inference: minor refactoring on abstractinterpretation.jl #58154
Conversation
| (b::Core.Binding, bpart::Core.BindingPartition, world::UInt)-> | ||
| Pair{WorldRange, Pair{Core.Binding, Core.BindingPartition}}(WorldRange(bpart.min_world, bpart.max_world), b=>bpart), | ||
| interp, g, wwr) | ||
| function scan_partitions(query::Function, interp::AbstractInterpreter, g::GlobalRef, wwr::WorldWithRange) |
There was a problem hiding this comment.
Do we mean for these all to be ::F {where F<:Function} ?
There was a problem hiding this comment.
Does ::F where F<:Function still lead to different specialization than ::Function? I'm just borrowing this annotation from what @Keno is using elsewhere.
There was a problem hiding this comment.
Yeah, get_compileable_sig still implements the de-specialization heuristics.
I'd change these signatures in a separate PR.
| return Pair{Any,Any}(newty, ErrorException) | ||
| end | ||
| (valid_worlds, ret) = scan_partitions((interp, _, partition)->global_assignment_binding_rt_exct(interp, partition, newty), interp, g, sv.world) | ||
| newty′ = Ref{Any}(newty) |
There was a problem hiding this comment.
| newty′ = Ref{Any}(newty) | |
| newty′ = Core.Box(newty) |
I think that's a bit more canonical (esp for JET and other analysis tools)?
There was a problem hiding this comment.
JET can analyze both cases.
Since the frontend inserts Core.Box, are you suggesting we should consistently use Core.Box for captured variables?
I slightly prefer Ref because .contents is a bit long, but I don't have a strong preference.
- add missing `@nospecialize` annotation - add more type annotations to functions - fixed capturing uninferrable variables
aviatesk
force-pushed
the
avi/minor-absint-refactor
branch
from
86ea8e9 to
939ead0
Compare
- add missing `@nospecialize` annotation - add more type annotations to functions - fixed capturing uninferrable variables
KristofferC
removed
the
backport 1.12
…58154) - add missing `@nospecialize` annotation - add more type annotations to functions - fixed capturing uninferrable variables
…58154) - add missing `@nospecialize` annotation - add more type annotations to functions - fixed capturing uninferrable variables
@nospecializeannotation