Re: MPS: unable to build due to assertion violation in igc_dump_check_object_starts

[Pip Cet <pipcet@protonmail.com>]

On Wednesday, July 24th, 2024 at 08:01, Helmut Eller <eller.helmut@gmail.com> wrote:
> On Tue, Jul 23 2024, Pip Cet wrote:
> > I'm pretty sure it's the bignum thing, currently testing a fix.
> I think fix_wrapped_vec is also broken for 32-bit hardware:

Thanks for checking! I'm not sure I understand the problem, so please excuse the silly questions.

> First, is_aligned doesn't make any sense; client is neither a tagged
> pointer nor does it point to a header.

It points to the contents area of a struct Lisp_Vector. Since the Lisp vector header is aligned to 8 bytes, and 16 bytes in size including the MPS header, the client address also has to be aligned.

I believe the best thing to do here is to calculate the header address and assert, not check, that it is aligned. Unaligned addresses should never get here.

> Second, sizeof (struct Lisp_Vector) is 16 because the struct decl has
> GCALIGNED_STRUCT. The code should perhaps use header_size or
> offsetof (struct Lisp_Vector, content).

I'm not sure I understand. sizeof (struct Lisp_Vector), header_size, and offsetof (struct Lisp_Vector, content) are all 16.

(And, yes, that means wasting 32 bits per vector. Is that the problem?)

Pip