Update to changes to construct_ssa

[oxinabox]

This is an update to make things work after JuliaLang/julia#50943

Right now however it seems to makes tests hang

[codecov]

Codecov Report

Attention: 5 lines in your changes are missing coverage. Please review.

Comparison is base (7be0a92) 56.17% compared to head (d63aeed) 56.50%.
Report is 7 commits behind head on main.

Files	Patch %	Lines
src/stage2/abstractinterpret.jl	25.00%	3 Missing ⚠️
src/stage1/recurse.jl	66.66%	1 Missing ⚠️
src/stage2/interpreter.jl	66.66%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #234      +/-   ##
==========================================
+ Coverage   56.17%   56.50%   +0.32%     
==========================================
  Files          28       28              
  Lines        2848     2869      +21     
==========================================
+ Hits         1600     1621      +21     
  Misses       1248     1248              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[topolarity]

LGTM

The only thing that construct_ssa! inspects in the OptimizationState are the function- and block- level slot types (slottypes and bb_vartables, respectively)

[topolarity]

FWIW in the typical pipeline convert_to_ircode is used to convert CodeInfo-style statements to IRCode, but that's skipped here.

Most of the function is a no-op if you have no (informative) results from inference but there is some normalization (e.g. strip_trailing_junk!) that might be needed here for certain IR.

None of that is new in this PR though, so we can worry about it later.

[oxinabox]

@topolarity any ideas why this would make it hang?
Seen both locally and in CI

[topolarity]

Not sure, but here's a reduced MWE at least:

using Diffractor

function times_three_while(x)
    z = x
    i = 3
    while i > 1
        z += x
        i -= 1
    end
    z
end

let var"'" = Diffractor.PrimeDerivativeBack
    println(times_three_while'(1.0) == 3.0)
end

The IR that Diffractor generates is identical on latest Julia versus 2cee483bce (ir.txt), but inference appears to get stuck in a loop doing semi-concrete eval of > and -:

semi-concrete eval: (::Diffractor.∂⃖{1})(typeof(Base.:(-)), Int64, Int64) from (::Diffractor.∂⃖{N})(T, Any...) where {T, N}
semi-concrete eval: (::Diffractor.∂⃖{1})(typeof(Base.:(>)), Int64, Int64) from (::Diffractor.∂⃖{N})(T, Any...) where {T, N}
semi-concrete eval: (::Diffractor.∂⃖{1})(typeof(Base.:(-)), Int64, Int64) from (::Diffractor.∂⃖{N})(T, Any...) where {T, N}
semi-concrete eval: (::Diffractor.∂⃖{1})(typeof(Base.:(>)), Int64, Int64) from (::Diffractor.∂⃖{N})(T, Any...) where {T, N}
semi-concrete eval: (::Diffractor.∂⃖{1})(typeof(Base.:(-)), Int64, Int64) from (::Diffractor.∂⃖{N})(T, Any...) where {T, N}
semi-concrete eval: (::Diffractor.∂⃖{1})(typeof(Base.:(>)), Int64, Int64) from (::Diffractor.∂⃖{N})(T, Any...) where {T, N}
... (repeats infinitely)

[oxinabox]

I guess since the IR is identical, we could consider merging this.

[aviatesk]

Going to merge as the 1.10 CI passes.