From mboxrd@z Thu Jan  1 00:00:00 1970
Path: eternal-september.org!news.eternal-september.org!.POSTED!not-for-mail
From: Simon Wright <simon@pushface.org>
Newsgroups: comp.lang.ada
Subject: Re: spurious error with GNAT 13.2.0 on Intel macOS 17.2
Date: Fri, 15 Dec 2023 15:23:31 +0000
Organization: A noiseless patient Spider
Message-ID: <lyo7ermqvg.fsf@pushface.org>
References: <ku1ijrFhcjkU1@mid.individual.net>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 8bit
Injection-Info: dont-email.me; posting-host="2c895444440ef578c1d5ce6d6b06a357";
	logging-data="2059676"; mail-complaints-to="abuse@eternal-september.org";	posting-account="U2FsdGVkX18y58Pw4/QqKM17uDCEfw7RtTxfowscaiw="
User-Agent: Gnus/5.13 (Gnus v5.13) Emacs/28.2 (darwin)
Cancel-Lock: sha1:ZErcHW3vAD5CbgXzY5Mj9nxjRF4=
	sha1:D6N3DtETV9zGq9UBWtR7hIbcYco=
Xref: news.eternal-september.org comp.lang.ada:65918
List-Id: <comp.lang.ada>

moi <findlaybill@blueyonder.co.uk> writes:

> The 17.2 update is accompanied by updated Command Line Tools,
> so, having made a copy of the current CLTs, I let it update.

I think you mean 14.2!

GCC 13/CLT 15.* still has issues with exceptions, eg this unresolved
issue: https://github.com/iains/gcc-13-branch/issues/10, and the
ld-classic dance fixes them as far as I can tell.

I'm working with this fix: https://github.com/simonjwright/xcode_15_fix
which is included in my GCC 13.2.0 build:
https://github.com/simonjwright/distributing-gcc/releases/tag/gcc-13.2.0-aarch64

> I get:
>
> A. with  ”-largs -Wl,-ld_classic” in the linker parameters:
>
> /Users/wf/KDF9/emulation/Testing: failure
> raised PROGRAM_ERROR : Open failed to get RW access
>
> This is what should happen.
>
> B. recompiled and relinked without ”-largs -Wl,-ld_classic”:
>
> /Users/wf/KDF9/emulation/Testing: failure
> raised CONSTRAINT_ERROR : erroneous memory access
>
> OOPS!
>
> Strangely, this is now the ONLY one out of dozens of regression tests
> that fails in this software configuration.
> Previously, they all failed at the link stage.

Yes, the 15.1 betas fixed the linking problem. I've no idea why some
exceptions don't get caught/raise memory access errors. I do think that
the aarch64 GCC 14 has fixed the problem.