Re: pure-fns in byte-opt.el

[Mark Oteiza <mvoteiza@udel.edu>]

On 29/07/17 at 10:24am, Eli Zaretskii wrote:
>> Date: Fri, 28 Jul 2017 21:24:07 -0400
>> From: Mark Oteiza <mvoteiza@udel.edu>
>> Cc: emacs-devel@gnu.org
>>
>> >> #3  0x000000000060f09b in Ffuncall (nargs=4, args=0x7fffffffb688) at eval.c:2742
>> >>         fun = XIL(0x17aee65)
>> >>         original_fun = XIL(0x20b0da0)
>> >>         funcar = XIL(0xce8cc0)
>> >>         numargs = 3
>> >>         val = make_number(897379714801469509)
>> >>         count = 58
>> >
>> >In this frame #3, what are the values of the important variables?
>> >
>> >  (gdb) frame 3
>> >  (gdb) pp original_fun
>>
>> unidata-gen-table-word-list
>>
>> >  (gdb) pp args[1]
>> >  (gdb) pp args[2]
>> >  (gdb) pp args[3]
>>
>> decomposition
>> 5
>> unidata-split-decomposition
>>
>> >  (gdb) pp funcar
>> >  (gdb) pp fun
>>
>> #<INVALID_LISP_OBJECT 0x00ce8cc0>
>
>Thanks.
>
>So I think the problem happens in unidata-word-list-compress, and it
>happens because make-vector, which that function calls always returns
>the same vector, so the vectors used by that function and created by
>it are all messed up.

I don't follow: why does it always return the same vector?  That
particular call to make-vector should be unaffected by make-vector being
marked pure, because its arguments aren't constants.  OTOH, the form
(make-vector 128 nil) would indeed get byte-compiled into the constant
[nil nil … nil]