When using HandleInner() not all AggregateExceptions are being checked

[sideproject]

When handling exceptions that that are wrapped in an AggregateException, it looks like only the first exception is recursively searched for a matching predicate. If there are two (or more) exceptions in the InnerExceptions collection, they are not searched.

//Example Policy
var policy = Policy
    .HandleInner<DivideByZeroException>()
    .Retry(3, (exception, _) => passedToOnRetry = exception);

//The policy will find and retry the nested DivideByZeroException
Exception willRetryException = new AggregateException(
    new Exception("First: With Inner Exception"),
        new DivideByZeroException()));
    new Exception("Second: Without Inner Exception",

//The policy will NOT find and retry the nested DivideByZeroException
Exception willNotRetryException = new AggregateException(
    new Exception("First: Without Inner Exception"),
    new Exception("Second: With Inner Exception",
        new DivideByZeroException()));

I was able to make this work by modifying the HandleInner function in src/Polly/PolicyBuilder.OrSyntax.cs:55.

internal static ExceptionPredicate HandleInner(Func<Exception, bool> predicate)
{
    return exception =>
    {
        if (exception is AggregateException aggregateException)
        {
            foreach (var innerException in aggregateException.Flatten().InnerExceptions)
            {
                var matchedInAggregate = HandleInnerNested(predicate, innerException);
                if (matchedInAggregate != null)
                    return matchedInAggregate;
            }
        }

        return HandleInnerNested(predicate, exception);
    };
}

I'm happy to submit a pull request with tests for this if you're willing to entertain it.

[sideproject]

Just checking in - is this change something that you'd entertain? I'm happy to submit a pull request for discussion.

[martincostello]

This seems reasonable to me, but I'm not sure whether or not it could be considered a breaking behavioural change or not.

I guess if it's a minor enough code change to make you could open a PR for it and solicit further comment/opinion.

[sideproject]

Understood, I've created PR #822 for your review.

The CI appears to have failed but doesn't seem to be related to my change - Could not load file or assembly 'YamlDotNet,

(Apologies for the second PR, I had some trouble with the first one so I started over)

[martincostello]

Fixed in v7.2.2.