NetBSD 6.1 (and 5.1 being checked)

Note that these changes probably have to be carried forward into the MariaDB 10 series.

Can you clarify the purpose of (intptr_t) casting in threadpool_unix code? I do not have NetBSD here thus I ask.

I'm asking about it, because udata, the last parameter of EV_SET macro is listed as void *

• in OpenBSD documentation http://www.manualpages.de/OpenBSD/OpenBSD-5.0/man2/EV_SET.2.html,
• in OSX docs https://developer.apple.com/library/mac/documentation/Darwin/Reference/ManPages/man2/EV_SET.2.html ,
• in FreeBSD online man pages http://www.freebsd.org/cgi/man.cgi?query=kevent&apropos=0&sektion=0&manpath=FreeBSD+9.1-RELEASE&arch=default&format=html

The only relevant place I have found that mentions NetBSD being different is here
http://ftp.ics.uci.edu/pub/centos0/ics-custom-build/BUILD/dovecot-1.1.7/src/lib/ioloop-kqueue.c, in this piece of code

/* kevent.udata's type just has to be different in NetBSD than in
FreeBSD and OpenBSD.. */
#ifdef _NetBSD_

1. define MY_EV_SET(a, b, c, d, e, f, g) \
   EV_SET(a, b, c, d, e, f, (intptr_t)g)
   #else
2. define MY_EV_SET(a, b, c, d, e, f, g) \
   EV_SET(a, b, c, d, e, f, g)
   #endif

If this is lke the comment says I'd rather port the above hack from ioloop-kqueue.c, than implement casting void* as inrteger types elsewhere

NetBSD (5.1, 6.1) appears to allow for udata to be either an int or a pointer. From /usr/include/sys/event.h

#define EV_SET(kevp, a, b, c, d, e, f) \
do

while (/* CONSTCOND */ 0)

struct kevent {
uintptr_t ident; /* identifier for this event */
uint32_t filter; /* filter for event */
uint32_t flags; /* action flags for kqueue */
uint32_t fflags; /* filter flag value */
int64_t data; /* filter data value */
intptr_t udata; /* opaque user data identifier */
};

If you create you replace the EV_SET macro, be sure to #undef the previous one.

Removing one of the casts and compiling causes this compile error; from NetBSD 5.1 - the 6.1 is at home and unreachable just now, but had similar build problem:

[ 58%] Building CXX object sql/CMakeFiles/sql.dir/threadpool_unix.cc.o
/usr/local/src/mariadb-5.5.30/sql/item.h: In constructor 'Item_direct_ref_to_ident::Item_direct_ref_to_ident(Item_ident*)':
/usr/local/src/mariadb-5.5.30/sql/item.h:3008: warning: dereferencing type-punned pointer will break strict-aliasing rules
/usr/local/src/mariadb-5.5.30/sql/item.h:3012: warning: dereferencing type-punned pointer will break strict-aliasing rules
/usr/local/src/mariadb-5.5.30/sql/gcalc_slicescan.h: In member function 'Gcalc_scan_iterator::point** Gcalc_scan_iterator::point::next_ptr()':
/usr/local/src/mariadb-5.5.30/sql/gcalc_slicescan.h:400: warning: dereferencing type-punned pointer will break strict-aliasing rules
/usr/local/src/mariadb-5.5.30/sql/threadpool_unix.cc: In function 'int io_poll_start_read(int, int, void*)':
/usr/local/src/mariadb-5.5.30/sql/threadpool_unix.cc:297: error: invalid conversion from 'void*' to '__intptr_t'
/usr/local/src/mariadb-5.5.30/sql/threadpool_unix.cc: In function 'int io_poll_disassociate_fd(int, int)':
/usr/local/src/mariadb-5.5.30/sql/threadpool_unix.cc:315: warning: converting to non-pointer type 'int' from NULL

Thanks. I checked I have to fix it the same way as it was done in my previous comment in ioloop-keueue.c source. If I would just apply your patch, compilation would fail on every BSD except NetBSD

Ok, now I have a question about _func_ in your patch. What was the reason the code is there? #if _STDC_VERSION_ < 199901L would be true? IS this a special compiler, or compiler setting?

