New StackOverflow in jl_has_free_typevars, intersect_all, intersect at,

[KristofferC]

In the latest PkgEval (https://github.com/JuliaCI/NanosoldierReports/blob/5401256da7d60e8b1ea5ec7c7a1a3becc543201b/pkgeval/by_hash/d97ccd7_vs_c4acbf9/report.md) a lot of packages started failing with logs like

Internal error: encountered unexpected error in runtime:
StackOverflowError()
jl_has_free_typevars at /workspace/srcdir/src/jltypes.c:190
intersect_union at /workspace/srcdir/src/subtype.c:2156
intersect at /workspace/srcdir/src/subtype.c:2979
intersect_all at /workspace/srcdir/src/subtype.c:3110
intersect_aside at /workspace/srcdir/src/subtype.c:2146
intersect_var at /workspace/srcdir/src/subtype.c:2337
intersect at /workspace/srcdir/src/subtype.c:2963
intersect_var at /workspace/srcdir/src/subtype.c:2257

The packages failing are:

AnyMOD
BifurcationKit
BilevelJuMP
BilevelOptimization
BirkhoffDecomposition
BundleMethod
CDDLib
COSMO
Cbc
ClimateMARGO
Clp
ConstraintSolver
Convex
CutPruners
DataEnvelopmentAnalysis
DensityRatioEstimation
DiffEqBayes
DiffEqCallbacks
DiffEqJump
DifferentialEquations
DiscreteFunctions
DualDecomposition
ECOS
ErdosExtras
Fri
GLPK
GasModels
GasPowerModels
GraphicalModelLearning
GuessworkQuantumSideInfo
InfiniteOpt
InfrastructureModels
Ipopt
JuliaGrid
Juniper
LatinSquares
LazySets
LightGraphsFlows
LightGraphsMatching
MINLPTests
MathOptInterface
ModelingToolkit
MosekTools
MultiScaleArrays
NaiveGAflux
NaiveNASflux
NaiveNASlib
OPFSampler
OpSel
ParameterJuMP
PiecewiseLinearOpt
PolyhedralRelaxations
PowerDynamics
PowerModelsACDC
PowerModelsAnnex
PowerModelsDistribution
PowerModelsRestoration
PowerModelsSecurityConstrained
PowerSimulations
PowerWaterModels
ProxSDP
QHull
QuadraticToBinary
RandomizedProgressiveHedging
SCS
SDDP
SDPA
SDPAFamily
Seleroute
SemidefiniteOptInterface
SetProg
SimplePosetAlgorithms
StochDynamicProgramming
StochasticPrograms
Sudoku
TravelingSalesmanExact
Tulip
UnbalancedOptimalTransport
WaterModels
sparseQFCA
vOptGeneric

[KristofferC]

Looks like this happened ~5 days ago https://github.com/JuliaCI/NanosoldierReports/blob/master/pkgeval/by_date/2020-09/02/report.md

[JeffBezanson]

Bisected to #36962. That probably more exposed it than caused it, but I think this has caused enough problems that at this point we should revert it.