From: Alan van den Bosch
Date: October 31 2000 1:11pm
Subject: Re: ha_berkeley.[ch] files
List-Archive: http://lists.mysql.com/internals/86
Message-Id: <00110100113408.13992@odin>
MIME-Version: 1.0
Content-Type: text/plain
Content-Transfer-Encoding: quoted-printable

On Tue, 31 Oct 2000, you wrote:
>On Tue, Oct 31, 2000 at 02:32:15PM +1100, Alan van den Bosch wrote:
>> I Downloaded the src tarballs for mysql-3.23.27-beta-win-src.zip and
>> db-3.2.3b.tar.gz had had a poke around slq/ha_berkeley.[ch] and the db=
=2Eh
>> files. I've the impression that transaction support has been successfu=
lly
>> compiled and tested to a degree. However, there are references to a me=
mber
>> of the DB type (__DB struct) that doesnt exist in that structure, but =
is a
>> member of the DBT type ( does app_private ring any bells? ). An expamp=
le
>> of this is the berkley_cmp_packed_key() function in ha_berkeley.c
>
>There's an app_private in __db, __db_env and __db_dbt.  We
>haven't done any work on getting MySQL on Windows to build
>with BDB, though, so ... I have no idea what troubles you'll
>run into there.  We will get to that ... sometime.  Monty is
>on a trip to the US right now, and probably he's the one who
>will need to handle it.  Unless you do first!
>
>What exactly is happening?
>
>Tim

Hi Tim,

I managed to get mysql-opt built statically linked with libdb32s.lib usin=
g=20
MSVC++ 6.0 with sp3 on win95b. Have done a quick test (single user/single=
=20
transaction) and it seems to work - after I remembered to set autocommit =
off.=20
Well the server doesnt fall over anyway :-)

The db-3.2.3b tarball includes a few different directories specific to=20
building on different platforms (win32,vms...) and the db.h file in the=20
build_win32 directory is from sleepycat ver 2 -> app_private only in=20
__db_dbt. I added app_private to __db and __db_env by hand. I cant see a =
*nix=20
version of db.h in the tarball either so I assume that the first db.h alr=
eady=20
on a *nix system the compiler comes accross will be used. This tarball ma=
y=20
not be enough by itself to get db-3 happening on *nix :-(

Some other issues I came across;

1) both mysql (mysys.lib) and libdb32s.lib have 'getopt' implemented. I t=
hrew=20
a #ifndef HAVE_GETOPT ... #endif around most of the contents of=20
db-3.2.3b/clib/getopt.c, starting from just below where db_config.h is=20
included (was tempting to just remove it from the project, which prolly w=
ould=20
have been easier, but HAVE_GETOPT is mentioned in db_config.h already for=
=20
what its worth )

2) the use of the type 'byte' in ha_berkeley.[ch] caused MSVC++ to barf. =
I=20
went sick with type casting arguments till the compiler was content - hop=
e I=20
havent broken too much ;-)

3) db-3.2.3b/os_win32/os_rename.c uses MoveFileEx(), which is not availab=
le=20
on win95/98 - they have also used a hack to get around a bug in this func=
tion=20
that occurs if the destination file file allready exists which involves=20
calling MoveFile() if MoveFileEx() fails. The files comments explain it=20
nicely. I replaced this with a call to DeleteFile() then a call to=20
MoveFile(). DeleteFile() will close the file before deleting it if needed=
=20
(alegedly).

Thankyou for your assistance and I hope this is of some help.

--=20
Regards,

Alan van den Bosch
Sanguis Pty Ltd


/* All generalizations are false. */