Segfault with small time limit

[odow]

Originally reported as jump-dev/HiGHS.jl#240

MPS file: time_crash.mps.txt

Very reproducible in HiGHS v1.7.2 and v1.8.0, on both Windows and Mac, and every time.

julia> HiGHS_jll.highs() do exe
       run(`$exe time_crash.mps --time_limit 0.1`)
       end
Running HiGHS 1.8.0 (git hash: fcfb53414): Copyright (c) 2024 HiGHS under MIT licence terms
QP   time_crash has 8440 rows; 5871 cols; 24016 matrix nonzeros; 5871 Hessian nonzeros
Coefficient ranges:
  Matrix [2e-01, 1e+03]
  Cost   [8e+02, 7e+03]
  Bound  [4e-03, 1e+03]
  RHS    [2e-03, 1e+00]
  Iteration        Objective     NullspaceDim
ERROR: failed process: Process(`/Users/oscar/.julia/artifacts/6c052094bcce703903faeccc27a66f3453fd061e/bin/highs time_crash.mps --time_limit 0.1`, ProcessSignaled(11)) [0]

The Julia log makes it look like is related to HighsTimerClock during the buildSimple call in factorization:

[1610] signal (11.1): Segmentation fault: 11
in expression starting at REPL[6]:1
_ZN7HFactor11buildSimpleEv at /Users/oscar/.julia/artifacts/6c052094bcce703903faeccc27a66f3453fd061e/lib/libhighs.1.8.dylib (unknown line)
_ZN7HFactor5buildEP15HighsTimerClock at /Users/oscar/.julia/artifacts/6c052094bcce703903faeccc27a66f3453fd061e/lib/libhighs.1.8.dylib (unknown line)
_ZN5Basis5buildEv at /Users/oscar/.julia/artifacts/6c052094bcce703903faeccc27a66f3453fd061e/lib/libhighs.1.8.dylib (unknown line)
_ZN5BasisC2ER7RuntimeNSt3__16vectorIiNS2_9allocatorIiEEEENS3_I11BasisStatusNS4_IS7_EEEES6_ at /Users/oscar/.julia/artifacts/6c052094bcce703903faeccc27a66f3453fd061e/lib/libhighs.1.8.dylib (unknown line)
_Z14solveqp_actualR8InstanceR8SettingsR21QpHotstartInformationR10StatisticsR13QpModelStatusR10QpSolutionR10HighsTimer at /Users/oscar/.julia/artifacts/6c052094bcce703903faeccc27a66f3453fd061e/lib/libhighs.1.8.dylib (unknown line)
_Z7solveqpR8InstanceR8SettingsR10StatisticsR16HighsModelStatusR10HighsBasisR13HighsSolutionR10HighsTimer at /Users/oscar/.julia/artifacts/6c052094bcce703903faeccc27a66f3453fd061e/lib/libhighs.1.8.dylib (unknown line)
_ZN5Highs11callSolveQpEv at /Users/oscar/.julia/artifacts/6c052094bcce703903faeccc27a66f3453fd061e/lib/libhighs.1.8.dylib (unknown line)
_ZN5Highs3runEv at /Users/oscar/.julia/artifacts/6c052094bcce703903faeccc27a66f3453fd061e/lib/libhighs.1.8.dylib (unknown line)

[fwesselm]

I think the QP code does not check that HiGHS hit the time limit in computeStartingPointHighs. Also, I think the code may try to set a negative time limit for HiGHS. I have tried to fix this (https://github.com/fwesselm/HiGHS/tree/QpTimeLimit), but I do not know the QP code well.

[jajhall]

Thanks @odow and @fwesselm , I'll look at this today

[jajhall]

I think the QP code does not check that HiGHS hit the time limit in computeStartingPointHighs. Also, I think the code may try to set a negative time limit for HiGHS. I have tried to fix this (https://github.com/fwesselm/HiGHS/tree/QpTimeLimit), but I do not know the QP code well.

I see where a negative value of use_time_limit might be set. Other than to modification in

const double use_time_limit =

    settings.time_limit < kHighsInf
        ? std::max(settings.time_limit - timer.readRunHighsClock(), 0.0)
        : kHighsInf;

I don't see any other meaningful changes - every other change seems to be due to formatting.

Now that @feldmeier is not supporting the QP solver, I'll apply clang formatting to it.

Could the qpsolver/* files be added to the clang test on GitHub @galabovaa ?

[fwesselm]

Sorry, I realized that the code was auto-formatted while I was editing it.

The second change is to check whether computeStartingPointHighs returned an error:

if (qp_model_status == QpModelStatus::kError) return QpAsmStatus::kError;

[fwesselm]

I think the QP code does not check that HiGHS hit the time limit in computeStartingPointHighs. Also, I think the code may try to set a negative time limit for HiGHS. I have tried to fix this (https://github.com/fwesselm/HiGHS/tree/QpTimeLimit), but I do not know the QP code well.

I see where a negative value of use_time_limit might be set. Other than to modification in

const double use_time_limit =

    settings.time_limit < kHighsInf
        ? std::max(settings.time_limit - timer.readRunHighsClock(), 0.0)
        : kHighsInf;

I don't see any other meaningful changes - every other change seems to be due to formatting.

Now that @feldmeier is not supporting the QP solver, I'll apply clang formatting to it.

Could the qpsolver/* files be added to the clang test on GitHub @galabovaa ?

I have updated the branch now. Sorry for creating the extra work. I could create a PR, if you think the changes are reasonable.

[jajhall]

Sorry @fwesselm I've done a lot more - the code wasn't handling interrupts or timeouts - and will run clang formatting so I don't conflict with the changes @galabovaa has made on latest

[jajhall]

Closed by #1995