DecodeRFC2047 fails to decode parameter list correctly

[mgnsk]

I've created the following testcase which fails:

diff --git a/internal/coding/headerext_test.go b/internal/coding/headerext_test.go
index 05d35ff..f92fd73 100644
--- a/internal/coding/headerext_test.go
+++ b/internal/coding/headerext_test.go
@@ -144,6 +144,7 @@ func TestRfc2047Decode(t *testing.T) {
 		{"quoted", "=?US-ASCII?q?Hello=20World?=", "Hello World"},
 		{"base64", "=?US-ASCII?b?SGVsbG8gV29ybGQ=?=", "Hello World"},
 		{"nested qp+b64", "=?utf-8?b?PT9VUy1BU0NJST9xP0hlbGxvPTIwV29ybGQ/PQ==?=", "Hello World"},
+		{"qp special header", "=?UTF-8?Q?s=3Dgreen;d=3Ddomain;i=3Dsender+=C3=A4@domain?=", "s=green;d=domain;i=sender+ä@domain"},
 	}
 
 	for _, tt := range ttable {

It outputs s="green;d=domain;i=sender+ä@domain" instead of the expected s=green;d=domain;i=sender+ä@domain.

In isolation, I expected the decoding function to return the exact original UTF-8 text but when I looked into it, I found that mediatype parsing depends on this function being aware it is being used to parse only a single parameter pair, hence the quote adding logic.
I tried moving the quote logic to mediatype parsing but got stuck on the "voice call" testcases.

When I run my string through mime.WordDecoder, I get the correct result.

Should this be fixed? i.e. move the quote logic to mediatype parsing and keep this function as a pure RFC2047 decoder or should its documentation mention that you can't just decode any random header value with it?

[jhillyerd]

I'd be OK with renaming the existing to RFC2047DecodePair or similar, and removing the single parameter pair specialization from the originally named func, if you think that would make more sense to folks using it.

[mgnsk]

I'd be OK with renaming the existing to RFC2047DecodePair or similar, and removing the single parameter pair specialization from the originally named func, if you think that would make more sense to folks using it.

That would make sense. I think just in case, it should be marked as a breaking change, should anyone depend on the pair specialization in the current func.