Check for sideffects from invalid manipulations.
Interestingly, the codegen'd map ends up doing this in a different way than the generic map: since it *is* actually wrapping and delegating to another assembler to handle the child value, it has to make sure *that* is invalidated; otherwise the fact we hand calls to the child assembler right away before doing our 'flush' would be problematic. Splitting the 'flush' in half would also work, of course. At present it seems like six of one and half a dozen of the other. We can subject this to microbenchmarking later if it seems relevant.
Showing
Please register or sign in to comment