[SYCL][clang] Clarify warning message for potential property conflicts

[steffenlarsen]

The compiler currently warns when properties are used in conjunction with SYCL 2020 properties, warning about potentially ignored conflicting properties. This commit makes the warning messages a little clearer to help users understand what particular properties may be of concern.

[bader]

Thanks!

[aelovikov-intel]

Do we need to extend the test with a scenario when multiple conflicting properties are possible?

[steffenlarsen]

Do we need to extend the test with a scenario when multiple conflicting properties are possible?

We could! I would expect it to just issue multiple warnings for the same line. That said, it is not directly related to these changes, so would you mind if we did it in a follow-up?

[aelovikov-intel]

it is not directly related to these changes

I somewhat disagree with that statement. "conflicting properties are ignored" could be read as anything conflicting will be ignored (even if a stretch, we can still justify if more is ignored). Now that you're specific in what is being ignored, it would be harder to justify if more properties are being ignored yet no warnings are emitted for them.

[steffenlarsen]

it is not directly related to these changes

I somewhat disagree with that statement. "conflicting properties are ignored" could be read as anything conflicting will be ignored (even if a stretch, we can still justify if more is ignored). Now that you're specific in what is being ignored, it would be harder to justify if more properties are being ignored yet no warnings are emitted for them.

Right, but functionality nothing has changed, so we would not be testing any logic introduced by this PR, just making sure that the messages fit the current behavior. If you really feel strongly about it, I can do it as part of this patch, but it does feel a little misleading to bundle them together as it might make it look like there's a functional change to this. @elizabethandrews - What are your thoughts on this?

[aelovikov-intel]

How urgent is this PR? We could just make it the second one in the series, after extra tests.

[steffenlarsen]

How urgent is this PR? We could just make it the second one in the series, after extra tests.

Faster to implement than to argue. #20652 adds the cases.

[elizabethandrews]

it is not directly related to these changes

I somewhat disagree with that statement. "conflicting properties are ignored" could be read as anything conflicting will be ignored (even if a stretch, we can still justify if more is ignored). Now that you're specific in what is being ignored, it would be harder to justify if more properties are being ignored yet no warnings are emitted for them.

Right, but functionality nothing has changed, so we would not be testing any logic introduced by this PR, just making sure that the messages fit the current behavior. If you really feel strongly about it, I can do it as part of this patch, but it does feel a little misleading to bundle them together as it might make it look like there's a functional change to this. @elizabethandrews - What are your thoughts on this?

Follow-up PR works for me.