gmane.os.netbsd.devel.toolchain

Re: make: wilcard targets and .PHONY

Simon J. Gerraty — 2013-05-17T19:54:44

On Fri, 17 May 2013 17:38:25 +0000, David Holland writes:

No, just fixing a bug wrt the documented behavior.
Obviously the combination of wildcard targets and .PHONY is rare,
but there are a number of examples in the FreeBSD ports tree,
and having to use gmake rather than bmake to build them is a pitty.

Re: make: wilcard targets and .PHONY

David Holland — 2013-05-17T17:38:25

 > Back in 2000, the callpath through Suff_FindDeps was skipped for .PHONY
 > targets.
 > It turns out though that that is the only place where wildcard targets
 > are expanded.
 > Assuming there is something in ./sub/ the makefile:
 > 
 > .MAIN: all
 > all: subs
 > 
 > SUBS= sub/*
 > subs:${SUBS}
 > < at >echo yay ${SUBS}
 > 
 > .ifdef PHONY
 > .PHONY: subs
 > .endif
 > 
 > works so long as PHONY isn't defined.

Do you have a model in mind for wildcard expansion in make? (Not just
here, but elsewhere in the language...)

make: wilcard targets and .PHONY

Simon Gerraty — 2013-05-16T18:40:25

Back in 2000, the callpath through Suff_FindDeps was skipped for .PHONY
targets.
It turns out though that that is the only place where wildcard targets
are expanded.
Assuming there is something in ./sub/ the makefile:

.MAIN: all
all: subs

SUBS= sub/*
subs:${SUBS}
< at >echo yay ${SUBS}

.ifdef PHONY
.PHONY: subs
.endif

works so long as PHONY isn't defined.

The patch below (sans fixing the indentation) pushes the PHONY check
lower so that wildcards can be expanded correctly, while still skipping 
most of the suffix related effort.

Index: suff.c
===================================================================
RCS file: /cvsroot/src/usr.bin/make/suff.c,v
retrieving revision 1.69
diff -u -p -r1.69 suff.c
--- suff.c29 Sep 2011 23:38:04 -00001.69
+++ suff.c16 May 2013 18:34:29 -0000
< at >< at > -2058,6 +2058,10 < at >< at > SuffFindNormalDeps(GNode *gn, Lst slst)
      * children, then look for any overriding transformations they imply.
      * Should we find one, we discard the one we found before.
      */
+    bottom = NU

Missing sysinst/sysinstmsgs.*

Emmanuel Dreyfus — 2013-03-25T17:45:58

Hi

When I build netbsd-6 using build.sh -Uuom i386 release, I get an error
because of missing
src/distrib/i386/ramdisks/ramdisk-big/sysinst/sysinstmsgs.es
Other sysinstmsgs.* are missing as well.

Is there anything wrong with that build.sh usage?

Re: CVS commit: src/external/bsd/mdocml/dist

Greg Troxel — 2013-03-22T01:45:27

  [introduction of __arraycount into mdocml]

It seems __arraycount is not POSIX, and thus can't be used in tools (in
src/external/bsd/mdocml/dist/roff.c).  The man page says it appeared in
NetBSD 4, and does not mention any posix spec.  And it's not there on os
x 10.7.

Re: netbsd-6 build failure

Lloyd Parkes — 2013-03-15T22:42:35

On 16/03/2013, at 11:36 AM, Greg Troxel <gdt< at >ir.bbn.com> wrote:

What do you mean "old days"? it was just last year when I had a conversation where I was told "Oh him! He caused Black Tuesday when he made the mainframe crash twice in one day".

Cheers,
Lloyd

Re: netbsd-6 build failure

Greg Troxel — 2013-03-15T22:36:47

Edgar Fuß <ef< at >math.uni-bonn.de> writes:


The right place is releng< at >netbsd.org and perhaps current-users (we don't
have stable-users).  But, I already told releng and they have (I
believe) already fixed it.

It's quite a testament to how good our releng processes and people are
that netbsd-6 not building is almost like the old days when workers at a
company would tell each other about the day the mainframe crashed years
ago.

netbsd-6 build failure

Edgar Fuß — 2013-03-15T20:49:34

Running an unpriviledged build of a current netbsd-6 source tree fails with

======  3 missing files in DESTDIR  ========
Files in flist but missing from DESTDIR.
File wasn't installed ?
------------------------------------------
./usr/share/zoneinfo/Asia/Khandyga
./usr/share/zoneinfo/Asia/Ust-Nera
./usr/share/zoneinfo/Europe/Busingen
========  end of 3 missing files  ==========
*** [checkflist] Error code 1

I'm not sure whether this is the right place to report this.

Re: Unnoticed host-mkdep failure with binstall

Nicolas Joly — 2013-03-12T18:28:27

Thanks Christos.

