Allow system level metadata information #3892
Conversation
| struct Author end | ||
| struct MyVersion end | ||
| struct License end | ||
| struct Category end | ||
| struct Tags end | ||
| struct MyBool end | ||
| struct NewInt end | ||
|
|
||
| @mtkmodel TestMetadataModel begin | ||
| @metadata begin | ||
| Author = "Test Author" | ||
| MyVersion = "1.0.0" | ||
| License = "MIT" | ||
| Category => "example" | ||
| Tags = ["test", "demo", "metadata"] | ||
| MyBool => false | ||
| NewInt => 1 | ||
| end |
There was a problem hiding this comment.
Why do these have to be structs? Can we have some docs around this?
There was a problem hiding this comment.
I was requested this API to avoid potential key overwrites. The setmetadata utility also explicitly only has dispatches for ::DataType. The intial version of the PR was a more standard Dict{Symbol, Any}. We could have checked for existing keys when setting too imo, and reject overwrites that way.
There was a problem hiding this comment.
This is in line with how symbolic metadata is stored. They don't have to be structs, just types. The reason is that this automatically avoids name conflicts since identically named types in different modules are different types, avoiding the necessity for name mangling.
Checklist
contributor guidelines, in particular the SciML Style Guide and
COLPRAC.
Additional context
Ref #3734
Add any other context about the problem here.
Are there cases where there are expression evaluations anywhere for the model parsing? Currently everything is delayed, but that means local variables are not referred to properly, and
escwill simply convert the escaped expression, whereas we need it as parse/ expansion time.