Callsite `@inline` doesn't like `val` argument

[maleadt]

MWE:

macro foo()
    quote
        function bar(val)
            @inline identity(nothing)
        end
    end
end
@foo

ERROR: LoadError: syntax: local variable name "#9#val" conflicts with an argument

Using any other variable name than val makes this work. Probably suspect:

julia/base/expr.jl

Lines 846 to 859 in 7eb9615

	function annotate_meta_def_or_block(@nospecialize(ex), meta::Symbol)
	inner = unwrap_macrocalls(ex)
	if is_function_def(inner)
	# annotation on a definition
	return esc(pushmeta!(ex, meta))
	else
	# annotation on a block
	return Expr(:block,
	Expr(meta, true),
	Expr(:local, Expr(:(=), :val, esc(ex))),
	Expr(meta, false),
	:val)
	end
	end

[TakshDhabalia]

anyone working on this issue ? would like to talk about this and make changes
Thanks!

[maleadt]

The issue isn't assigned, so I don't think anybody is working on this.

[simeonschaub]

This is probably not a great first issue though, since the source is likely a macro hygiene bug

[xlxs4]

For completeness' sake annotate_meta_def_or_block is also called when @noinline is used, so this error is present if you substitute @inline with @noinline. The problem has to do with annotate_meta_def_or_block itself.
The expanded macro will be

quote
    function var"#<x gensym>#<function name>"(var"#<y gensym>#<argument name>")
        begin
            $(Expr(:<meta>, true))
            local var"#<z gensym>#val" = <escaped expression>
            $(Expr(:<meta>, false))
            var"#<z gensym>#val"
        end
    end
end

The gensym for the argument in the case of val is the same inside the begin block

[xlxs4]

Same issue for @inbounds, defined as:

macro inbounds(blk)
    return Expr(:block,
        Expr(:inbounds, true),
        Expr(:local, Expr(:(=), :val, esc(blk))),
        Expr(:inbounds, :pop),
        :val)
end

[xlxs4]

I wonder, why use true/false for @(no)inline, but true/:pop for @inbounds?

[xlxs4]

I think this is unrelated to hygiene, and is intended behavior, as it has to do with what gets escaped and what doesn't. If that's the case, we can leave everything as-is, or prefix val with an underscore? In case we'd do the latter, we should remember to also prefix the respective variable in @noinline and @inbounds.

[Pangoraw]

Probably related to #43151. @foo and @inline share the same gensym for val.