workaround for #40048, stack overflow in type intersection

[JeffBezanson]

This is yet another case where we end up creating circular variable constraints inside intersection. I have tried some approaches to fixing it, but so far it is proving tricky to both fully avoid the problem and preserve precision.

For some reason this call site has tended to lead to these sorts of tricky intersection cases, and does not seem to be 100% necessary, so I propose disabling it for now.

[palday]

This seems to fix the problem in MixedModels.jl, but tests will still fail on this because a deprecation in LinearAlgebra means we get the wrong warning in one test. (That will be fixed soon, now that we can actually run tests against 1.7-dev.)

cc: @dmbates

[Sacha0]

This patch also appears to work around the instance of #40048 that we've been hitting. Thanks Jeff! 🎉

[vtjnash]

This was added to avoid unreliable computations in method-ambiguity detection that was making Base tests fail periodically (it seemed they would fail most often on Windows, as most particularly cursed heisenbugs do). Is type-intersection now accurate enough to avoid those cases?

[JeffBezanson]

I don't know, but the failures in #40048 seem to be a larger problem. Maybe the failures I triggered in #39722 (comment) are a more consistent version of the heisenbug?

[JeffBezanson]

@nanosoldier runtests(ALL, vs = ":master")

[nanosoldier]

Your package evaluation job has completed - possible new issues were detected. A full report can be found here. cc @maleadt

[KristofferC]

@nanosoldier runtests(["Convex1d", "EarlyStopping", "EcoSISTEM", "FastFloat16s", "GtkObservables", "KLDivergences", "LoggingExtras", "MultiStochGrad", "PLCTag", "Pitaya", "SalesForceBulkApi", "StatsFuns", "Theta", "YAActL"], vs = ":master")

[nanosoldier]

Your package evaluation job has completed - possible new issues were detected. A full report can be found here. cc @maleadt

[KristofferC]

Looks like at least Gtk and Theta are related?

[JeffBezanson]

I could not reproduce them locally. Looks like a general memory corruption kind of issue, but I guess I can't rule out the PR as a cause completely.

[KristofferC]

I can't repro locally either.

[ararslan]

@nanosoldier runtests(["Gtk", "Theta"], vs=":master")

[nanosoldier]

Something went wrong when running your job:

ProcessExitedException(2)

cc @maleadt