In Mpp_ir_to_bir.wrap_m_code_call, some assignments of the Mir program are set as input definition, regardless of the M actual definition, before converting them as Bir instructions.
When we consider M call as rules this has no effect on the generated Bir code as body of rules are converted elsewhere. In case on rule inlining (such as in an optimization setting) however, this will lead to affectations disappearing from the code. To my knowledge, this is not what the current INTER files would do, as it will call rule functions with all their assignments in place regardless of context.
I fear the only reason we haven't caught on this semantic discrepancy is because M++ files only assign variables that were already considered input by the M code.
In
Mpp_ir_to_bir.wrap_m_code_call, some assignments of the Mir program are set as input definition, regardless of the M actual definition, before converting them as Bir instructions.When we consider M call as rules this has no effect on the generated Bir code as body of rules are converted elsewhere. In case on rule inlining (such as in an optimization setting) however, this will lead to affectations disappearing from the code. To my knowledge, this is not what the current INTER files would do, as it will call rule functions with all their assignments in place regardless of context.
I fear the only reason we haven't caught on this semantic discrepancy is because M++ files only assign variables that were already considered input by the M code.