errors

Review all changed and surrounding code in the current diff for error handling violations, then fix each one.

Rules

1. Never create dynamic errors

Do not use fmt.Errorf(...) or errors.New(...) to construct ad-hoc error values at the call site. Every error must be a package-level sentinel.

Wrong:

return fmt.Errorf("user %s not found", id)
return errors.New("connection failed")

2. Use sentinel errors

Each package declares its sentinels in a dedicated errors.go file (see internal/sensors/errors.go). Keep them unexported (errFoo) unless callers in other packages need errors.Is on them:

package sensors

import "errors"

var (
    errCPUInfo = errors.New("host_processor_info failed")
    errSMCRead = errors.New("SMC read failed")
)

Wrap sentinels to add context using fmt.Errorf with %w:

return fmt.Errorf("%w: kern_return_t %d", errCPUInfo, int(kr))

Wrapping an underlying error you received (syscall, stdlib) with %w is also fine: fmt.Errorf("sysctl hw.memsize: %w", err).

Do not create anonymous errors — always wrap a sentinel or a received error.

Check errors with errors.Is:

if errors.Is(err, errSMCRead) {
    // handle SMC read failure
}

3. Test files

Test files (*_test.go) are exempt from these rules. They may use fmt.Errorf or errors.New freely to simulate error conditions in mocks and assertions.

Checklist

For every piece of code you write or modify:

• No fmt.Errorf(...) or errors.New(...) producing anonymous errors in production code
• All packages declare their sentinels (var errFoo = errors.New(...)) in the package's errors.go
• Errors are wrapped with fmt.Errorf("...: %w", errSentinel) when adding context
• New error scenarios have a corresponding sentinel var