WIP on #798: nicer error logs #830
Conversation
| } | ||
| } | ||
|
|
||
| pub type DynAnchorResult = std::result::Result<(), Box<dyn AnchorError>>; |
There was a problem hiding this comment.
Not super into these names, happy to change them
| @@ -16,7 +16,7 @@ pub fn generate(program: &Program) -> proc_macro2::TokenStream { | |||
| // on chain. | |||
| #[inline(never)] | |||
| #[cfg(not(feature = "no-idl"))] | |||
| pub fn __idl_dispatch(program_id: &Pubkey, accounts: &[AccountInfo], idl_ix_data: &[u8]) -> ProgramResult { | |||
| pub fn __idl_dispatch(program_id: &Pubkey, accounts: &[AccountInfo], idl_ix_data: &[u8]) -> ::anchor_lang::DynAnchorResult { | |||
There was a problem hiding this comment.
These return type changes are where I'm not sure how to do the enum/no-dynamic dispatch approach—what should these return types be? It would depend on whether or not you've generated your own #[error] enum, right? I definitely agree that it's possible to have a nice static return type here (only a static number of error enums in your program), but I wasn't sure how to actually do it.
There was a problem hiding this comment.
This is really nice to use to get rid of dynamic dispatch
https://docs.rs/enum_dispatch/0.3.7/enum_dispatch/
| e | ||
| let program_error = e.to_program_error(); | ||
| anchor_lang::solana_program::msg!(&program_error.to_string()); | ||
| if let ::anchor_lang::solana_program::program_error::ProgramError::Custom(code) = program_error { |
There was a problem hiding this comment.
This printing is late-night hackyness, it seems that the anchor ts code isn't super lenient with error parsing, so I'm just printing both the old Custom error code: 0x thing as well as the new one.
|
One problem with this approach is that it doesn't allow us to add a filename + line number to the error message. So we'll probably need to change this when we want to add that feature. Tag @paul-schaaf. |
|
Closing in favor of #1462. |
#798
This is pretty rough but I think I need to hurry up and start a PR, oof. I think just about all tests are passing except for
lockup, which doesn't even build, but I can't seem to find why its registry#[program]is unhappy.There are two ideas:
First idea: Box up behind a dyn trait the various different error enums that can show up in an anchor program (low-level Solana ProgramErrors, Anchor's framework ErrorCode enum, Anchor's wrapper Error enum, your own custom errors, etc.). In principle it ought to be possible to avoid this dynamic dispatch approach but I'm not sure how to actually do it :/
Second idea/huge hack: Anchor's built-in traits (
Accounts,AccountsExit, etc.) all returnProgramResults. This means that internal Anchor ErrorCodes get smooshed intoProgramError::Custom(u32)s, throwing away their messages. I think probably the right fix for this is to change these return types, either toResult<(), crate::error::Error>, or I don't know, maybe by adding an associatedtype Error: Into<ProgramError>. I tried the first option. It ends up being a bit of a slog that I never quite managed to finish, and it's a breaking change that requires changing a lot of tests (there are some test ix handlers that rely on auto-into-ingProgramErrors in functions returningResult<()>, but that breaks if Anchor traits start usingcrate::error::Error).So, I thought it was worth starting a PR with something way hackier, based on the original idea I tried: macro generate a mapping from Custom(u32) --> Anchor's ErrorCode enum (we don't need to do this for your own custom
#[error]enums because you can just use a good return type for your ix handlers, and then the dyn trait stuff handles the rest). This feels so gross but it does seem to work?