Re: User-defined integers: Asking for feedback

I thought about compact-instance-header, a lot of platforms can't
utilize the whole 64-bit address space, so pointers only take up 56
bits or lower. Some even ignore high bits (but that would require
unaligned loads).

On Sun, Dec 7, 2025 at 9:24 PM Douglas Katzman via Sbcl-devel
<[email protected]> wrote:
>
> Hi Philipp, a concern that I've not voiced before is that I also want to claim the "available" lowtag that you're taking, but in a way that is more broadly applicable to SBCL as a whole. Specifically I want to revise the implementation of immobile-symbol and compact-instance-header to work on platforms other than x86-64 and with the address space restriction removed.  Currently these only work on x86-64 because a space can be forcibly mapped below 2GiB.  To lift the restriction, and represent symbols, layouts, and perhaps other important primitive objects as NIL-relative pointers, SBCL itself will quite likely need to use a new lowtag or 2.  (I've never actually made progress on this; it's theoretical.)  But when or if I do it, you can't reserve lowtags for user code, because compressed pointers (to borrow the JVM term), i.e. NIL-relative objects, will become more important than the accidental discovery that you can take lowtag N for yourself.
> Since I never did anything, I didn't suggest you stop trying to work on this patch, but I'm unconvinced that your effort to land it in the main tree is going to reach a consensus.  This is a separate question from whether instructing random users how to take advantage of a particular detail of the object representation is a generally good idea.
> Doug
> _______________________________________________
> Sbcl-devel mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/sbcl-devel


_______________________________________________
Sbcl-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/sbcl-devel