Re: Native compilation on MPS branch

["Gerd Möllmann" <gerd.moellmann@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 325 bytes --]

Andrea Corallo <acorallo@gnu.org> writes:

> Cool, keep us updated.

Daily update ;-).

Haven't got much further, and can use a little break now.

I don't know if it makes sense to attach this, but maybe it does... I'm
sometimes keeping notes when debugging (which I find useful when
debugging more than 1 thing at a time).


[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: debugging --]
[-- Type: text/x-org, Size: 3900 bytes --]

:PROPERTIES:
:ID:       544E65F2-9EED-4D0E-812F-79205DA0040F
:END:
#+title: Debugging Ideas
* Assertion in string_char_and_length

In read_make_string a string is made that later leads to the assertion
in string_char_and_len.

#+begin_src sh
(lldb) p s
(const char *) 0x000000016fdf8588 "macroexp--cons"
#+end_src

Symbol whose name is that string is made in read:

#+begin_src sh
result = dwim-print result
(Lisp_Object) 0x00000000080935f8 (struct Lisp_Symbol *) $7 = 0x0000000108cc7568
#+end_src

Whole thing read looks like

#+begin_src sh
(lldb) xdebug_print retval
[let byte-compile-form-stack cons (byte-compile-form-stack) nil make-symbol "shared" "unshared" "tail" "new-el" let* (nil) while...
#+end_src

The read is in from load_static_obj, which is called from
load_comp_unit.

#+begin_src c
      if (!loading_dump)
	{
	  comp_u->optimize_qualities =
	    load_static_obj (comp_u, TEXT_OPTIM_QLY_SYM);
	  comp_u->data_vec = load_static_obj (comp_u, TEXT_DATA_RELOC_SYM);
	  comp_u->data_impure_vec =
	    load_static_obj (comp_u, TEXT_DATA_RELOC_IMPURE_SYM);

	  if (!NILP (Vpurify_flag))
	    /* Non impure can be copied into pure space.  */
	    comp_u->data_vec = Fpurecopy (comp_u->data_vec);

	  if (!NILP (Vpurify_flag))
	    /* Non impure can be copied into pure space.  */
	    comp_u->data_vec = Fpurecopy (comp_u->data_vec);
#+end_src

The cu is allocated in MPS, purify is nil

#+begin_src sh
(lldb) p comp_u
(Lisp_Native_Comp_Unit *) 0x0000000108cc51f0
(lldb) p is_mps (comp_u)
(bool) true
(lldb) p globals.f_Vpurify_flag
(Lisp_Object) NULL
#+end_src

Objects are copied

#+begin_src c
    Lisp_Object *data_relocs = dynlib_sym (handle, DATA_RELOC_SYM);
    Lisp_Object *data_imp_relocs = comp_u->data_imp_relocs;
    ...
    EMACS_INT d_vec_len = XFIXNUM (Flength (comp_u->data_vec));
    for (EMACS_INT i = 0; i < d_vec_len; i++)
      data_relocs[i] = AREF (comp_u->data_vec, i);
#+end_src

data_relocs points somewhere into the loaded dylib, I'd say

#+begin_src sh
  (lldb) p data_relocs
  (Lisp_Object *) 0x0000000100ebcfd8 (struct Lisp_Symbol *) $26 = 0x0000000201af0f48
  (lldb) p is_mps (data_relocs)
  (bool) false
  (lldb) p pdumper_object_p (data_relocs)
  (bool) false
  (lldb) p is_pure (data_relocs)
  (bool) false
#+end_src

data_imp_relocs too.

#+begin_src sh
  (lldb) p *comp_u
(Lisp_Native_Comp_Unit) {
  header = (size = 4611686018863607815)
  file = 0x0000000108cc50d4 (struct Lisp_String *) $28 = 0x0000000108cc50d0
  optimize_qualities = 0x0000000108cc53bb (struct Lisp_Cons *) $29 = 0x0000000108cc53b8
  lambda_gc_guard_h = 0x0000000108cc5255 (struct Lisp_Vector *) $30 = 0x0000000108cc5250
  lambda_c_name_idx_h = 0x0000000108cc52ad (struct Lisp_Vector *) $31 = 0x0000000108cc52a8
  data_fdoc_v = NULL
  data_vec = 0x0000000108cd000d (struct Lisp_Vector *) $32 = 0x0000000108cd0008
  data_impure_vec = NULL
  data_imp_relocs = 0x0000000100ebda00 (struct Lisp_Symbol *) $33 = 0x0000000201af1970
  loaded_once = false
  load_ongoing = true
  handle = 0x0000000088eb8210
}
#+end_src

That all looks like I would have expected.
(Note to self: remember register_comp_unit...)
That all happens in (require macroexp)

Continue. Landing in copy_sequence with invalid string.

#+begin_src sh
(struct Lisp_String) {
  u = {
    s = {
      size = 4546911032
      size_byte = -1
      intervals = NULL
      data = 0x0000000108cca740 "H\xdd\U00000003\U0000000f\U00000001"
    }
    next = 0x000000010f045b38
    gcaligned = '8'
  }
}
#+end_src

(Note to self: xbacktrace broken. By LLVM 18?)

Coming from print_prune_charset <- print_object <- print <-
Fprin1 of that string, ..., backtrace_frame_apply <- mapbacktrace ....
<- signal_or_quit.

From (error "Not enough aruments in format string") in styled_format.
args[0] is the kaputte string.

From readevalloop_eager_expand_eval, call of macroexpand.

And finito by calling xdebug_print, process was killed.