Presolve bugs using "netlib" dataset

[sschnug]

While doing some first-hour experiments with this project (very interesting; especially the license!), the following was observed:

sascha@sascha-VirtualBox:~/Downloads/HiGHS/build/bin$ ./highs -p On -f ~/Downloads/greenbea.mps
Presolve is set to On
Reading file /home/sascha/Downloads/greenbea.mps
Setting default value crashMode = Off
Setting default value edWtMode = DSE1
====================================================================================
Running HiGHS
------
terminate called after throwing an instance of 'std::out_of_range'
  what():  vector::_M_range_check: __n (which is 18446744073709551615) >= this->size() (which is 30877)
Error during presolve: no variable found in singleton row 281.Aborted (core dumped)

It seems presolve is a relatively new addition and maybe bugs are to be expected. While probably not much relevant, i'm using a Linux + GCC + OpenMP=0 setup here and the exact mps-file in use can be obtained from this link (gz compressed).

Sascha

-----------
UPDATE
-----------
I wrote a small testing-script (available here).

Using above setup against the whole netlib-dataset (as described in the link above; README describes setup), there are:

• 2 crashes in presolve
• 1 timeout (with presolve on only) (with 30 seconds; should be more than enough according to the citation mentioned in README)

Output of the script (2 configs: default vs. default + presolve with 94 netlib instances each):

sascha@sascha-VirtualBox:~/Documents/HiGHS_eval$ python3 run.py
----------------
Run all 188 tests...
run 1/188
run 2/188
... (omitted in example)
run 188/188
    ...all tests finished!
-------------------
Check for errors...
Observed error:
    Instance: /home/sascha/Documents/HiGHS_eval/netlib/greenbea.mps
    Arguments: ['-p', 'On']
    Error-type: SOLVER ERROR
    -> returncode: -6
    -> stdout:
 
        Presolve is set to On
        Reading file /home/sascha/Documents/HiGHS_eval/netlib/greenbea.mps
        Setting default value crashMode = Off
        Setting default value edWtMode = DSE1
        ====================================================================================
        Running HiGHS
        ------
        Error during presolve: no variable found in singleton row 281.
 
    -> stderr:
 
        terminate called after throwing an instance of 'std::out_of_range'
          what():  vector::_M_range_check: __n (which is 18446744073709551615) >= this->size() (which is 30877)
 
 
Observed error:
    Instance: /home/sascha/Documents/HiGHS_eval/netlib/cycle.mps
    Arguments: ['-p', 'On']
    Error-type: TIMEOUT
    -> returncode: -999
    -> stdout:
 
 
 
    -> stderr:
 
 
 
Observed error:
    Instance: /home/sascha/Documents/HiGHS_eval/netlib/greenbeb.mps
    Arguments: ['-p', 'On']
    Error-type: SOLVER ERROR
    -> returncode: -6
    -> stdout:
 
        Presolve is set to On
        Reading file /home/sascha/Documents/HiGHS_eval/netlib/greenbeb.mps
        Setting default value crashMode = Off
        Setting default value edWtMode = DSE1
        ====================================================================================
        Running HiGHS
        ------
        Error during presolve: no variable found in singleton row 281.
 
    -> stderr:
 
        terminate called after throwing an instance of 'std::out_of_range'
          what():  vector::_M_range_check: __n (which is 18446744073709551615) >= this->size() (which is 30877)
 
 
    ...all checks finished!

[galabovaa]

Thank you for the info Sasha, we will look into it. Presolve is a newer feature, yes, but it still should be able to handle greenbea and greenbeb and a previous version of presolve did so. We extended the list of presolve rules so that must have changed something. Cycle I had not tested on so I will check that too when I get a chance.

[mattmilten]

We should add all those instances to check/CMakeLists.txt to test them with TravisCI.

[galabovaa]

The bug is now fixed. Both problems are in check/instances and presolve is on in ctest too.