Numeric instability in some tests

[jmarrec]

Issue overview

@mitchute reported a weird test error after making a completely unrelated IDD change in UnitarySystemModel_MultiSpeedDXCoolCoil_Only_NoFan. Turns out I can actually reproduce this in develop, even in Release mode, provided I run the test enough times (I'm on Ubuntu 24.04, gcc 13)

./Products/energyplus_tests --gtest_repeat=50 --gtest_filter=*UnitarySystemModel_MultiSpeedDXCoolCoil_Only_NoFan*

I'll get some failures if I do that (ratio about 1 in 20).

Repeating all tests (iteration 21) . . .

Note: Google Test filter = *UnitarySystemModel_MultiSpeedDXCoolCoil_Only_NoFan*
[==========] Running 1 test from 1 test suite.
[----------] Global test environment set-up.
[----------] 1 test from ZoneUnitarySysTest
[ RUN      ] ZoneUnitarySysTest.UnitarySystemModel_MultiSpeedDXCoolCoil_Only_NoFan
/home/julien/Software/Others/EnergyPlus/tst/EnergyPlus/unit/UnitarySystem.unit.cc:1715: Failure
The difference between thisSys->m_CoolingCycRatio and 0.36056 is 0.0191571141065352, which exceeds 0.001, where
thisSys->m_CoolingCycRatio evaluates to 0.37971711410653519,
0.36056 evaluates to 0.36055999999999999, and
0.001 evaluates to 0.001.
[  FAILED  ] ZoneUnitarySysTest.UnitarySystemModel_MultiSpeedDXCoolCoil_Only_NoFan (90 ms)
[----------] 1 test from ZoneUnitarySysTest (90 ms total)

[----------] Global test environment tear-down
[==========] 1 test from 1 test suite ran. (90 ms total)
[  PASSED  ] 0 tests.
[  FAILED  ] 1 test, listed below:
[  FAILED  ] ZoneUnitarySysTest.UnitarySystemModel_MultiSpeedDXCoolCoil_Only_NoFan

 1 FAILED TEST

This happens in a function that calls SolveRoot, so I guess there's a numeric instability there. My initial guess was that in some edge case an extra iteration would be required (I have investigated such a problem before) but that's not even what is seemingly happening here.

Anyways, should:

• Investigate why this specific test fails and see if we can mitigate that
• Investigate whether other tests are unstable as well

Operating System (Multiple choices)

Any

Operating System Version

N/A

Version of EnergyPlus

25.1.0-IOFreeze

Unmethours link or helpdesk ticket number

N/A

Defect file

No response

[jmarrec]

I've isolated the issue to the DXCoils::CalcMultiSpeedDXCoilCooling.

Interestingly enougjh, I can repeat the tests as many times as I want forking a new process everytime and it works fine.

$ i=1; while true; do   echo "🚀 Starting iteration $i...";   ./Products/energyplus_tests --gtest_brief=1 --gtest_filter='*UnitarySystemModel_MultiSpeedDXCoolCoil_Only_NoFan*' || break;   ((i++)); done

I killed it after 160 iterations.

Smells GOOD like a global not being reset somewhere. It also makes this extremely fun to track down, since it's hard to reproduce.