The idea is to look at the code as-is (that is, NOT suggesting new or alternate syntax and semantics) in the review / commentary and try to add (more) tests and start to (re)package / modular-ize code in "place holder" gems (waiting for adoption by the founders) such as 0xfacet and 0xfacet-typed and 0xfacet-rubidity
Yes, you can. You are welcome to join in and comments / discuss / debate.
The idea here is break the "majestic rails rubidity monolith" also known as "facet vm" (formerly "ethscriptions vm") up into easier to (re)use modules.
For example, why not bundle up a "core" language "rubidity" gem with no dependencies on any blockchain and break out "core / standard" contracts samples and database (SQL) and runtime modules or such.
The first published "place holder" modules / gems include:
0xfacet - facet vm base foundation
0xfacet-typed - solidity-like value and reference types for rubidity o.g.
0xfacet-rubidity - rubidity o.g. facet vm contract programming language interpreter
Why not bundle up contracts in gems and use gem packages for versioning? Let's try:
0xfacet-contracts - rubidity o.g. standard contracts (incl. ERC20, PublicMintERC20, ERC721, GenerativeERC721, etc.)
- ERC20
- ERC721
- EtherBridge
- EtherBridgeV2
- EthscriptionERC20Bridge
- GenerativeERC721
- OpenEditionERC721
- PublicMintERC20
- UnsafeNoApprovalERC20
- Upgradeable
0xfacet-uniswap - rubidity o.g. uniswap v2 contracts (incl. UniswapV2Factory, UniswapV2Pair, UniswapV2Router, etc.)
- UniswapSetupZapV2
- UniswapV2Callee
- UniswapV2ERC20
- UniswapV2Factory
- UniswapV2Pair
- UniswapV2Router
- UniswapV2RouterWithRewards
Let's say it out loud - Rails is great. It's magic. It's amazing.
However, to make the rubidity core execution engine (interpreter) more secure (and smaller and simpler) less dependencies is more!
Thus, allow me to preach again and again to break out ("isolate") the rubidity core code to make it run "off-rails", that only requires "plain ruby" with "zero-dependencies" on any activesupport (rails-y) "magic".
This makes testing easier because the "isolated" core can get tested with "zero-dependency" and as the saying goes a strong solid foundation will help you for sure in the future to build out big big things.
the (core typed) code is:
the great news - this code for the type machinery is basically "isolated" and i recommend to move it to its own module as the foundation types (library) incl. the rspec tests.
Show respc test results
rubidity.review\0xfacet-typed> rspec
..............
Finished in 0.14714 seconds (files took 13.25 seconds to load)
14 examples, 0 failures
or
rubidity.review\0xfacet-typed> rspec --format documentation
Type
#can_be_assigned_from?
returns true if types are the same
returns true if both types are integer types and the number of bits of the first type is greater than or equal to the number of bits of the second type
returns false otherwise
returns true if a literal can be assigned to the type
raises a VariableTypeError if a literal cannot be assigned to the type
#values_can_be_compared?
returns true if types are compatible
returns true if both types are integer types
returns false otherwise
returns true if a literal can be compared with the type
returns false if a literal cannot be compared with the type
TypedVariable
.create_or_validate
returns the same TypedVariable if the value is a TypedVariable and its type can be assigned from the specified type
is fine to go up in bits
creates a new TypedVariable if the value is not a TypedVariable
raises a VariableTypeError if the value is a TypedVariable and its type cannot be assigned from the specified type
Finished in 0.08964 seconds (files took 4.97 seconds to load)
14 examples, 0 failures
The current code uses a "typed var holder" class with "replace" semantics. "Replace" semantics is flawed because you CANNOT replace, for example, a bool true or false or (if added enums that are integer constants).
The better approach is to ALWAYS create a new typed var (and never "replace" inline). Once created you CANNOT change the value inplace. Of course, if the typed var is a reference type such as array or mapping you CAN change / replace the mapping or array items.
The deserialize SHOULD not be method of typed variable because it assumes "replace" semantics.
Aside: "replace" semantics?!
def value=(new_value)
# ...
end
def deserialize(serialized_value)
self.value = serialized_value
end
the value gets replaced inline using value=
. the value should only
get set only once on init! and than is unreplaceable (immutable). create a new typed variable instead of replace inline.
Method Missing Magic & "Explicit" One-to-One Type Mappings Instead Of "Quick & Dirty" Generic Type "Wrappers"
For now only mapping and array get an explicit typed class. All other "value" types are (re)using a kind of genric typed class that than delegates via method_missing to wherever. Yes, a security and performance nightmare. While this approach may look like ingenious for being "quick & dirty" it is a clasic anti-pattern / major design flaw, that is, do NOT use method_missing UNLESS it is the last option. And options are many with the preference of doing the hard work coding and writing out all the types one by one!
Aside - method_missing?!
def method_missing(name, *args, &block)
if value.respond_to?(name)
result = value.send(name, *args, &block)
if result.class == value.class
begin
result = type.check_and_normalize_literal(result)
rescue ContractErrors::VariableTypeError => e
if type.is_uint?
result = TypedVariable.create(:uint256, result)
else
raise
end
end
end
result
if name.to_s.end_with?("=") && !%w[>= <=].include?(name.to_s[-2..])
self.value = result if type.is_value_type?
self
else
result
end
else
super
end
end
The array and mapping typed variable use "proxy classes" inside the typed variable. That proxy / layer inside the typed classes is redundant. The typed variable / class is already the "proxy" or "front" to define the API for the type, thus, less is more. better security by keeping it simple.
Twin Typed Classes:
ArrayType
&ArrayType::Proxy
MappingType
&MappingType::Proxy
ContractType
&ContractType::Proxy
- ...
Only code what you use. Remove unnecessary fluff. All contracts in /app/models/contracts do NOT use any multiple-inheritance, thus, do NOT add the baggage that no one uses. Use a simple linear parents contract inhertiance chain! Makes you code more secure by simplifying the rule for what function (contract method) gets called. No "diamond-shaped" surprise possible.
Let's say it out lout - multiple-inheritance is a security nightmare and design flaw in solidity - possibly inspired by python? (origin roots) definitely NOT ruby ;-).
Triva: Vyper - a "pythonic" more secure and simpler ("modern") solidity alternative is the other extreme:
Q: What is not included in Vyper?
The following constructs are not included because their use can lead to misleading or difficult to understand code:
- Modifiers
- Class inheritance
- Inline assembly
- ...
To be continued ...