champion this change, I'd want to know why
$(MINIPERL_EXE) and miniperl both were listed as dependencies in the
all target in the first place, which should be discernible from git's
blame feature.

Anyone fluent with git care to enlighten me how to do that efficiently?

--
  Andy Dougherty doughera@​lafayette.edu