Infinite recursion when promote_rule depends on order of arguments
#13193
Assignees
Labels
bug
Indicates an unexpected problem or unintended behavior
Comments
JeffBezanson
added
the
bug
|
I believe this no longer causes infinite recursion in the compiler, but rather just during execution. If Here's an example of a case in Base that takes advantage of the current behavior: However, that could be resolved by adding a disambiguating definition. After all, the first result doesn't really make sense --- a Bool cannot be converted to an |
JeffBezanson
changed the title
promote_rule depends on order of arguments
|
Confirmed that's still true: |
|
On v1.11.3 this still causes a stack overflow only during execution, but on nightly (v1.13) this sadly again causes a stack overflow in the compiler itself. |
nsajko
added a commit
to nsajko/julia
that referenced
this issue
Feb 24, 2025
Don't use recursion, delete the `promote_result` function. Use loop known to terminate, because of a hardcoded limit on the iteration count. Closes JuliaLang#57507, this PR encompasses the fixes from that PR and is more comprehensive. Fixes JuliaLang#13193
nsajko
added a commit
to nsajko/julia
that referenced
this issue
Feb 24, 2025
Don't use recursion, delete the `promote_result` function. Use loop known to terminate, because of a hardcoded limit on the iteration count. Closes JuliaLang#57507, this PR encompasses the fixes from that PR and is more comprehensive. Fixes JuliaLang#13193
nsajko
added a commit
to nsajko/julia
that referenced
this issue
Feb 24, 2025
Don't use recursion, delete the `promote_result` function. Use loop known to terminate, because of a hardcoded limit on the iteration count. Closes JuliaLang#57507, this PR encompasses the fixes from that PR and is more comprehensive. Fixes JuliaLang#13193
nsajko
added a commit
to nsajko/julia
that referenced
this issue
Feb 26, 2025
Don't use recursion, delete the `promote_result` function. Use loop known to terminate, because of a hardcoded limit on the iteration count. Closes JuliaLang#57507, this PR encompasses the fixes from that PR and is more comprehensive. Fixes JuliaLang#13193
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
This is a more complicated case than #13183, and its root cause seems to be different since it persists after 9f72db1. But if you're interested in squashing these, here's an interesting case: given two
Numberwrappers that are themselves numbers and promote numbers to their type, we get an infinite recursion in the compiler/inference if they ever come into contact with each other:immutable SIQuantity{T<:Number} <: Number; end Base.promote_rule{T, S}(::Type{SIQuantity{T}}, ::Type{SIQuantity{S}}) = SIQuantity{promote_type(T,S)} Base.promote_rule{T, S<:Number}(::Type{SIQuantity{T}}, ::Type{S}) = SIQuantity{promote_type(T,S)} immutable Interval{T<:Number} <: Number; end Base.promote_rule{T, S}(::Type{Interval{T}}, ::Type{Interval{S}}) = Interval{promote_type(T,S)} Base.promote_rule{T, S<:Number}(::Type{Interval{T}}, ::Type{S}) = Interval{promote_type(T,S)} # note that promote_rule can hide the ambiguity here; no warnings are thrown julia> promote_type(Interval{Int}, SIQuantity{Int}) ^C^C^C signal (11): Segmentation fault: 11 jl_is_type at /Users/mbauman/Code/julia-0.4/src/jltypes.c:68 arg_type_tuple at /Users/mbauman/Code/julia-0.4/src/gf.c:1384 jl_apply_generic at /Users/mbauman/Code/julia-0.4/src/gf.c:1688 promote_type at promotion.jl:119 jl_apply at /Users/mbauman/Code/julia-0.4/src/gf.c:1691 jl_apply at /Users/mbauman/Code/julia-0.4/src/./julia.h:1332 abstract_call_gf at ./inference.jl:659 abstract_call at ./inference.jl:877 abstract_eval_call at ./inference.jl:932 abstract_eval at ./inference.jl:959 abstract_eval_call at ./inference.jl:904 abstract_eval at ./inference.jl:959 typeinf_uncached at ./inference.jl:1620 jlcall_typeinf_uncached_36 at /Users/mbauman/Code/julia-0.4/usr/lib/julia/sys.dylib (unknown line) typeinf at ./inference.jl:1337 jlcall_typeinf_4 at /Users/mbauman/Code/julia-0.4/usr/lib/julia/sys.dylib (unknown line) typeinf at ./inference.jl:1287 abstract_call_gf at ./inference.jl:737 abstract_call at ./inference.jl:877 abstract_eval_call at ./inference.jl:932 abstract_eval at ./inference.jl:959 abstract_eval_call at ./inference.jl:904 abstract_eval at ./inference.jl:959 …(via JuliaArrays/AxisArrays.jl#20 (comment))
The text was updated successfully, but these errors were encountered: