Fix hasmethod for worlds greater than the latest world
#59808
Conversation
base/reflection.jl
Outdated
| end | ||
|
|
||
| function hasmethod(f, t, kwnames::Tuple{Vararg{Symbol}}; world::UInt=get_world_counter()) | ||
| function hasmethod(f, t, kwnames::Tuple{Vararg{Symbol}}; world::UInt=typemax(UInt)) |
There was a problem hiding this comment.
I think get_world_counter is still the preferred default, typemax is a bit of a hack.
There was a problem hiding this comment.
typemax(UInt) as a token for "latest world" though
This has been deprecated for several releases now (it was a legacy hack) . I would rather not bring back this hack, since it is not correctly reflecting what this actually means, and is a nonsense special case.
There was a problem hiding this comment.
get_world_counter is not accessible everywhere (sometimes it is defined to be typemax, indicating that the world is unbounded dynamically), so this will lead to potentially more confusing errors and more buggy code. I really don't think we should do attempt to make a special case to make hasmethod error.
|
Fair enough, although we should still update the docstring and either remove the "code reflection cannot be used from generated functions" check or add it to the non-keyword-argument case. |
mlechu
force-pushed
the
hasmethod-world
branch
from
40b0c1b to
df22587
Compare
3 participants
hasmethodreturns false when theworldargument specified is greater than the global latest world. We often usetypemax(UInt)as a token for "latest world" though; JuliaLowering uses this to check if there's a provenance-preserving macro of a given name before falling back to the Expr macro (a bit of a hack, but losing provenance with every@nospecializewould be unfortunate for JETLS, and we want to call our own version of@ccall).@Keno suggested throwing an error for too-large worlds other than typemax, which is implemented here.