Re: bk commit into 5.0 tree (tnurnberg:1.2504) BUG#10776

[Stewart Smith]

On Thu, 2007-08-09 at 17:17 +0200, Joerg Bruehe wrote:
> Hi Stewart !
> 
> Stewart Smith wrote:
> > On Wed, 2007-08-01 at 11:50 +0200, Joerg Bruehe wrote:
> >> All your changeset comments refer to AIX 5.2,
> >> but the file changes affect all AIX versions.
> > 
> > For NDB, we shouldn't care about:
> > a) AIX
> > b) AIX < 5
> > 
> > I'd prefer a, but will fight to the death for B.
> 
> I have no intention for you to die, but you better forget a):
> AFAIK, there are talks with some customer who wants MySQL AB to build 
> for AIX 5.3 including NDB.

Until we have autotest machines for a platform, I'll fight for a)

otherwise it's about as good as releasing the product untested.

Of course, mine is a sensible world view... so has little relation to
what platforms end up being "supported".

> > 
> >>> diff -Nrup a/config/ac-macros/large_file.m4
> b/config/ac-macros/large_file.m4
> >>> --- a/config/ac-macros/large_file.m4	2004-12-04 19:48:50 +01:00
> >>> +++ b/config/ac-macros/large_file.m4	2007-08-01 04:44:54 +02:00
> >>> @@ -127,14 +127,9 @@ AC_DEFUN([MYSQL_SYS_LARGEFILE],
> >>>  	hpux10.[2-9][0-9]* | hpux1[1-9]* | hpux[2-9][0-9]*)
> >>>  	  ac_cv_sys_largefile_source=1 ;;
> >>>  	esac])
> >>> -     AC_SYS_LARGEFILE_MACRO_VALUE(_LARGE_FILES,
> >>> -       ac_cv_sys_large_files,
> >>> -       [Large files support on AIX-style hosts.],
> >>> -       [case "$host_os" in
> >>> -	# AIX 4.2 and later
> >>> -	aix4.[2-9]* | aix4.1[0-9]* | aix[5-9].* | aix[1-9][0-9]*)
> >>> -	  ac_cv_sys_large_files=1 ;;
> >>> -	esac])
> >>> +
> >>> +     # AIX 4.2 and later -- do nothing, include standards.h instead.
> >>> +     # this goes for both GNU and IBM C and C++ compilers.
> >> The deleted lines take effect starting from AIX 4.1.
> >> We do not build our 5.0 (and up) on any AIX 4, so it is not important 
> >> for MySQL AB - but we do not test build there, so it is possible your 
> >> change breaks something. Or did you test on our "aix43" host ?
> > 
> > AIX 4.1.5 End of Service: Mar 31, 1999
> > 
> > Totally dead. Not just a little bit dead, way totally dead.
> > 
> > AIX 4.3 End of Service: Dec 32, 2001
> > AIX 4.3.3 (a limited fee based extension) End Of Service: Dec 31, 2003
> > 
> > Dead.
> > 
> > Looks like we may be stuck with 5L for a while yet though...
> 
> Yes, AIX 5 is still current, and AIX 4 is dead.
> But your remarks are beside the point:
> 
> AFAIK, we have no machine running AIX 4.2 (only 4.3 and 5.2 currently, 
> and we do not build our 5.0 on AIX 4.3), so I have my doubts whether the 
> code was tested on AIX 4.2 like the comment implies.

Either way.... although if configure magic comes from reliable
sources... personally I prefer these types of things to be as similar as
possible to the configure magic commonly used by others... 

> We build on/for AIX 5.2, that is a supported platform for MySQL 5.*
> We even still support AIX 4.3 with our MySQL 4.* families.

We shouldn't.

> If you want to change that (not that I would object), discuss it with 
> Support, Sales, and Product Management.

We've decided to drop NDB on AIX now.

The fact that this bug was open for 2 years without anybody complaining
is surely a sign that both users of NDB on AIX don't exist :)

> 
> > 
> >>>     fi
> >>>    ])
> >>>  
> >>> diff -Nrup a/configure.in b/configure.in
> >>> --- a/configure.in	2007-06-21 18:28:48 +02:00
> >>> +++ b/configure.in	2007-08-01 04:44:54 +02:00
> >>> @@ -2890,6 +2890,12 @@ then
> >>>    ndb_opt_subdirs="$ndb_opt_subdirs docs"
> >>>    ndb_bin_am_ldflags=""
> >>>  fi
> >>> +# building dynamic breaks on AIX. (If you want to try it and get
> unresolved
> >>> +# __vec__delete2 and some such, try linking against libhC.)
> >>> +case "$host_os" in
> >>> +  aix3.* | aix4.0.* | aix4.1.*) ;;
> >>> +  *) ndb_bin_am_ldflags="-static";;
> >>> +esac
> >> Is it worth caring about these old versions ?
> >> We don't have them in our environment, AFAIK.
> > 
> > Quite possibly... not too fussed.
> 
> Now here, handling for AIX 4 gets added.
> Relating this to the above changes, to me that seems inconsistent.
> Also, it may cause the impression we care about these AIX versions, and 
> I would rather not raise false hopes of the few who still run them.

either way... although I prefer anything that's closer to common use.
i.e. if this is the standard autoconf foo for "if aix", then keep it.

> > 
> >>>  
> >>> [[...]]
> >>> diff -Nrup a/ndb/test/ndbapi/benchronja.cpp
> b/ndb/test/ndbapi/benchronja.cpp
> >>> +#define NDB_MAXTHREADS 256
> >>> [[...]]
> >>> diff -Nrup a/ndb/test/ndbapi/flexAsynch.cpp
> b/ndb/test/ndbapi/flexAsynch.cpp
> >>> +#define NDB_MAXTHREADS 128
> [[...]]
> >>> [[...]]
> >>> diff -Nrup a/ndb/test/odbc/SQL99_test/SQL99_test.cpp
> b/ndb/test/odbc/SQL99_test/SQL99_test.cpp
> >>> +#define NDB_MAXTHREADS		24
> >>> [[...]]
> >> I find it somewhat confusing to see wildly varying values of 
> >> NDB_MAXTHREADS at different places, but maybe there are convincing reasons.
> > 
> > Patches welcome :)
> 
> I know Tatjana only renamed the symbols, to avoid the name conflict.
> As long as I do not know why this value varies from 24 to 256, I cannot 
> comment in detail, let alone propose a patch.

you have the source :) Probably as likely to find out the true meaning
as anybody else... quite possible that nobody's touched some of those
for a few years at least.

Symbol rename alone is fine for me.


I'm okay on this patch as-is. Although don't have a problem with
removing the references to AIX < 5 - as long as this doesn't deviate
from standard autoconf foo for AIX.

-- 
Stewart Smith, Senior Software Engineer
MySQL AB, www.mysql.com
Office: [phone] Ext: 6616
VoIP: 6616@stripped
Mobile: [phone]

Jumpstart your cluster:
http://www.mysql.com/consulting/packaged/cluster.html

Attachment: [application/pgp-signature] This is a digitally signed message part signature.asc

Attachment: [application/pgp-signature]