From mboxrd@z Thu Jan  1 00:00:00 1970
X-Spam-Checker-Version: SpamAssassin 3.4.5-pre1 (2020-06-20) on
	ip-172-31-74-118.ec2.internal
X-Spam-Level: 
X-Spam-Status: No, score=-1.9 required=3.0 tests=BAYES_00 autolearn=ham
	autolearn_force=no version=3.4.5-pre1
Path: eternal-september.org!reader02.eternal-september.org!aioe.org!yy9MKEJN2ULhWGfnfq4v5w.user.gioia.aioe.org.POSTED!not-for-mail
From: Simon Wright <simon@pushface.org>
Newsgroups: comp.lang.ada
Subject: Elaboration code, aggregates
Date: Sun, 28 Mar 2021 20:41:25 +0100
Organization: Aioe.org NNTP Server
Message-ID: <lylfa712xm.fsf@pushface.org>
NNTP-Posting-Host: yy9MKEJN2ULhWGfnfq4v5w.user.gioia.aioe.org
Mime-Version: 1.0
Content-Type: text/plain
X-Complaints-To: abuse@aioe.org
User-Agent: Gnus/5.13 (Gnus v5.13) Emacs/27.1 (darwin)
Cancel-Lock: sha1:BQRyQoIkRRdDHz5dbcqF8KP31Q0=
X-Notice: Filtered by postfilter v. 0.9.2
Xref: reader02.eternal-september.org comp.lang.ada:61662
List-Id: <comp.lang.ada>

In June 2020, Luke A. Guest was having trouble with getting the compiler
to place constant data into the data section without elaboration code.
https://groups.google.com/g/comp.lang.ada/c/B2NA-qjCJuM/m/4ykywZWZAgAJ

During preliminary work for FSF GCC 11, I found that this ARM interrupt
vector (which used to compile happily without needing elaboration code)
no longer would:
https://github.com/simonjwright/cortex-gnat-rts/blob/master/stm32f4/adainclude/startup.adb#L231

   Vectors : array (-14 .. Ada.Interrupts.Names.FPU_IRQ) of Handler :=
     (-9 .. -6 | -4 .. -3 => null,                      -- reserved
      -14                 => Dummy_Handler'Access,      -- NMI
      -13                 => HardFault_Handler'Access,  -- HardFault
      -12                 => Dummy_Handler'Access,      -- MemManagement
      -11                 => Dummy_Handler'Access,      -- BusFault
      -10                 => Dummy_Handler'Access,      -- UsageFault
      -5                  => SVC_Handler'Access,        -- SVCall
      -2                  => PendSV_Handler'Access,     -- PendSV
      -1                  => SysTick_Handler'Access,    -- SysTick
      others              => IRQ_Handler'Access)
     with
       Export,
       Convention         => Ada,
       External_Name      => "isr_vector";
   pragma Linker_Section (Vectors, ".isr_vector");

and Arduino Due clock startup ddn't:
https://github.com/simonjwright/cortex-gnat-rts/blob/master/arduino-due/adainclude/startup-set_up_clock.adb#L48

   PMC_Periph.CKGR_MOR := (KEY      => 16#37#,
                           MOSCXTEN => 1,     -- main crystal oscillator enable
                           MOSCRCEN => 1,     -- main on-chip rc osc. enable
                           MOSCXTST => 8,     -- startup time
                           others   => <>);

On investigating, it turns out that FSF GCC 11 **AND** GNAT CE 2020 have
lost the ability to assign aggregates as a whole; instead, they assign
the record components one-by-one.

The reason for the Arduino Due failure is that the PMC hardware requires
that each write to the CKGR_MOR register contain that value of KEY! so
the sequence is

read the register (KEY is always returned as 0)
overwrite the KEY field
write the register back
read the register, KEY is 0
overwrite the MOSCXTEN field
write the register back, KEY is 0 so inoperative
etc (including the 'others => <>' components).

Bug report raised:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99802