From what I understand, in HTTP/1.1, headers and always sent in plaintext, and message bodies can optionally be compressed using an off-the-shelf algorithm like gzip.

In HTTP/2.0, a dedicated header compression algorithm was introduced to compress headers (HPACK).

This is a naïve question, but I’m wondering why an HTTP-specific was necessary. Wouldn’t it make more sense to simply apply gzip over everything sent over TCP? (i.e. compress together all headers and bodies of all messages sharing a connection, using a compression algorithm that can encode/decode in a streaming fashion)

Are the compression gains of a custom-built solution really that significant, or am I not seeing some additional constraint making a naïve solution infeasible?