julia type inference crashes (typeerror) #41908
Assignees
Labels
bug
Indicates an unexpected problem or unintended behavior
compiler:inference
Type inference
types and dispatch
Types, subtyping and method dispatch
Comments
|
Is there a tool that would allow me to debug this myself? I tried sprinkling in println's in some compiler/*.jl files but not only did compilation appear to hang, this will also take a really long time to get to the bottom off, with all that recompilation. |
|
Can you post the full stack trace? If I could see which line of base/compiler/*.jl was throwing the error, it could likely be fixed easily, and an example case generated, from knowing that |
|
|
this should have been the stack trace for julia master |
|
Heh, trying to find a simple reproducer I stumbled upon a different but likely similar issue: using LinearAlgebra
function foo()
x = Any[1][1]
x = x::(Diagonal{Int, T} where Int<:T<:Int)
size(x)
end
foo()Works (=only throws a runtime-error) on 1.6, but fails on 1.7 and master. Worth its own issue given that it is a regression? |
|
They are slightly different, but I think we can keep them both in this issue, since they make sense to fix together. original: julia> bar(x::Diagonal{Int, T}) where {Int<:T<:Int} = 1
julia> foo() = bar(Any[1][1])
julia> foo()
Internal error: encountered unexpected error in runtime:
TypeError(func=:Diagonal, context="V", expected=V<:AbstractArray{T, 1}, got=Int64)
jl_type_error_rt at /Users/jameson/julia1/src/rtutils.c:119
check_datatype_parameters at /Users/jameson/julia1/src/jltypes.c:1205
inst_datatype_inner at /Users/jameson/julia1/src/jltypes.c:1368
inst_type_w_ at /Users/jameson/julia1/src/jltypes.c:1760
inst_tuple_w_ at /Users/jameson/julia1/src/jltypes.c:1653
jl_instantiate_unionall at /Users/jameson/julia1/src/jltypes.c:1018 [inlined]
jl_apply_type at /Users/jameson/julia1/src/jltypes.c:972
subst_trivial_bounds at ./compiler/utilities.jl:159
normalize_typevars at ./compiler/utilities.jl:169
specialize_method at ./compiler/utilities.jl:181
specialize_method at ./compiler/utilities.jl:180 [inlined]
typeinf_edge at ./compiler/typeinfer.jl:780 [inlined]and yours I see as "works" on master too: julia> foo()
Internal error: encountered unexpected error in runtime:
TypeError(func=:Diagonal, context="V", expected=V<:AbstractArray{T, 1}, got=Int64)
jl_type_error_rt at /Users/jameson/julia1/src/rtutils.c:119
check_datatype_parameters at /Users/jameson/julia1/src/jltypes.c:1205
inst_datatype_inner at /Users/jameson/julia1/src/jltypes.c:1368
inst_type_w_ at /Users/jameson/julia1/src/jltypes.c:1760
jl_instantiate_unionall at /Users/jameson/julia1/src/jltypes.c:1018 [inlined]
normalize_unionalls at /Users/jameson/julia1/src/jltypes.c:1313
inst_datatype_inner at /Users/jameson/julia1/src/jltypes.c:1330
jl_apply_tuple_type_v_ at /Users/jameson/julia1/src/jltypes.c:1538 [inlined]
jl_apply_tuple_type_v at /Users/jameson/julia1/src/jltypes.c:1548
_jl_invoke at /Users/jameson/julia1/src/gf.c:0 [inlined]
jl_apply_generic at /Users/jameson/julia1/src/gf.c:2427
jl_apply at /Users/jameson/julia1/src/./julia.h:1768 [inlined]
do_apply at /Users/jameson/julia1/src/builtins.c:713
argtypes_to_type at ./compiler/typeutils.jl:53 [inlined] |
|
Looks like the OP can pragmatically be solved with diff --git a/base/compiler/utilities.jl b/base/compiler/utilities.jl
index 5d4d52172a..852ec1e216 100644
--- a/base/compiler/utilities.jl
+++ b/base/compiler/utilities.jl
@@ -166,7 +166,11 @@ end
# f(x::S) where S<:Int
# gets cached and matches a concrete dispatch case.
function normalize_typevars(method::Method, @nospecialize(atypes), sparams::SimpleVector)
- at2 = subst_trivial_bounds(atypes)
+ try
+ at2 = subst_trivial_bounds(atypes)
+ catch
+ return atypes, sparams
+ end
if at2 !== atypes && at2 == atypes
atypes = at2
sp_ = ccall(:jl_type_intersection_with_env, Any, (Any, Any), at2, method.sig)::SimpleVector(Is there a nice way of finding out whether The other case is slightly trickier as julia> Tuple{LinearAlgebra.Diagonal{Int64, T} where Int64<:T<:Int64}
ERROR: TypeError: in Diagonal, in V, expected V<:AbstractArray{T, 1}, got Type{Int64}I guess we would want to actually form that type? (Alternatively, maybe we should forbid |
|
Ah, I already needed that patch in #41795. |
JeffBezanson
added a commit
that referenced
this issue
Aug 23, 2021
JeffBezanson
added a commit
that referenced
this issue
Aug 25, 2021
Co-authored-by: Martin Holters <[email protected]>
KristofferC
pushed a commit
that referenced
this issue
Aug 25, 2021
Co-authored-by: Martin Holters <[email protected]> (cherry picked from commit e2aeefb)
63 tasks
KristofferC
pushed a commit
that referenced
this issue
Oct 29, 2021
Co-authored-by: Martin Holters <[email protected]> (cherry picked from commit e2aeefb)
KristofferC
pushed a commit
that referenced
this issue
Nov 11, 2021
Co-authored-by: Martin Holters <[email protected]> (cherry picked from commit e2aeefb)
LilithHafner
pushed a commit
to LilithHafner/julia
that referenced
this issue
Feb 22, 2022
…ng#41976) Co-authored-by: Martin Holters <[email protected]>
LilithHafner
pushed a commit
to LilithHafner/julia
that referenced
this issue
Mar 8, 2022
…ng#41976) Co-authored-by: Martin Holters <[email protected]>
staticfloat
pushed a commit
that referenced
this issue
Dec 23, 2022
Co-authored-by: Martin Holters <[email protected]> (cherry picked from commit e2aeefb)
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
my tests started crashing with the stacktrace:
happens at least on julia 1.6.2, and also on master.
I wasn't able to really reduce it down very far, as it's a reasonably large codebase. To run it, you need my fork of TensorKit.jl, and the planocalypse branch of MPSKit.jl . I will try to reduce it down further, but so far I didn't have much luck.
The text was updated successfully, but these errors were encountered: