[FVM] Handle cadence ParentErrors

[peterargue]

Cadence errors.ParentErrors contain a list of errors encountered during execution:
https://github.com/onflow/cadence/blob/5ce1f36f95bb458f05d03283c0af307ff7f6fe1e/runtime/errors/errors.go#L101-L104

These errors have an Unwrap() method that returns an []error instead of a plain error. This results in a tree of errors instead of a chain. While this is supported by the standard errors.Is and errors.As methods, the CodedError type is overloaded for use by both normal errors and failures. This means that if there is a failure mixed with other errors, the failure could be missed since searching would stop at the first instance of a CodedError.

This PR refactors the FVM CodedError to add a new type for failures: CodedFailure. This allows errors.As to find any instance of a failure within the error tree without custom logic.

[codecov-commenter]

Codecov Report

Attention: 14 lines in your changes are missing coverage. Please review.

Comparison is base (ed23d21) 55.59% compared to head (8a9af22) 55.62%.

Files	Patch %	Lines
fvm/errors/errors.go	92.50%	5 Missing and 1 partial ⚠️
fvm/errors/execution.go	33.33%	4 Missing ⚠️
fvm/errors/failures.go	69.23%	4 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #5272      +/-   ##
==========================================
+ Coverage   55.59%   55.62%   +0.03%     
==========================================
  Files        1002     1002              
  Lines       96658    96708      +50     
==========================================
+ Hits        53733    53795      +62     
+ Misses      38862    38845      -17     
- Partials     4063     4068       +5     

Flag	Coverage Δ
unittests	55.62% <86.66%> (+0.03%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[sideninja]

nice

[ramtinms]

I have to spend some time reviewing how the new error handling (coded) works on the FVM side before reviewing this, but before that out of curiosity why not fixing this on the cadence runtime?

[peterargue]

I have to spend some time reviewing how the new error handling (coded) works on the FVM side before reviewing this, but before that out of curiosity why not fixing this on the cadence runtime?

So there were 2 issues addressed here:

1. Some cadence errors contain a list of child errors and don't implement an Unwrap() method, so we need to handle them explicitly. This should be fixed in cadence. I tried to make the handler code in this PR lightweight and easy to remove later. (Update: this is fixed in cadence now, so I removed those bits)
2. Given some errors can have a list of child errors, there is an error tree instead of an error chain. The logic we have today expects a chain and would not find Failures if they were in another branch from a different CodedError.

for example, in the below case the existing code would only see CheckerError. After cadence is updated to unwrap these, the existing FVM code would only find someFVMError, but miss someFVMFailure since As and Is stop searching at the first instance of an error.

runtime.Error
└── sema.CheckerError
    ├── someCadenceError
    ├── someFVMError
    └── someFVMFailure

By separating CodedError and CodedFailure types, the stdlib's methods can now find any CodedFailure in the tree.

[turbolent]

I don't know enough about FVM, so can't really review the PR, but in general, thank you for cleaning up and improving the categorization of errors 🙏

As far as I understand, "errors" are implementation errors (e.g. file can't be opened) and "failures" are user errors (e.g. the user ran out of computation units)?

[peterargue]

As far as I understand, "errors" are implementation errors (e.g. file can't be opened) and "failures" are user errors (e.g. the user ran out of computation units)?

It's the opposite. "failures" are exceptions, and "errors" are recoverable errors.