Dear Martin,
Thanks for the input. After serching fora for similar problems I found some
stating issues between Solaris 10 include files and gcc compatibility.
In fact in /usr/include/iso/math_c99.h Solaris defines:
#define HUGE_VAL __builtin_huge_val
Now, this may be fine for their own compiler, but clashes with gcc. gcc
expects the following:
#define HUGE_VAL __builtin_huge_val()
By making this change in /usr/include/iso/math_c99.h the CVS compiles. Since
I am not using sun's compiler I made the change.
-or- even better
The difference between CVS and 1.4.1 lies in the compiler definitions. CVS
defines _STDC_C99 and ends up using /usr/include/iso/math_c99.h's broken
definition, whereas 1.4.1 doesn't define it and uses
/usr/include/iso/math_iso.h's definition that works:
#define HUGE_VAL __huge_val._d
Both includes are first referenced in math.h and are applied conditionally
based on that flag. Anybody knows what is the purpose of _STDC_C99 and why
did it start been used in CVS? It could be that in Solaris it exists solely
to flag Sun's compiler.
BR,
Nikos
----- Original Message -----
From: "Nikos Balkanas" <[EMAIL PROTECTED]>
To: "Martin Conte Mac Donell" <[EMAIL PROTECTED]>
Cc: <[email protected]>
Sent: Wednesday, October 29, 2008 5:35 AM
Subject: Re: Problem with CVS source
Dear Martin,
gcc is the standard stock 64bit gcc shipped with Solaris 10.
Server2:~-> gcc -v
Reading specs from /usr/sfw/lib/gcc/i386-pc-solaris2.10/3.4.3/specs
Configured with:
/builds/sfw10-gate/usr/src/cmd/gcc/gcc-3.4.3/configure --prefix=/usr/sfw --with-as=/usr/sfw/bin/gas
--with-gnu-as --with-ld=/usr/ccs/bin/ld --without-gnu-ld --enable-languages=c,c++
--enable-shared
Thread model: posix
gcc version 3.4.3 (csl-sol210-3_4-branch+sol_rpath)
It doesn't seem to be a gcc issue, inasmuch the same exactly source code
compiles without a problem with gateway-1.4.1. Includes are also identical
upto 1 level below (cannot follow the whole tree).
You have a better acquaintance with the code at that point. I don't
disagree that a correct solution is needed, but in a 64-bit compilation
what is the use of cheking the limits on a 32bit float? Of course you have
the same problem with a 32bit compile and this time it may be more
crucial.
Nikos
----- Original Message -----
From: "Martin Conte Mac Donell" <[EMAIL PROTECTED]>
To: "Nikos Balkanas" <[EMAIL PROTECTED]>
Cc: <[email protected]>
Sent: Wednesday, October 29, 2008 5:04 AM
Subject: Re: Problem with CVS source
2008/10/28 Nikos Balkanas <[EMAIL PROTECTED]>:
Hi,
After looking at it a bit longer, I think I need some help with it. In
Solaris 10.5 the cvs is broken. The build failes in wmlscript/wlexer.c
(1037):
if (*result == *HUGE_VAL || *result == -(*HUGE_VAL)>.>..>..// NIKOS!!
|| ws_ieee754_encode_single(*result, buf) != WS_IEEE754_OK)
ws_src_error(compiler, 0, "floating point literal too large");
Are these lines really important? We are talking for a 64bit build here
(of
course the same lines will affect 32bit builds as well). HUGE_VAL is
undefined in the CVS version:
Undefined first referenced
symbol in file
__builtin_huge_val libwmlscript.a(wslexer.o)
ld: fatal: Symbol referencing errors. No output written to
wmlscript/wmlsc
This is after I commented out the relevant code in wslexer.c and filled
in
for a simple printf. Of interest is that in the CVS version HUGE_VAL is
a
undefined pointer, whereas in the 1.4.1 version it is a float(correct).
