Fix bugs in ResourceCodeDomSerializer.SerializationResourceManager

[halgab]

As discussed in #9457.

#8332 introduced bugs in ResourceCodeDomSerializer.SerializationResourceManager, in method GetResourceSet. When we fail to get a resource reader, the original code used to return a resource set for the invariant culture, and null otherwise. After #8332,,

• we're returning null the first time we fill the ResourceTable cache
• for the non-invariant culture case, we started filling the ResourceTable cache with a static dictionary, which looks like a pretty big issue

These bugs were introduced in .Net 8 and we still have a chance to fix them before the release

@dreddy-work, @elachlan

Microsoft Reviewers: Open in CodeFlow

[dreddy-work]

• we started filling the ResourceTable cache with a static dictionary, which looks like a pretty big issue

@halgab, i see original code before @elachlan's change was also doing this. Can you point me to the original code where it wasn't?

[halgab]

Before #8332, s_resourceSetSentinel was not a dictionary or a hashtable but an object (see line 21). Then on line 505, we checked that the content of ResourceTable was actually a Hashtable, which proved to be false when we were returning s_resourceSetSentinel. The assert on line 509 is here to support the fact that if ResourceTable returned anything but a Hashtable, then it must be s_resourceSetSentinel

[Tanya-Solyanik]

@halgab - is it possible to add a unit test for this scenario?

[halgab]

I agree that would be ideal, but right now code coverage of serializers is so scarce I wouldn't really know where to start...

[dreddy-work]

We can track test debt under separate issue. @Tanya-Solyanik, you good with changes here?

[elachlan]

Related: #719, #966

[lonitra]

LGTM