auto spatial inertia calculation fallback to documented default density is treated as an error

[SeanCurtis-TRI]

Environment

• OS Version: Ubuntu 24.04
• Source or binary build? Source from 15.3.0 (although the points below apply equally well to main).

Description

• Expected behavior: if <inertial auto="true"> is specified, density is documented as not required with a default value. Therefore, if I skip the child element <density> in my specification, I expect things to proceed apace, picking up the default value as documented (and as occurs throughout when we devolve to the default value).

• Actual behavior: When the default density is used, an error is dispatched. For any user of SDFormat that is taking errors seriously, this renders (what should be) a valid file, invalid.

Steps to reproduce

Simply parse something akin to:

<?xml version="1.0"?>
<sdf xmlns:drake="http://drake.mit.edu" version="1.8">
  <model name='model'>
    <link name='body'>
      <inertial auto="true"/>
      <collision name="collision">
        <geometry>
          <sphere><radius>1</radius></sphere>
        </geometry>
      </collision>
    </link>
  </model>
</sdf>

and look at the errors.

It seems the error should be demoted to warning (at most). Although, the warning can possibly be reconsidered completely in light of how other non-required elements there are with their own default values.

[SeanCurtis-TRI]

Caveat:

I've since read a bit more about the ParserConfig and the nature of Errors. I recognize the user of SDFormat configures the policy. If we have it configured to ERR, then every such Error will be treated as an error and then we have to filter them on the back end (for example, if I don't want the error indicated above to actually report as an error on my side, I have to filter on the ErrorCode and filter out the ones I want to ignore.

That said, I'm still slightly confused why an optional parameter with a default value would trigger an Error that requires disposition at all.

[jwnimmer-tri]

For the record, I read through #1299 as best I could and I couldn't find any discussion about why the use of the default value should be a warning. That seems antithetical the purpose of a default.