Conference azur::mcc

1600.0. "trouble linking with nDBm and MIR rtns" by RIVAGE::BACCAR () Fri Oct 04 1991 11:15

   I'm writting an application on the top of nDBm package, and because the MIR
is also built on the top of nDBm, I thinck that my application could interfere
with the MIR. In others word, I protect the nDBm routines from thread re-entrance
using a CMA mutex ( jacket package), and from concurrent process write access 
using the ULTRIX flock system call ( and I thinck that the MIR do the same 
thing ). Then we share the same nDBm routines with two different mutexes, AIE..
( bad in french ..).  Is there a way to avoid this conflict ?
Following are suggestions:
  1 - : I use the MIR mutex to mutex-lock my routines before calling to
nDBm routines.
  2 - : I link my program with an nDBm protected routines (the same routines
as used to build the MIR routines)
 
note: I used nDBm and the MIR ( in the same process) in my application.

    Aucun probleme.
    
    All non-reentrant operating system routines are protected with the SAME
    mutex, which is the one hidden behind mcc_mutex_lock_global
    (cma_mutex_lock_global).   That's why the global mutex was invented.
    Do not use a private mutex to protect OS routines because it won't
    work, as your note correctly deduces.  Bonne chance.
    

    And as an aside, aren't you (we) glad they invented ndbm?   With the old dbm
    routines, you couldn't open more than one DBM file at a time PER
    PROCESS.  Yuck.
    

  the global mutex of course ...
   thanck you for your help .