Re: Unnoticed host-mkdep failure with binstall

Christos Zoulas — 2013-03-12T16:48:32

In article <20130311195417.GA12036< at >lynche.sis.pasteur.fr>,
Nicolas Joly  <njoly< at >pasteur.fr> wrote:

Fixed.

christos

Unnoticed host-mkdep failure with binstall

Nicolas Joly — 2013-03-11T19:54:17

Hi,

While trying to update my -current amd64 workstation, i luckily found
a non-fatal(?) error while early building binstall ...

dependall ===> binstall
     create  binstall/getid.d
     create  binstall/xinstall.d
In file included from /local/src/NetBSD/src/tools/binstall/../../usr.bin/xinstall/xinstall.c:77:0:
/local/src/NetBSD/src/tools/binstall/../compat/sys/sha1.h:4:27: fatal error: nbtool_config.h: No such file or directory
compilation terminated.
     create  binstall/.depend
    compile  binstall/xinstall.lo
    compile  binstall/getid.lo
       link  binstall/xinstall
xinstall.lo: In function `do_symlink':
xinstall.c:(.text+0x604): warning: warning: mktemp() possibly used unsafely,use mkstemp() or mkdtemp()

Looks like host_mkdep tool does not return a non zero status upon
error, letting build continue :

#    create  binstall/xinstall.d
/local/src/NetBSD/src/tools/host-mkdep/obj.amd64/host-mkdep -f xinstall.d  --     -I/local/src/NetBSD/src/tools/binstall/../compat/sys -DTARGET_STRIP=\"/local/s

Re: g++ -pedantic on 6.0.1/amd64 fails to include xmmintrin.h

Christos Zoulas — 2013-03-11T12:52:33

In article <20797.49528.393898.445783< at >ficus.home>,
Anthony Mallet  <anthony.mallet< at >laas.fr> wrote:

I am fine with this too :-)

christos

Re: g++ -pedantic on 6.0.1/amd64 fails to include xmmintrin.h

Anthony Mallet — 2013-03-11T11:35:20

| On Sat, Mar 09, 2013 at 04:14:56PM +0100, Anthony Mallet wrote:
| The extern C is slightly hidden in the stdlib.h declaration, but the "throw()"
| of course is a lie. Look at the start of mm_malloc.h:
| 
| #include <stdlib.h>

Just a thought: shouldn't that mm_malloc.h header include <cstdlib> in C++,
instead of stdlib.h?

Then maybe it's easier to add the throw() to posix_memalign there.


| My vote: remove the bogus declaration from the gcc header file and ignore
| the issue.

As far as I am concerned, I'm fine with this.

Re: g++ -pedantic on 6.0.1/amd64 fails to include xmmintrin.h

Martin Husemann — 2013-03-11T08:40:54

Sorry, this was unclear: I am talking about something like what Christos
suggested:

extern "C" __attribute__((nothrow)) {
 // function declarations
}


However, this does not seem to work, so we are back at header file pollution
or just ignoring the issue.

My vote: remove the bogus declaration from the gcc header file and ignore
the issue.

Martin

Re: g++ -pedantic on 6.0.1/amd64 fails to include xmmintrin.h

Martin Husemann — 2013-03-11T08:31:47

I don't see how this would be the case.

Adding (in C++) arbitrary throw declarations has this effect [which is
why sane programs never use them, and I'm saving you the "avoid
exceptions by all means" rant here for brevity], but adding throw()
[i.e. the "does not throw any C++ exception] declarations on
non-inlinable functions (like the public exports from libc) should not
cause the compiler to add any additional runtime checks nor incur any
costs.

Please provide detailed references how this excact declaration can cause
costs.

Martin

Re: g++ -pedantic on 6.0.1/amd64 fails to include xmmintrin.h

Valeriy E. Ushakov — 2013-03-11T02:53:05


As far as I understand, nothrow attribute is different and can't be
expressed in proper C++.  It says the function will not throw.  OTOH,
throw() says that compiler should make the program die (too lazy to
look up specific details in the standard) if any exception is thrown
out of that function.

-uwe

Re: g++ -pedantic on 6.0.1/amd64 fails to include xmmintrin.h

Christos Zoulas — 2013-03-11T02:20:09

In article <20130310081553.GA14441< at >mail.duskware.de>,
Martin Husemann  <martin< at >duskware.de> wrote:

I was thinking of something like:

extern "C" __attribute__((__nothrow__)) {
};

christos

Re: g++ -pedantic on 6.0.1/amd64 fails to include xmmintrin.h

Joerg Sonnenberger — 2013-03-11T01:01:11

I disagree on this. Adding throw() in many cases costs more than it
provides and I am not talking about the header pollution here.

Joerg

Re: g++ -pedantic on 6.0.1/amd64 fails to include xmmintrin.h

Anthony Mallet — 2013-03-10T13:43:43

| Scratch that - we should indeed mark most of our standard library functions
| as "nothrow" similar to the visibility and extern C stuff in cdef.h, and
| then explicitly exclude functions taking a function pointer.

For what it's worth, Linux has a __THROW after every single libc function...
So I guess this means that gcc has no way to "push" a default __nothrow__
attribute like for visibility. Depressing...

Re: g++ -pedantic on 6.0.1/amd64 fails to include xmmintrin.h

Martin Husemann — 2013-03-10T08:15:53

Linking without name mangling and exceptions are orthogonal. However, I surely
agree that adding a __nothrow at the end of every function declaration is
unlikely to happen.

My question was: like with the visibility pragama, is there a way to
apply "nothrow" to a whole block of declarations?

Gcc documentation did not tell me, and I don't even want to start thinking
about clang compatibility.

Martin

Re: g++ -pedantic on 6.0.1/amd64 fails to include xmmintrin.h

Christos Zoulas — 2013-03-09T22:56:43

In article <20130309195448.GA19234< at >mail.duskware.de>,
Martin Husemann  <martin< at >duskware.de> wrote:

I understand the motivation, but this dirties the header files because an
unfortunate language decision (why extern "C" requires an explicit "nothrow"
instead of defaulting to "nothrow" and requiring "throws" instead?). Isn't
there a better way?

christos