feat: descriptive error messages via IDL, not binary

[L0STE]

Problem

#[error_code] enums currently have no way to attach human-readable messages to error variants. Anchor solves this with #[msg("...")] attributes, but that approach embeds string literals into the on-chain binary — bloating program size for zero runtime benefit. Error messages are never read on-chain; they only matter for debugging in explorers and test output.

Proposal

Support an optional #[msg("...")] attribute on #[error_code] variants, but strip the strings at compile time so they never appear in the binary. Instead, emit them exclusively into the IDL:

#[error_code]
pub enum VaultError {
    #[msg("signer does not match vault authority")]
    Unauthorized,
    #[msg("insufficient lamports for withdrawal")]
    InsufficientFunds,
}

On-chain, this compiles to exactly what we have today — ProgramError::Custom(N) with no string data. The IDL gains an errors section that maps discriminants to names and messages:

{
  "errors": [
    { "code": 6000, "name": "Unauthorized", "msg": "signer does not match vault authority" },
    { "code": 6001, "name": "InsufficientFunds", "msg": "insufficient lamports for withdrawal" }
  ]
}

Integration

• Explorers: Read the IDL errors section to display descriptive messages for failed transactions instead of raw Custom(6000)
• QuasarSVM / test frameworks: Resolve ProgramError::Custom codes against the IDL to show the message in test failure output
• CLI (quasar idl): Already generates the IDL — extend it to include the errors section

Why this is better than Anchor's approach

Anchor bakes #[msg] strings into the compiled binary via std::fmt::Display. This costs binary size on every program deploy for strings that are never used on-chain. By keeping messages IDL-only, we get the same developer experience with zero binary overhead — programs stay as small as possible.

Scope

1. quasar-derive: Parse #[msg("...")] on #[error_code] variants, discard the strings during macro expansion (no codegen change)
2. quasar idl: Extract #[msg] strings and emit them in the IDL errors array
3. QuasarSVM test runner: Resolve error codes against the IDL for readable test output
4. CLI templates: Update template add-error to show #[msg] in the generated scaffold

[subhdotsol]

can i have a try on this @L0STE ?

[L0STE]

Hey @subhdotsol thanks for expressing interest but this is probably one of the most critical and "hard" thing that are currently on-going here.

We need to make sure that this is extremely based because it will be a change from the status quo that needs a lot of upgrades from explorer, sdk's for CPIs, ... so we're taking it slow and internal.