Design Consideration.

[jaredponn]

Sometimes scripts are written to include useless computation to "salt" the script. For example, the following script includes the integer salt in the body s.t. different salts will induce the script to have different script hashes, but the script has the same functionality.

myScript :: Int -> ScriptContext -> Redeemer -> ()
myScript salt scriptContext redeemer = (\_ -> ... ) salt

Clearly, this poses a problem for certain optimizations -- namely, if we optimize after the salt is applied, then evaluating the expression (\_ -> ...) salt leads to just the body ..., and we've clearly removed the salt from the script. Thus, scripts with different salts now all have the same script hash.

This can have disastrous consequences if other parts of the protocol crucially rely on these scripts having distinct hashes.

So, the design considerations for covenant are as follows:

• we can ignore this in optimization passes and say "don't run covenant AFTER the script has been parameterized"; or
• we can introduce "volatile" constants which aren't optimized, and still let covenant run AFTER the script has been parameterized

See this slack thread for more details https://mlabs-corp.slack.com/archives/C02QQHLFB6Y/p1738182211804239.

[kozross]

Covenant itself performs no optimizations other than hash consing. In this case, since salt is an argument, myScript would be a top-level application node, rather than just the lambda, so it shouldn't be in danger of optimizing anything out.

The issue stems primarily in codegen, specifically from CTE. In general, CTE is something we very much want to do where possible, as it reduces script size and costs. However, in this exact case, CTE could be disastrous, as myScript would go from an application to a lambda, thus causing exactly the problem described. We kind of get this problem to some degree already (for example, if salt had effects, we can't CTE it away), but there's also a confounding factor in CBPV at play here.

Depending on how CBPV works out here, we may need a 'volatile' designator to specifically inhibit CTE. I believe this question can be kicked down the road a little, at least until we've figured out #8 and #9 .