some fixes for compilation #27898
some fixes for compilation #27898
Conversation
| @@ -601,7 +601,7 @@ static void jl_compilation_sig( | |||
| for (i = 0; i < np; i++) { | |||
| jl_value_t *elt = jl_tparam(tt, i); | |||
| jl_value_t *decl_i = jl_nth_slot_type(decl, i); | |||
| size_t i_arg = (i <= nargs ? i : nargs); | |||
| size_t i_arg = (i < nargs - 1 ? i : nargs - 1); | |||
There was a problem hiding this comment.
Readability+C? You're funny. The min function can't exist in C since it doesn't have templates, and fmin is a floating point function.
src/gf.c
Outdated
| if (definition->isva ? np <= nargs : np != nargs) | ||
| return 0; | ||
| if (np != nargs) { | ||
| if (!definition->isva || (np <= nargs && jl_is_vararg_type(jl_tparam(type, np - 1)))) { |
There was a problem hiding this comment.
I don't quite understand this condition. Why does the vararg matter?
There was a problem hiding this comment.
Only Vararg functions may be compiled for Vararg signatures
There was a problem hiding this comment.
Right, but isn't this asking the opposite?
There was a problem hiding this comment.
This (should?) check the condition up to the "only" clause. We should also check for the rest of that somewhere, but currently appear to be probably missing that check. It's not particularly important.
There was a problem hiding this comment.
Right, so this disallows anything of the form say Tuple{typeof(f), Int...}, for say f(a::Int, b::Int...) which may be what you intended, but I thought you may have intended the opposite.
There was a problem hiding this comment.
Right, this check is for ensure that we have enough parameters to form a Vararg. For np == nargs, we expect to just compile a concrete signature
A type-assert is usually a bad idea, unless the RHS is a concrete type.
No description provided.