[Clp] clp/cbc bomb on 10teams on OS X 10.6.7 : not clp's fault

[Sven de Vries]

Hi,

tried to narrow down the problem. It seems like gcc 4.5.2 is the culprit.
With the other compilers I tried, it seems fine, see below.

best
sven


gcc version 4.0.1 (Apple Inc. build 5494): OK
gcc version 4.2.1 (Apple Inc. build 5666) (dot 3): OK
gcc-Version 4.4.4 (GCC) (fink): OK
gcc-Version 4.5.0 (GCC) (fink): OK

gcc-Version 4.5.2 (GCC): FAILS




On Wed, 18 May 2011, Sven de Vries wrote:

> Hi,
>
> noticed [a month ago] that current trunk clp (Coin LP version trunk, build 
> Apr 11 2011) bombs on 10teams [actually noticed this in cbc -miplib, but clp 
> does the same]
>
>
> ./bin/clp
> Coin LP version trunk, build Apr 11 2011
> Clp takes input from arguments ( - switches to stdin)
> Enter ? for list of commands or help
> Clp:log 63
> logLevel was changed from 1 to 63
> Clp:import 10teams
> At line 15 NAME          10TEAMS
> At line 16 ROWS
> At line 248 COLUMNS
> At line 8351 RHS
> At line 8467 BOUNDS
> At line 10493 ENDATA
> Problem 10TEAMS has 230 rows, 2025 columns and 12150 elements
> Model was imported from ./10teams in 0.011676 seconds
> Clp:solve
> 20 rows dropped after presolve pass 1
> 20 rows dropped after presolve pass 2
> Presolve 210 (-20) rows, 1600 (-425) columns and 9600 (-2550) elements
> Presolve took 0.01 seconds (total 0.01)
> Bounds were tightened 0 times
> Initial range of elements is 1 to 1
> Not bothering to scale as good enough
> Absolute values of scaled rhs range from 1 to 2, minimum gap 1e+100
> Absolute values of scaled objective range from 3 to 86
> Absolute values of scaled bounds range from 1 to 1, minimum gap 1
>        length of U 0, length of L 0
> Primal error 0, dual error 0
> Perturbing problem by 0.001 % of 86 - largest nonzero change 0.00024935072 (% 
> 0.0037644719) - largest zero change 0
> Primal error 0, dual error 0
> 0  Obj 20 Primal inf 119.99999 (120)
> 0  Obj 20 Primal inf 119.99999 (120)
> Pivot row 102
> dirOut 1, dirIn 1, theta 11.000135, out 1, dj 11.000135, alpha -1
> 1 31.000135 In: C13 Out: R102 dj ratio 1 distance 11.000135
> Pivot row 179
> Segmentation fault
>
> this version was build without me setting any -O or -g...
>
>
> if I build with -g, I get: [~/coin-Cbc/g] % bin/clp
> Coin LP version trunk, build Apr 11 2011
> Clp takes input from arguments ( - switches to stdin)
> Enter ? for list of commands or help
> Clp:log 63
> logLevel was changed from 1 to 63
> Clp:import 10teams
> At line 15 NAME          10TEAMS
> At line 16 ROWS
> At line 248 COLUMNS
> At line 8351 RHS
> At line 8467 BOUNDS
> At line 10493 ENDATA
> Problem 10TEAMS has 230 rows, 2025 columns and 12150 elements
> Model was imported from ./10teams in 0.019515 seconds
> Clp:solve
> 20 rows dropped after presolve pass 1
> 20 rows dropped after presolve pass 2
> Presolve 210 (-20) rows, 1600 (-425) columns and 9600 (-2550) elements
> Presolve took 0.02 seconds (total 0.02)
> Bounds were tightened 0 times
> Initial range of elements is 1 to 1
> Not bothering to scale as good enough
> Absolute values of scaled rhs range from 1 to 2, minimum gap 1e+100
> Absolute values of scaled objective range from 3 to 86
> Absolute values of scaled bounds range from 1 to 1, minimum gap 1
>        length of U 0, length of L 0
> Primal error 0, dual error 0
> Perturbing problem by 0.001 % of 86 - largest nonzero change 0.00024935072 (% 
> 0.0037644719) - largest zero change 0
> Primal error 0, dual error 0
> 0  Obj 20 Primal inf 119.99999 (120)
> 0  Obj 20 Primal inf 119.99999 (120)
> Pivot row 102
> dirOut 1, dirIn 1, theta 11.000135, out 1, dj 11.000135, alpha -1
> 1 31.000135 In: C13 Out: R102 dj ratio 1 distance 11.000135
> Pivot row 179
> dirOut 1, dirIn 1, theta 4.0001182, out 1, dj 4.0001182, alpha -1
> 2 35.000254 In: C231 Out: R179 dj ratio 1 distance 4.0001182
> Pivot row 144
> dirOut 1, dirIn 1, theta 9.0001461, out 1, dj 9.0001461, alpha -1
> 3 44.0004 In: C1201 Out: R144 dj ratio 1 distance 9.0001461
> Pivot row 174
>
> and it finishes without error.
>
> ... about 2 months ago it was still working fine.
>
> did anybody observe this too or knows what's wrong?
>
> best
> sven
>
> ps: similar with gcc 452...
> pps: happens also with current build: Coin LP version trunk, build May 17 
> 2011
>