ASAN fixes

[maleadt]

No description provided.

[Keno]

I'm confused why the datalayout commit is necessary. All of these have nfields set to zero, so it should never attempt to load any pointer past the end of the declared struct. Which code is doing this reading?

[maleadt]

nfields set to zero

The number of pointers isn't set to 0, so this adds one:

julia/src/staticdata.c

Lines 873 to 874 in 08ebedc

	char *flddesc = (char*)dt->layout;
	size_t fldsize = sizeof(jl_datatype_layout_t) + nf * fieldsize + (np << dt->layout->fielddesc_type);

[Keno]

I think the idea was that if first_ptr is -1, then those trailing bytes aren't allocated.

[maleadt]

I think the idea was that if first_ptr is -1, then those trailing bytes aren't allocated.

Ah, OK. Yeah, that wasn't clear to me at all. I'll have a look at avoiding those reads then.

[maleadt]

I don't really get the layout code (i.e. the specific semantics of nfields/npointers/first_ptr, @vtjnash?), so I just implemented #34389 (comment).