Fixes to run the bootloader integration tests #37
Commits on Feb 22, 2024
-
Some issues occurred when compiling the project with the Madara prover API.
-
Fix: use correct type when setting
simple_bootloader_inputStore the `simple_bootloader_input` field of the bootloader input instead of storing the whole bootloader_input.
-
Fix: use correct hint code for %{ 0 %}
The hint is compiled to a different code.
-
Fix: use correct hint code for validate_hash
A backslash was encoded the wrong way in the hint code.
-
Inner select builtins hint: select builtin
Implement the hint that determines whether a specific builtin is selected for the given program.
-
Fix: consider output_ptr as a relocatable instead of a value
`pre_execution_builtin_ptrs` is an array of pointers, not of values as we previously understood.
-
Fix: do not force integer type when copying builtins
Problem: `write_return_builtins` copies data from the pre-execution builtins for each builtin not used by the program. We specify that the value of the pre-execution builtins is integer, which is not the case. Solution: just copy the memory without enforcing a specific type.
-
Fix: output_runner_data is an optional
Implemented `get_task_fact_topologies` for Cairo PIE tasks at the same time.
-
-
Fix: use correct hint code for nondet %{ ids.num // 2 %}
The hint is compiled to a different code.
-
Fix: use correct hint code for isinstance(packed_output, PlainPackedO…
…utput) The hint is compiled to a different code.
-
-
-
Fix: use program identifiers to determine ret_pc for Cairo PIEs
Remove the hardcoded value for `ret_pc` and compute it using the program identifiers instead. We now expect the bootloader program to be accessible as the `bootloader_program` variable in the root execution scope.
-
Fix: deserialize PIE additional data as struct instead of hashmap
Problem: Deserializing the PIE additional data as a hashmap of `BuiltinAdditionalData` enums because of an issue with deserializing untagged unions in `serde` (see serde-rs/json#1103). Solution: add a new `AdditionalData` struct with explicit fields for each builtin, circumventing the untagged union issue. This solution has the advantage of always associating the correct data type for each builtin (it's not possible anymore to associate a builtin with a different data type), but requires modifications if a new builtin is added.
-
-
Fix: deserialize ECDSA builtin data from Python VM format
Problem: the ECDSA/signature builtin additional data is stored internally as a hashmap, but the Python VM stores it as a vector of tuples. Solution: Add a `SignatureBuiltinAdditionalData` struct and implement a custom deserializer for it that can take either a hashmap or a vector.
-
Fix: support loading PIE memory files > 32KB
Fixed a bug in the memory file loader implementation.
-
Feature: support output pages in public input memory
Problem: output page IDs do not appear when exporting the builtin memory, but default to page 0 instead. Solution: add a `get_public_memory` to the output builtin to export page IDs correctly.
-
Feature: deserialize Cairo PIE from bytes
Added a `from_bytes` class method to build a `CairoPie` method in addition to the existing `from_file` method.
