Duplicate device instances are ignored

[m8pple]

If a graph contains devices with duplicate device instances, then one of them
is silently ignored rather than raising an error.

[m8pple]

See PR #326 for test-case.

[mvousden]

If a graph contains devices with duplicate device instances, then one of them is silently ignored rather than raising an error.

Yeah, that's not supposed to happen.

@DrongoTheDog thoughts?

[mvousden]

On Sun, Jul 17, 2022 at 12:15:57PM +0100, Andrew Brown wrote:

You know, I have even have a vague recollection of this, but I thought
"Screw it, what's the harm in ignoring it?".

Also - philosophical point - the natuiral place to get this is in the
Validator, but it's not really a validation issue - you can't capture
name uniqueness in the validation grammar.

I think:

1. It could/should be ignored; it have never have any adverse effects
   (other than it indictates the monkey has cocked up).

2. It should therefore be posted as a Warning.

3. The natural place to do this is by the conditional that traps it,
   but I don't have the time right now to hunt it.

4. OK, you knew I would:

DevI_t * DS_XML::_DevI_t( .... )

My source, line 150:

if (pD->Def()!=0) { // Already defined?
fprintf(fd,"(%3u,-) W%3u: Device %s already defined on line %u\n",
pn->lin,++wcnt,dname.c_str(),pD->Def()); // Cockup. Bail.

So it is reported, in the microlog for the datastructure build, as a
build warning.

If you Post a warning to the LogServer, that's a bit more "in yer face",
but on the other hand if you do that you'll have to Post every other
microlog message for consistency, and that's exactly what the microlog
is there to avoid.

ADB

[mvousden]

Agree, up until:

If you do [Post a warning to the LogServer] you'll have to Post every other
microlog message for consistency, and that's exactly what the microlog is
there to avoid.

We should really be posting that there are warnings, and point the user to
the microlog if so. Let the operator walk to their own destruction if they
wish, but at least we told them so.

Nobody reads the micrologs unless prompted, anyway (by design).

[mvousden]

On Mon, Jul 25, 2022 at 11:01:22AM +0100, ADB wrote:

On 18/07/2022 10:01, Mark Vousden wrote:

Agree, up until:

If you do [Post a warning to the LogServer] you'll have to Post every other
microlog message for consistency, and that's exactly what the microlog is
there to avoid.
We should really be posting that there are warnings, and point the user to
the microlog if so. Let the operator walk to their own destruction if they
wish, but at least we told them so.
Fair point
Nobody reads the micrologs unless prompted, anyway (by design).
I do?

Anyway, by all means create a "microlog not empty" flag for each command
and Post a warning at the end if it's set.........

ADB