Consider referencing System.Memory

[Rob-Hague]

Perhaps you've already considered it, but in case you haven't:

The library has many duplicate code paths, one using a Span implementation for higher targets (.NET) and one using arrays for lower targets (.NET Framework and .NET Standard 2.0)

Microsoft provide (and support) the System.Memory package, which allows use of Span etc. on .NET Framework and .NET Standard 2.0. It is fairly common for multi-targeting libraries to reference the package on lower targets. The source code is at https://github.com/dotnet/maintenance-packages/tree/main/src/System.Memory.

This reference would allow many code paths (and public API) to be unified.

Practically, the initial change would look like:

diff --git a/crypto/src/BouncyCastle.Crypto.csproj b/crypto/src/BouncyCastle.Crypto.csproj
index 8b7c8852..166dc1ae 100644
--- a/crypto/src/BouncyCastle.Crypto.csproj
+++ b/crypto/src/BouncyCastle.Crypto.csproj
@@ -88,6 +88,10 @@
     <None Include="..\..\README.md" Pack="true" PackagePath="\" />
   </ItemGroup>

+  <ItemGroup Condition=" '$(TargetFramework)' == 'net461' or '$(TargetFramework)' == 'netstandard2.0' ">
+    <PackageReference Include="System.Memory" Version="4.5.5" />
+  </ItemGroup>
+
   <ItemGroup>
     <PackageReference Include="Microsoft.NETFramework.ReferenceAssemblies" Version="1.0.3">
       <PrivateAssets>all</PrivateAssets>

Example unified diff (note Index/Range operators [..32] not supported):

diff --git a/crypto/src/crypto/modes/ChaCha20Poly1305.cs b/crypto/src/crypto/modes/ChaCha20Poly1305.cs
index 56bef5e9..ca982f9d 100644
--- a/crypto/src/crypto/modes/ChaCha20Poly1305.cs
+++ b/crypto/src/crypto/modes/ChaCha20Poly1305.cs
@@ -783,29 +783,16 @@ private ulong IncrementCount(ulong count, uint increment, ulong limit)
 
         private void InitMac()
         {
-#if NETCOREAPP2_1_OR_GREATER || NETSTANDARD2_1_OR_GREATER
             Span<byte> firstBlock = stackalloc byte[64];
             try
             {
                 mChacha20.ProcessBytes(firstBlock, firstBlock);
-                mPoly1305.Init(new KeyParameter(firstBlock[..32]));
+                mPoly1305.Init(new KeyParameter(firstBlock.Slice(0, 32)));
             }
             finally
             {
                 firstBlock.Fill(0x00);
             }
-#else
-            byte[] firstBlock = new byte[64];
-            try
-            {
-                mChacha20.ProcessBytes(firstBlock, 0, 64, firstBlock, 0);
-                mPoly1305.Init(new KeyParameter(firstBlock, 0, 32));
-            }
-            finally
-            {
-                Array.Clear(firstBlock, 0, 64);
-            }
-#endif
         }
 
         private void PadMac(ulong count)

[Rob-Hague]

@peterdettman would you be open to this? Happy to get it started. As it's a big job, it can be done in pieces starting in the lower levels of the library.

It's also possible to keep the Index/Range syntax using a polyfill of those types and a reference to System.ValueTuple on net461.

[scott-xu]

Suggest ref https://www.nuget.org/packages/Microsoft.Bcl.Memory#readme-body-tab which includes Index and Range. It also depends on System.Memory.

[Rob-Hague]

Good idea. Unfortunately this library targets .NET Framework 4.6.1 which is out of support and that library targets 4.6.2, so we would at least still need an explicit reference to System.ValueTuple on net461

[scott-xu]

@peterdettman Can you please share your thoughts on this?

[paulomorgado]

Good idea. Unfortunately this library targets .NET Framework 4.6.1 which is out of support and that library targets 4.6.2, so we would at least still need an explicit reference to System.ValueTuple on net461

I don't think .NET Framework 4.6.1 should be supported because no operating system where it can run is supported.

Users using that are unlikely to be change the software.

Spans could be supported on all targets above that, anyway. It would be better for all others and not worst for this one.

But I'd still drop it.

[paulomorgado]

@peterdettman, anything I can do to help?