Use of @assert in Core.Compiler can result in unintended errors

[maleadt]

For example

julia/base/compiler/types.jl

Lines 348 to 350 in 9b73611

	# If they didn't pass typemax(UInt) but passed something more subtly
	# incorrect, fail out loudly.
	@assert world <= get_world_counter()

Triggering this shows

world = Base.get_world_counter() = 0x0000000000007b86
Internal error: encountered unexpected error in runtime:
MethodError(f=Base.string, args=(Expr(:call, :<=, :world, Expr(:call, :get_world_counter)),), world=0x00000000000015aa)
jl_method_error_bare at /home/tim/Julia/src/julia/src/gf.c:2204
jl_method_error at /home/tim/Julia/src/julia/src/gf.c:2222
jl_lookup_generic_ at /home/tim/Julia/src/julia/src/gf.c:3053 [inlined]
ijl_apply_generic at /home/tim/Julia/src/julia/src/gf.c:3068
macro expansion at ./error.jl:231 [inlined]
#NativeInterpreter#287 at ./compiler/types.jl:360
NativeInterpreter at ./compiler/types.jl:349 [inlined]
typeinf_ext_toplevel at ./compiler/typeinfer.jl:1078

It's actually the rendering by @assert that fails, but the point remains.

[vtjnash]

There are a lot of features not available in the Compiler. We don't need an issue for each of them. In this case, only the 2 argument version of @assert false "message" is available.

[maleadt]

We don't need an issue for each of them.

I wasn't filing an issue about the missing feature, but about the unintended error leaking into user code.

[Keno]

Could we just define string(args...) = invokelatest(Core.Main.Base.string, args...) in Core? It still won't work during bootstrap, but at least it'd give reasonable errors once everything is running.

[maleadt]

I guess this isn't considered an issue.

[Keno]

I guess this isn't considered an issue.

No, I agree it's annoying, but we should just make the @assert macro do something reasonable rather than rewriting all the uses of it to something worse.