Re: How to add pseudo vector types

[Stephen Leake]

Eli Zaretskii <eliz@gnu.org> writes:

>> From: Yuan Fu <casouri@gmail.com>
>> Date: Fri, 23 Jul 2021 16:22:59 -0400
>> Cc: Stefan Monnier <monnier@iro.umontreal.ca>,
>>  Clément Pit-Claudel <cpitclaudel@gmail.com>,
>>  emacs-devel@gnu.org
>> 
>> > We must replace this function, if only because the MS-Windows build of
>> > Emacs uses a custom malloc implementation.  Does TS allow the client
>> > to use its own malloc?
>> 
>> Yes, in that case, we need to embed tree-sitter into Emacs, instead
>> of using it as a dynamic library, I think.
>> 
>> // Allow clients to override allocation functions
>> #ifndef ts_malloc
>> #define ts_malloc  ts_malloc_default
>> #endif
>> #ifndef ts_calloc
>> #define ts_calloc  ts_calloc_default
>> #endif
>> #ifndef ts_realloc
>> #define ts_realloc ts_realloc_default
>> #endif
>> #ifndef ts_free
>> #define ts_free    ts_free_default
>> #endif
>> 
>> How do we handle such thing in Emacs?
>
> We use xmalloc, which calls memory_full when allocation fails, which
> releases some spare memory we have for this purpose, and tells the
> user to save the session and exit.

I'm thinking about how this applies to wisi, when migrating to a module.

Ada has a built-in allocator; it's probably possible to change that, but
I'd like to understand exactly why we need to do that.

The Ada allocator throws an exception on allocation fail; is it
sufficient to turn that exception into an elisp signal, and arrange for
elisp to call memory_full (or take some other action, like killing the
parser)?

Another possible reason to change the Ada allocator is if we want to
expose Ada memory pointers directly to elisp, as Yuan Fu wants to do for
tree-sitter (I don't plan to do this for wisi). Does that require that
the pointers be allocated by the same allocator? I'm not clear what that
would mean for the garbage collector; is it then expected to recover the
tree-sitter-allocated memory for the tree? or does it ignore those lisp
objects?

-- 
-- Stephe