Jonas Hahn

Ok so I think we've solved the IDL problem forever™ 1. Anchor 1.1 will add back legacy IDL support, so you can leave the legacy IDL up (and old clients functioning) until everyone has upgraded. 2. Anchor 1.1 client will check the major version of the IDL it receives from Program Metadata. If the major version is higher than what is supported by the client, it will search for the now legacy IDL at seed "ANCHOR_IDL_V1". Any future breaking changes to Anchor IDL will not break old clients because they will be coded to find their compatible version. Backwards compatibility ftw!! 3. A tool will be released to crawl IDL history through time, for anyone who needs to index historical changes. 4. Codama and Blueshift people are going to work together on a unified IDL that works for all of Anchor, Quasar, and Codama. No more IDL sprawl! And we'll get extra information like account types and PDAs that explorers can leverage. Worth noting that while this fixes a lot of backwards compatibility issues, we mostly agreed that best practice for modern Solana dev is likely to put all of you instruction building logic behind an API. While the IDL can get you flexibility to make program changes without breaking consuming clients, an API provides maximum flexibility. Your API to interact with your program should be open source, so anyone can trustlessly run that API themselves and not have to trust your centralized implementation to build instructions. And obviously people can still just used an IDL based client. But when you're thinking about building user-facing applications, an API is a very useful layer.

One of the most painful pieces of DevEx on Solana continues to be a complete disregard for client backwards compatibility in Anchor. You are infinitely better off building all transactions/instructions from a centralized server, even though it increases security risk. Otherwise everyone using your libraries will be forced to coordinate a big bang upgrade every time Anchor decides to change IDL format, where IDLs are stored, etc. This could have been solved once and for all with the metadata program, but for some reason it wasn't so we get to continue dealing with this issue forever.