[Coin-lpsolver] building a model correctly

[Scott Drader]

Hi,

I'm using CLP to solve some geometry problems, but have had problems with segmentation faults. These problems need only be loaded from memory and never written to disk. I suspect my problem might be in how I'm building and loading my model, but I am unsure. The sequence of operations I perform are roughly:

1. Create bounds and objective arrays.
2. Create a CoinPackedMatrix and populate it with several appendRow() operations.
3. Create a ClpSimplex model with loadProblem(), passing in the matrix and bounds/objective arrays.
4. Do a solve with initialSolve().

I've attached a stripped-down source file demonstrating the above sequence in more detail.

With the above procedure I get consistent segmentation faults. There are however two ways that I have found will indefinitely prevent the segmentation faults.

1. Writing the problem to a file and reading it back in with writeMps()/readMps() before solving (inbetween step 3 and 4 above).
2. Disabling the presolve (this causes several of my problems to solve less than ideally however, maybe beacause of their degenerate nature).

I'm currently using the read/write fix above, but this is less than ideal since I'm doing iterative real-time solves (many many solves).

Is there an obvious problems with this means of constructing and solving a problem?

Thanks,

Scott




-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: coin_crash_example.cpp
URL: <http://list.coin-or.org/pipermail/clp/attachments/20050726/32b94365/attachment.ksh>