Improve nothrow analysis of :new with missing sparam #46754
Conversation
Similar to #46693, but for :new, rather than getfield. Unfortunately, this is somewhat limited as we don't really have the ability to encode type equality constraints in the lattice. In particular, it would be nice to analyze: ``` struct Foo{T} x::T Foo(x::T) where {T} = new{T}(x) end ``` and be able to prove this nothrow. If it's really important, we could probably pattern match it, but for the moment, this is not easy to do. Nevertheless, we can do something about the similar, but simpler pattern ``` struct Foo{T} x Foo(x::T) where {T} = new{T}(x) end ``` which is what this PR does.
| typ, isexact = instanceof_tfunc(atyp) | ||
| if !isexact | ||
| atyp = unwrap_unionall(widenconst(atyp)) | ||
| if isType(atyp) && isTypeDataType(atyp.parameters[1]) |
There was a problem hiding this comment.
I thought at some point we had to deal with this check being insufficient and incomplete due to the existence of free-type-vars, for example if the user called:
julia> Val(Val.body);
which I thought used to be an issue for convert and Fix1 calls (particularly those implied by fieldtypes)
But perhaps we don't handle that case anyways, since that call segfaults, and this call poisons the dispatch cache permanently:
julia> Base.Fix1(Val.body, 1)
ERROR: UndefVarError: F not defined
There was a problem hiding this comment.
(see test/core.jl for TypeError("new", ...)
There was a problem hiding this comment.
Types with missing sparams are not valid ::Types, because they error in subtyping. They can only dispatch as ::DataType. There's a few places in the code, where we aren't careful about it, but we just need to clean those up. I don't think :new with a type with a free tvar is legal either, but absent other bugs, I don't think this code allows that.
Similar to #46693, but for :new, rather than getfield. Unfortunately, this is somewhat limited as we don't really have the ability to encode type equality constraints in the lattice. In particular, it would be nice to analyze:
and be able to prove this nothrow. If it's really
important, we could probably pattern match it, but for the moment, this is not easy to do. Nevertheless, we can do something about the similar, but simpler pattern
which is what this PR does.