Avoid some stackoverflow during typeintersect.

[N5N3]

When I dig in #46736, I found that subtype_in_env_existential is somehow dangous as we flip all typevar to the right side.
I tried to follow #48006 style but it caused many test failure. So this PR just makes the following change to avoid some known stackoverflow:

1. Make sure env is restored between 2 adjacent subtype_in_env_existential. Thus the second subtype check wont be influenced by the first one. (close SIGSEGV and StackOverflow from dispatching on ::Type{T}, ::Array{Union{T,Missing},N}) #36185)
2. Add a fast path to skip the repeated intersect_aside (close Healpix.jl tests hang on Julia v1.9 #46736.)

[giordano]

On this branch tests of Healpix.jl crash instead of hanging (which is a slight improvement):

Internal error: stack overflow in type inference of readPolarizedMapFromFITS(String, Int64, Type{Float32}).
This might be caused by recursion over very long tuples or argument lists.
Map I/O: Error During Test at /Users/mose/.julia/packages/Healpix/UmwcZ/test/runtests.jl:48
  Got exception outside of a @test
  LoadError: TypeError: in new, expected TypeVar, got Type{Float32}
  Stacktrace:
    [1] readPolarizedMapFromFITS(fileName::AbstractString, column::Any, t::Type{T}) where T
      @ Healpix ~/.julia/packages/Healpix/UmwcZ/src/map_io.jl:107
    [2] top-level scope
      @ ~/.julia/packages/Healpix/UmwcZ/test/test_mapio.jl:22
    [3] include(fname::String)
      @ Base.MainInclude ./client.jl:478
    [4] macro expansion
      @ ~/.julia/packages/Healpix/UmwcZ/test/runtests.jl:49 [inlined]
    [5] macro expansion
      @ ~/repo/julia/usr/share/julia/stdlib/v1.10/Test/src/Test.jl:1502 [inlined]
    [6] top-level scope
      @ ~/.julia/packages/Healpix/UmwcZ/test/runtests.jl:49
    [7] include(fname::String)
      @ Base.MainInclude ./client.jl:478
    [8] top-level scope
      @ none:6
    [9] eval
      @ ./boot.jl:370 [inlined]
   [10] exec_options(opts::Base.JLOptions)
      @ Base ./client.jl:280
   [11] _start()
      @ Base ./client.jl:522
  in expression starting at /Users/mose/.julia/packages/Healpix/UmwcZ/test/test_mapio.jl:22
Test Summary: | Pass  Error  Total  Time
Map I/O       |    7      1      8  0.7s

Is this a bug in the package or something else to fix on the julia side? CC: @ziotom78

[KristofferC]

FWIW, this seems to work around it:

diff --git a/src/map_io.jl b/src/map_io.jl
index 3bea696..d3c2699 100644
--- a/src/map_io.jl
+++ b/src/map_io.jl
@@ -104,7 +104,10 @@ function readPolarizedMapFromFITS(
     end
 
     f = CFITSIO.fits_open_table(fileName)
-    i, q, u = (readMapFromFITS(f, colidx, t) for colidx in (column_i, column_q, column_u))
+    i = readMapFromFITS(f, column_i, t)
+    q = readMapFromFITS(f, column_q, t)
+    u = readMapFromFITS(f, column_u, t)
+
     CFITSIO.fits_close_file(f)
 
     PolarizedHealpixMap(i, q, u)

[N5N3]

Not that suppressing as i, q, u = (readMapFromFITS(f, colidx, t) for ...... would meet #23618
and

julia> code_typed(Healpix.readMapFromFITS, Tuple{Healpix.CFITSIO.FITSFile,Int,DataType})
ERROR: StackOverflowError:

[N5N3]

Might be a smaller MWE
this gives stackoverflow

struct HealpixMap{T,AA<:AbstractArray{T,1}}
    pixels::AA
    function HealpixMap{Union{T,Nothing},AA}(a::Number) where {T,AA<:AbstractArray{Union{T,Nothing},1}}
        new(Union{T, Nothing}[nothing for i in 1:a])
    end
    function HealpixMap{T,AA}(arr::AA) where {T,AA<:AbstractArray{T,1}}
        new(arr)
    end
end
HealpixMap{T}(arr) where {T} = HealpixMap{T,Vector{T}}(collect(arr))
f(::Type{T}) where {T} = HealpixMap{T}(T[])
code_typed(f, Tuple{DataType})

but this gives the correct result

struct HealpixMap{T,AA<:AbstractArray{T,1}}
    pixels::AA
    function HealpixMap{Union{T,Nothing},AA}(a::Number) where {T,AA<:AbstractArray{Union{T,Nothing},1}}
        new{Union{T,Nothing},AA}(Union{T, Nothing}[nothing for i in 1:a])
    end
    function HealpixMap{T,AA}(arr::AA) where {T,AA<:AbstractArray{T,1}}
        new{T,AA}(arr)
    end
end

[vtjnash]

Seems good to me. I think there was a small typo in the commit message and it needs to re-run CI since it had some networking issues. Could you push an update?

[vtjnash]

@nanosoldier runtests()

[nanosoldier]

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

[vtjnash]

I don't think it needs to be blocking to merge this, but AngularMomentumAlgebra may be good to look into https://s3.amazonaws.com/julialang-reports/nanosoldier/pkgeval/by_hash/36cd9c6_vs_79e29e3/AngularMomentumAlgebra.primary.log (and maybe some of the other reported hangs)

[N5N3]

I don't think it needs to be blocking to merge this, but AngularMomentumAlgebra may be good to look into https://s3.amazonaws.com/julialang-reports/nanosoldier/pkgeval/by_hash/36cd9c6_vs_79e29e3/AngularMomentumAlgebra.primary.log (and maybe some of the other reported hangs)

I'll take a look at AngularMomentumAlgebra and other failure and push the fix (if needed) in #48006.