Base.summarysize(::Symbol) returns zero

[xlxs4]

From a conversation at Slack w/ @tecosaur:

julia> Base.summarysize(:hello)
0

summarysize is explicitly coded to return 0 for Symbols.

The docstring reads:

help?> Base.summarysize
  Base.summarysize(obj; exclude=Union{...}, chargeall=Union{...}) -> Int

  Compute the amount of memory, in bytes, used by all unique objects reachable from
  the argument.

Symbols are interned strings with a hash, and all subsequent references are a pointer to the interned value

julia/src/julia.h

Lines 140 to 148 in 86cba99

	// Symbols are interned strings (hash-consed) stored as an invasive binary tree.
	// The string data is nul-terminated and hangs off the end of the struct.
	typedef struct _jl_sym_t {
	JL_DATA_TYPE
	_Atomic(struct _jl_sym_t*) left;
	_Atomic(struct _jl_sym_t*) right;
	uintptr_t hash; // precomputed hash value
	// JL_ATTRIBUTE_ALIGN_PTRSIZE(char name[]);
	} jl_sym_t;

julia/src/symbol.c

Lines 35 to 49 in 86cba99

	static jl_sym_t *mk_symbol(const char *str, size_t len) JL_NOTSAFEPOINT
	{
	jl_sym_t *sym;
	size_t nb = symbol_nbytes(len);
	jl_taggedvalue_t *tag = (jl_taggedvalue_t*)jl_gc_perm_alloc(nb, 0, sizeof(void*), 0);
	sym = (jl_sym_t*)jl_valueof(tag);
	// set to old marked so that we won't look at it in the GC or write barrier.
	jl_set_typetagof(sym, jl_symbol_tag, GC_OLD_MARKED);
	jl_atomic_store_relaxed(&sym->left, NULL);
	jl_atomic_store_relaxed(&sym->right, NULL);
	sym->hash = hash_symbol(str, len);
	memcpy(jl_symbol_name(sym), str, len);
	jl_symbol_name(sym)[len] = 0;
	return sym;
	}

So why would we have summarysize report 0? Shouldn't summarysize(x) == sizeof(x) for x::Symbol?

That method instance was originally added in 8322d79, with a comment reading # these three cases override the exception that would be thrown by Core.sizeof. Looking at the current source code for summarysize, I can't find any sizeof calls that would throw if that definition wasn't there. Is this a historical artifact? I've tried to find the commit where the comment was removed (e.g., it's still there in 34236e7) but git log -c -S and co. failed me.

Removing the mi doesn't seem to break anything, and we get

julia> sizeof(:hello)
5

julia> Base.summarysize(:hello)
5

julia> sizeof("hello")
5

julia> Base.summarysize("hello")
13

instead, which to me feels more close to what's actually going on behind the hood.