Re: Changing SS_Pool

[Simon Wright <simon@pushface.org>]

Lucretia <laguest9000@googlemail.com> writes:

> On Wednesday, 16 November 2016 22:11:49 UTC, Lucretia  wrote:
>> Hi,
>> 
>> Is it possible to make SS_Pool be an array of storage_elements
>> instead of just an integer? Seems pointless creating a separate
>> array and having that random integer just because the compiler
>> expects it.
>
> It seems to work, but whether it's legal or not, I'm not so sure.

The declaration is

private
   SS_Pool : Integer;
   --  Unused entity that is just present to ease the sharing of the pool
   --  mechanism for specific allocation/deallocation in the compiler

which I think means either that in a full RTS the ompiler has to do
stuff which it doesn't in a restricted RTS, or that it's a fossil from
some previous design.

I just looked into (a) a build using GNAT GPL 2016, (b) a build using
one of my cortex-gnat-rts runtimes. In (a), ss_pool occupies 4 bytes, in
(b) it's in a discarded input section. Both RTSs are compiled with
-ffunction-sections -fdata-sections, the difference is that in (b) my
project GPR explicity has those two switches in package Builder, and
-Wl,-gc-sections in package Linker. I wonder whether they should be in
the RTS's runtime.xml, so as to be used everywhere? Will ask AdaCore.