Without the fix I get a warning for every compiled .c like:

[ 4%] Building C object strings/CMakeFiles/strings.dir/bchange.c.o
In file included from /usr/local/src/mariadb-5.5.30/strings/strings_def.h:20,
from /usr/local/src/mariadb-5.5.30/strings/bchange.c:39:
/usr/local/src/mariadb-5.5.30/include/my_global.h:1441:1: warning: "_func_" redefined
In file included from /usr/include/pthread.h:35,
from /usr/local/src/mariadb-5.5.30/include/my_global.h:285,
from /usr/local/src/mariadb-5.5.30/strings/strings_def.h:20,
from /usr/local/src/mariadb-5.5.30/strings/bchange.c:39:
/usr/include/sys/cdefs.h:248:1: warning: this is the location of the previous definition

As I recall _func_ came in ANSI C90, but you have

#if _STDC_VERSION_ < 199901L

Don't you mean

#if _STDC_VERSION_ >= 199901L

??

I greped for _STDC_VERSION_ and there are several references, but no #define. Checking the GCC predefines shown below I don't see the macro defined:

gcc -O0 -g -Wall -Wno-char-subscripts -m32 -I./../../include -dM -E -xc /dev/null
#define _DBL_MIN_EXP_ (-1021)
#define _FLT_MIN_ 1.17549435e-38F
#define _CHAR_BIT_ 8
#define _WCHAR_MAX_ 2147483647
#define _DBL_DENORM_MIN_ 4.9406564584124654e-324
#define _FLT_EVAL_METHOD_ 2
#define _DBL_MIN_10_EXP_ (-307)
#define _FINITE_MATH_ONLY_ 0
#define _GNUC_PATCHLEVEL_ 3
#define _SHRT_MAX_ 32767
#define _LDBL_MAX_ 1.18973149535723176502e+4932L
#define _UINTMAX_TYPE_ long long unsigned int
#define _LDBL_MAX_EXP_ 16384
#define _SCHAR_MAX_ 127
#define _USER_LABEL_PREFIX_
#define _STDC_HOSTED_ 1
#define _LDBL_HAS_INFINITY_ 1
#define _DBL_DIG_ 15
#define _FLT_EPSILON_ 1.19209290e-7F
#define _NetBSD_ 1
#define _LDBL_MIN_ 3.36210314311209350626e-4932L
#define _unix_ 1
#define _DECIMAL_DIG_ 21
#define _LDBL_HAS_QUIET_NAN_ 1
#define _GNUC_ 4
#define _DBL_MAX_ 1.7976931348623157e+308
#define _DBL_HAS_INFINITY_ 1
#define _DBL_MAX_EXP_ 1024
#define _LONG_LONG_MAX_ 9223372036854775807LL
#define __GXX_ABI_VERSION 1002
#define _FLT_MIN_EXP_ (-125)
#define _DBL_MIN_ 2.2250738585072014e-308
#define _DBL_HAS_QUIET_NAN_ 1
#define _tune_i386_ 1
#define _REGISTER_PREFIX_
#define _NO_INLINE_ 1
#define __i386 1
#define _FLT_MANT_DIG_ 24
#define _VERSION_ "4.1.3 20080704 prerelease (NetBSD nb2 20081120)"
#define i386 1
#define _i386_ 1
#define _SIZE_TYPE_ unsigned int
#define _ELF_ 1
#define _FLT_RADIX_ 2
#define _LDBL_EPSILON_ 1.08420217248550443401e-19L
#define _LDBL_DIG_ 18
#define _FLT_HAS_QUIET_NAN_ 1
#define _FLT_MAX_10_EXP_ 38
#define _LONG_MAX_ 2147483647L
#define _FLT_HAS_INFINITY_ 1
#define _LDBL_MANT_DIG_ 64
#define _WCHAR_TYPE_ int
#define _FLT_DIG_ 6
#define _INT_MAX_ 2147483647
#define _FLT_MAX_EXP_ 128
#define _DBL_MANT_DIG_ 53
#define _WINT_TYPE_ int
#define _LDBL_MIN_EXP_ (-16381)
#define _LDBL_MAX_10_EXP_ 4932
#define _DBL_EPSILON_ 2.2204460492503131e-16
#define _INTMAX_MAX_ 9223372036854775807LL
#define _FLT_DENORM_MIN_ 1.40129846e-45F
#define _FLT_MAX_ 3.40282347e+38F
#define _FLT_MIN_10_EXP_ (-37)
#define _INTMAX_TYPE_ long long int
#define _GNUC_MINOR_ 1
#define _DBL_MAX_10_EXP_ 308
#define _LDBL_DENORM_MIN_ 3.64519953188247460253e-4951L
#define _STDC_ 1
#define _PTRDIFF_TYPE_ int
#define _LDBL_MIN_10_EXP_ (-4931)
#define _GNUC_GNU_INLINE_ 1

