Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!swrinde!elroy.jpl.nasa.gov!turnkey!orchard.la.locus.com!devnet.la.locus.com!richp
From: richp@locus.com (Richard L. Pettit Jr.)
Newsgroups: comp.windows.x.motif
Subject: Re: Vanishing Server Resources with Motif1.0 Application
Message-ID: <1991Jun27.144542.2521257@locus.com>
Date: 27 Jun 91 14:45:42 GMT
Reply-To: richp@locus.com
Organization: Locus Computing Corporation, Los Angeles, California
Lines: 42

Clifford Luke sez:
>                                                  He is running Motif1.0
>and an R3 based Server.
>
>There have been reports of "memory leaks" in earlier versions of Motif,
>due to memory not being freed when widgets are destroyed, leading to
>clients which "grow in size".
>
>Has anybody observed similar problems on the server side caused by Motif
>not destroying widgets correctly (ie. server resources not being released)?
>If so will updating to Motif 1.1.2 alleviate these problems?

There are 3 problems at work here (that I know of in Motif 1.0 & X11R3).

1) If backing store is on, server memory vanishes.
   Start the server with -bs to defeat this.

2) If Mwm*saveUnder is true you will be in deep poop in no time flat.
   Make it false.

3) If the memory allocator in the X server is first-fit not best-fit
   then uglyisms like this happen:

   p = Xalloc(7000000);  /* not enough memory, sbrk() */
   Xfree(p);             /* return to pool */
   q = Xalloc(1000);     /* 1000 < 7000000 use this chunk and leave 6999000 */
   p = Xalloc(7000000);  /* not enough memory, sbrk() */

Really, 3 is the big problem and 1 and 2 simply give the server copious
opportunity for it to happen.  Life with X has become correct since I
started the X -bs and Mwm*saveUnder: false gig.

There may also be a problem in XmStringFree() not freeing all of the
resources allocated by XmStringCreate*().  Scribble a test program that
spins in a loop allocating and freeing and see if the process grows.

Rich
-- 
----
              Richard Pettit                       Locus Computing Corp.
                                richp@locus.com
        "Opinions expressed herein are of the author, not (LCC or IBM)"