Re: Re: FreeExpr called on live expr
- To: mathgroup at smc.vnet.net
- Subject: [mg6370] [mg6370] Re: [mg6351] Re: [mg6328] FreeExpr called on live expr
- From: David Withoff <withoff>
- Date: Fri, 14 Mar 1997 14:54:00 -0500 (EST)
- Sender: owner-wri-mathgroup at wolfram.com
> >The message "FreeExpr called on live expr" should never come up. > >This message means that the Mathematica kernel is releasing memory > >that is still being used. When you see this message, the Mathematica > >kernel will normally crash. > > > >We have not seen this problem on any computers other than SGI > >computers. Even on SGI computers, this problem is very rare. > >It seems to come up only in large programs that use a lot of > >memory and that take a long time to run. > > > >Dave Withoff > >Wolfram Research > > I must report that last year I ran into a "FreeExpr called on live expr" > error when using Mathematica 2.2 on a Sun machine. It was a very > large algebraic calculation. I did not keep a record of the code > that caused the error. In the message above, which I wrote in response to an individual technical support question, I was referring specifically to Mathematica Version 3.0. Since this message is now going to a wider audience, I should clarify this by saying that we have not seen this behavior while using Mathematica Version 3.0 on anything other than an SGI. If you find a repeatable example of this behavior while using Mathematica Version 3.0 on something other than an SGI, we would very much like to know about it. There is a note in the technical support section of the Wolfram Research web site which describes what we know about this problem, and which will in the future contain any additional information that we discover about it. http://www.wolfram.com/support/Systems/Unix/SGI/FreeExpr.html Regarding Version 2.2, although this problem was more widespread, I was not aware even in Version 2.2 of this problem ever coming up on a Sun machine. Perhaps it came up, but no one ever found a reportable example. It is worth pointing out that at least some instances of this problem may have little or nothing to do with Mathematica, but may instead be related to defects outside of Mathematica. The behavior of operating systems when a computer is about to run out of memory, and memory allocation in general, are notoriously fragile operations. It is entirely possible, especially in calculations that will exhaust the memory of the computer, that Mathematica will expose a defect somewhere else in the computer, and that the outcome in Mathematica is merely the visible manifestation of that defect. The problems of this type that we have been able to isolate are not in the Mathematica source code, but are instead in compilers and other tools that are used to produce particular versions of Mathematica. Dave Withoff Wolfram Research