Source, includes and dependencies are exactly the same:
gateway-1.4.1/.depend:
wmlscript/wslexer.o: wmlscript/wslexer.c wmlscript/wsint.h ./gw-config.h
\
./gwlib/gwassert.h ./gwlib/log.h wmlscript/ws.h wmlscript/wsutf8.h \
wmlscript/wserror.h wmlscript/wsieee754.h wmlscript/wsbuffer.h \
wmlscript/wsencode.h wmlscript/wsalloc.h wmlscript/wsfalloc.h \
wmlscript/wsstream.h wmlscript/wshash.h wmlscript/wsbc.h \
wmlscript/wsstree.h wmlscript/wsasm.h wmlscript/wsopt.h \
wmlscript/wsstdlib.h wmlscript/wsgram.h
CVS/.depend:
wmlscript/wslexer.o wmlscript/wslexer.i: wmlscript/wslexer.c \
wmlscript/wsint.h ./gw-config.h ./gwlib/gwassert.h ./gwlib/log.h \
wmlscript/ws.h wmlscript/wsutf8.h wmlscript/wserror.h \
wmlscript/wsieee754.h wmlscript/wsbuffer.h wmlscript/wsencode.h \
wmlscript/wsalloc.h wmlscript/wsfalloc.h wmlscript/wsstream.h \
wmlscript/wshash.h wmlscript/wsbc.h wmlscript/wsstree.h \
wmlscript/wsasm.h wmlscript/wsopt.h wmlscript/wsstdlib.h \
wmlscript/wsgram.h
Where is wmlscript invoked? I have some issues right now with the
client, so
I cannot run a debugger on it.
Any help would be appreciated.
Thanks,
Nikos
----- Original Message -----
From: Nikos Balkanas
To: Stipe Tolj
Cc: [email protected]
Sent: Monday, October 27, 2008 9:43 PM
Subject: Re: Problem with CVS source
Dear Stipe,
It seems that the problem is not what I first suspected. You see the
file
wslexer.c is exactly the same in both cases (gateway-1.4.1 and cvs
gateway).
Therefore it is not a HUGE_VAL problem I first suspected.
This is strange. I am using exactly the same environment, gcc,
configuration
as with the gateway-1.4.1 version. Yet make fails on exactly the same
file.
I will have to look more into it. Meanwhile if you have seen this
before,
pls let me know.
Nikos
On Mon, Oct 27, 2008 at 9:26 PM, Nikos Balkanas <[EMAIL PROTECTED]>
wrote:
Thanks, Stipe, for the fast reply. Please see my comments inlined.
Regards,
Nikos
On Mon, Oct 27, 2008 at 8:10 PM, Stipe Tolj <[EMAIL PROTECTED]> wrote:
Nikos Balkanas schrieb:
> Hi,
>
> I am rather new to the list. I tried to download and compile the
> latest
> CVS source and I got the following error:
>
> gcc -std=gnu99 -D_REENTRANT=1 -I. -Igw -O6 -s -m64 -D_LARGE_FILES=
> -I/usr/include/libxml2 -I/usr/local/include -Wall -Wmissing-prototypes
> -Wmissing-declarations -Wnested-externs -Winline -Wformat
> -Wformat-security -Wmissing-format-attribute -Wall
> -I/usr/local/ssl/include -o wmlscript/wslexer.o -c
> wmlscript/wslexer.c
> wmlscript/wslexer.c:222: warning: no previous prototype for
> 'ws_yy_lex'
> wmlscript/wslexer.c: In function `read_float_from_exp':
> wmlscript/wslexer.c:1037: error: invalid operands to binary ==
> wmlscript/wslexer.c:1038: error: wrong type argument to unary minus
> make: *** [wmlscript/wslexer.o] Error 1
there is something wrong with your local CVS checkout tree. Make sure
you
$ make clean
$ make
To be sure I rm -rf gateway and restored from CVS again.
in order to remove any previously build objects. The configure run was
clean?
Yes. Apart from docbook support which I don't particular care (This is
a
minimal server for security/performance reasons)
CFLAGS="-O6 -s -m64 -DSunOS" CPPFLAGS="-I/usr/local/ssl/include"
configure --build=ia64-pc-solaris2.10 --libdir=/usr/local/lib/64
--enable-ssl --enable-cookies --enable-debug --enable-warnings --enable-docs
--mandir=/usr/local/man --with-ssl=/usr/local/ssl --enable-pcre
gateway-1.4.1 compiled with the same configuration with no problems.
The issue seems to be with HUGE_VAL and Solaris. I don't have a linux
in
front of me to check, but in solaris 10.5 in
/usr/include/iso/math_iso.h:
#define HUGE_VAL __huge_val._d
HUGE_VAL is a standard C feature. And __builtin_huge_val() is a
GCC-specific extension for implementing this one.
gcc had some troubles with math.h in solaris.
Which gcc version are you using?
M
