Different outputs under net45 vs netcoreapp2.1

[Drew Noakes]

While working on new tooling for testing against the image library, I was using a netcoreapp2.1 executable and found that MetadataExtractor produces different output under net45 and netcoreapp2.1.

For example:

diff --git a/png/ImageTestSuite/metadata/dotnet/m2-71915ab0b1cc7350091ef7073a312d16.png.txt b/png/ImageTestSuite/metadata/dotnet/m2-71915ab0b1cc7350091ef7073a312d16.png.txt
index df52416..62a18f6 100644
--- a/png/ImageTestSuite/metadata/dotnet/m2-71915ab0b1cc7350091ef7073a312d16.png.txt
+++ b/png/ImageTestSuite/metadata/dotnet/m2-71915ab0b1cc7350091ef7073a312d16.png.txt
@@ -1,6 +1,8 @@
 FILE: m2-71915ab0b1cc7350091ef7073a312d16.png
 TYPE: PNG
 
+[ERROR: PNG-zTXt] Exception decompressing zTXt chunk with keyword "T�umb::URï": The archive entry was compressed using an unsupported compression method.
+
 [PNG-IHDR - 0x0001] Image Width = 60
 [PNG-IHDR - 0x0002] Image Height = 34
@@ -34,8 +36,6 @@ TYPE: PNG
 
 [PNG-zTXt - 0x000d] Textual Data = Thumb::Size: 10.9355kb
 
-[PNG-zTXt - 0x000d] Textual Data = T�umb::URï: 
-
 [File Type - 0x0001] Detected File Type Name = PNG
 [File Type - 0x0002] Detected File Type Long Name = Portable Network Graphics

It seems that most differences are related to decompression algorithm differences between frameworks.

This isn't likely a big deal as most of these differences appear related to invalid data, but it does make me wonder if our test harness should compare not just Java and .NET, but also different supported target frameworks.