add stream test to csv test case 497

[pjfanning]

#497 issue only seems to happen with byte array input - work ok with InputStream input.

When byte array input to UTF8Reader, the _inputSource is always null and the _ioContext is null which is what is checked in canModifyBuffer(). Is there any way that this code can be adjusted to work for byte array input - or should we take the byte array and wrap as ByteArrayInputStream? - which would also fix the broken test.

The code that causes the issue.

jackson-dataformats-text/csv/src/main/java/com/fasterxml/jackson/dataformat/csv/impl/UTF8Reader.java

Lines 406 to 418 in 9a38ea2

	if (!canModifyBuffer()) {
	// 15-Aug-2022, tatu: Occurs (only) if we have half-decoded UTF-8
	// characters; uncovered by:
	//
	// https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=50036
	//
	// and need to be reported as IOException
	if (_inputSource == null) {
	throw new IOException(String.format(
	"End-of-input after first %d byte(s) of a UTF-8 character: needed at least one more",
	available));
	}
	}

Needed for a fuzz test to work but also breaks the 497 test.

[cowtowncoder]

If I remember correctly, it's not really so much the case of InputStream actually working, but it not hitting AAOBE -- since there's empty input buffer extending beyond content end.
Or maybe put another way: it working by reading out of actual content buffer.

One way to test this would be to pass byte[] where content doesn't end at buffer end but couple of bytes before, having empty slots at end.

I'll merge this: I understand the bug itself but don't have time to fix it. Adding more tests may not help with that part significantly.