bugs in 07.03.200 - Mailing list pgsql-odbc

Hi, everybody:
Now that psotgres 7.4 is out, I decided to move to the next
driver version as well.
Unfortunately, I found that my earlier postings regarding
some enhancements had not found way into the source code.

(26.08.03: fixes for psqlodbc-07.03.0100 - part 1 and 2;
  changes:
    - EINTR safety for send/recv,
    - honour ODBCINI env var,
    - init vars before using them (most prominently in
      SQLGetPrivateProfileString calls),
    - avoid those notorious core dumps when logging on hitting
      "null" data...
)

I must have done something wrong submitting them (needed to
sign in; post to developers; who knows...)... well...

After downloading the new driver source I (again) merged
all my fixes/changes into it.
When trying to run it, however, I had to realize that it
would not work, hangs on mutex_lock.
This has been pointed out by others already; see: Eric Hallander,
New 0200 ODBC Driver issue?? et al.
However, the issiue, IHMO, is way more complicated than what
it is presented by him. His fix is another bug, because it
just kills the lock info (thus of cause enabling a new lock process)
Anyway, this is not what you should do, one sould wait for the
lock to be released. Of cause, in this case it will never happen.
This is because the driver is buggy. The whole "thread environment"
setup seems to be somewhat doubtful. (of cause, all this will happen
only when runnig threaded; "configure --enable-pthreads"...):
1) libiodbc will, when requested, shell every driver call w/i a
   lock; therfore the driver - in principle - only needs to safe-
   guard it's own data; just like it did before (e.g. conn. array)
   (I dont knwo about unixodbc...)
2) the PGAPI_x calls in the new driver all seem to be of the flavour
   ENTER_CONN_CS
   do_internal_call
   LEAVE_CONN_CS
   now; also certain internal funcs are shelled (CC_send_query e.g.)
3) as SQLDriverConnect, e.g., via PGAPI_DriverConnect calls CC_connect,
   which itself calls CC_send_query to send an initial empty query to
   test connection/backend response, we have a nice litte self-made,
   unavoidable dead-lock here. Hm...
4) I just hope, the way these ENTER_x_CS calls are spread over the
   source code now, that there isn't more of this.
5) I should say, THIS HAS NEVER BEEN TESTED ("configure --enable-pthreads").

As I mentioned above, I have merged my earlier fixes/updates into
07.03.200 and will now work on this issue (as just breaking the lock
is no solution). Perhaps disabling all this (except driver-data-safeguarding)
will be the 1st step (as I said you can bring libiodbc to shell every
driver call).

P.S.: Mon pessimisme va jusqu'à suspecter la sincérité des pessimistes.
        - Jean Rostand (1894-1977), Journal d'un caractère, 1931

--
Hartmut "Hardy" Raschick / Dept. t2
ke Kommunikations-Elektronik GmbH
Wohlenberstr. 3, 30179 Hannover
Phone: ++49 (0)511 6747-564
Fax: ++49 (0)511 6747-340
e-Mail: hartmut.raschick@ke-elektronik.de
http://www.ke-elektronik.de