From /usr/include/sys/cdefs.h

/*

• C99 defines _func_ predefined identifier, which was made available
• in GCC 2.95.
  */
  #if !(_STDC_VERSION_ >= 199901L)
  #if _GNUC_PREREQ_(2, 6)
  #define _func_ _PRETTY_FUNCTION_
  #elif _GNUC_PREREQ_(2, 4)
  #define _func_ _FUNCTION_
  #else
  #define _func_ ""
  #endif
  #endif /* !(_STDC_VERSION_ >= 199901L) */

" As I recall _func_ came in ANSI C90, but you have

#if _STDC_VERSION_ < 199901L

Don't you mean

#if _STDC_VERSION_ >= 199901L

??"

Hmm,no, not really. _func_ is a C99 feature ,which is what _STDC_VERSION_ tests is. But ok, your GCC compiler does not announce itself as C99, but system headers compensate for that. Ok, thanks for the info.

Hi,
could you lease check if patch here http://bazaar.launchpad.net/~maria-captains/maria/5.5-mdev4553/revision/3779 solves compilation problems (it should, but I'd prefer someone with readl NetBSD machine test it .

The _func_ patch doesn't appear to work (NetBSD 6.1, gcc 4.5.3), many of these warnings...

Scanning dependencies of target strings
[ 4%] Building C object strings/CMakeFiles/strings.dir/bchange.c.o
In file included from /usr/local/src/mariadb-5.5.30/strings/strings_def.h:20:0,
from /usr/local/src/mariadb-5.5.30/strings/bchange.c:39:
/usr/local/src/mariadb-5.5.30/include/my_global.h:1453:0: warning: "_func_" redefined
/usr/include/sys/cdefs.h:353:0: note: this is the location of the previous definition
[ 4%] Building C object strings/CMakeFiles/strings.dir/bmove_upp.c.o
In file included from /usr/local/src/mariadb-5.5.30/strings/strings_def.h:20:0,
from /usr/local/src/mariadb-5.5.30/strings/bmove_upp.c:38:
/usr/local/src/mariadb-5.5.30/include/my_global.h:1453:0: warning: "_func_" redefined
/usr/include/sys/cdefs.h:353:0: note: this is the location of the previous definition
[ 4%] Building C object strings/CMakeFiles/strings.dir/ctype-big5.c.o
In file included from /usr/local/src/mariadb-5.5.30/strings/strings_def.h:20:0,
from /usr/local/src/mariadb-5.5.30/strings/ctype-big5.c:30:
/usr/local/src/mariadb-5.5.30/include/my_global.h:1453:0: warning: "_func_" redefined
/usr/include/sys/cdefs.h:353:0: note: this is the location of the previous definition
[ 4%] Building C object strings/CMakeFiles/strings.dir/ctype-bin.c.o
In file included from /usr/local/src/mariadb-5.5.30/strings/strings_def.h:20:0,
from /usr/local/src/mariadb-5.5.30/strings/ctype-bin.c:22:
/usr/local/src/mariadb-5.5.30/include/my_global.h:1453:0: warning: "_func_" redefined
/usr/include/sys/cdefs.h:353:0: note: this is the location of the previous definition

My suggested fix was better.

I might overlook something trivial, but how

#if (_STDC_VERSION_ < 199901L) && !defined(_func_)
as in my patch is the very same as

#if _STDC_VERSION_ < 199901L

1. ifndef _func_

as in your patch? Could you please clarify?

Thanks,
Vladislav

You version falls through to the #else line 1453, whereas my version enters the #if _STDC_VERSION_ < 199901L section before applying #ifndef _func_.

Alternatively should work too:

#ifndef _func_
#if (_STDC_VERSION_ < 199901L)

1. if _GNUC_ >= 2
2. define _func_ _FUNCTION_
3. else
4. define _func_ "<unknown>"
5. endif
   #elif defined(_MSC_VER)
6. if _MSC_VER < 1300
7. define _func_ "<unknown>"
8. else
9. define _func_ _FUNCTION_
10. endif
    #elif defined(_BORLANDC_)
11. define _func_ _FUNC_
    #else
12. define _func_ "<unknown>"
    #endif
    #endif /* _func_ */

Yes, I like this most.