get rid of some invalidation sources in Artifacts #37605
Conversation
| @@ -610,24 +620,24 @@ end | |||
| # Support `AbstractString`s, but avoid compilers needing to track backedges for callers | |||
| # of these functions in case a user defines a new type that is `<: AbstractString` | |||
| with_artifacts_directory(f::Function, artifacts_dir::AbstractString) = | |||
| with_artifacts_directory(f, string(artifacts_dir)) | |||
| with_artifacts_directory(f, String(artifacts_dir)::String) | |||
There was a problem hiding this comment.
These are just fixes so that the code adheres to what the comment claims it does.
| @@ -270,7 +271,12 @@ function unpack_platform(entry::Dict, name::String, artifacts_toml::String)::Uni | |||
| end | |||
|
|
|||
| # Collect all String-valued mappings in `entry` and use them as tags | |||
| tags = Dict(Symbol(k) => v for (k, v) in entry if isa(v, String)) | |||
There was a problem hiding this comment.
Symbol(k)::Symbol => v::String might also work, but your version is good.
|
Looks like the keys to the dictionary passed to unpack_platform are not Strings, but possibly substrings. |
I tweaked things a bit. Basically, we have a "funnel" where the top methods accept wide types and convert them to known concrete types for which the core implementations are defined on. This means that the core parts will not be invalidated since they have concrete type information while things should still accept generic types. |
Found with SnoopCompile and loading Plots.
Not sure if this is the best way to resolve them but it does the job. @timholy?