The definition of memarg in the text format has regressed in the Wasm 3.0 spec, specifically with regard to alignment. The rule definition uses the variable m but claims to produce an alignment of n (out of thin air), and there is nothing to translate from the raw alignment number (e.g. "4" or "8") to a power of two (e.g. 2 or 3). Both of these were correct in the Wasm 2.0 spec.
Additionally, the order of align= and offset= have been reversed. A literal reading of the spec would imply that in 2.0, offset must come before alignment, while in 3.0, alignment must come before offset. It is not clear to me if this strictness is intended, but it is a likely unintended difference between the spec versions.
3.0 (current):
2.0:
The definition of memarg in the text format has regressed in the Wasm 3.0 spec, specifically with regard to alignment. The rule definition uses the variable
mbut claims to produce an alignment ofn(out of thin air), and there is nothing to translate from the raw alignment number (e.g. "4" or "8") to a power of two (e.g.2or3). Both of these were correct in the Wasm 2.0 spec.Additionally, the order of
align=andoffset=have been reversed. A literal reading of the spec would imply that in 2.0, offset must come before alignment, while in 3.0, alignment must come before offset. It is not clear to me if this strictness is intended, but it is a likely unintended difference between the spec versions.3.0 (current):
2.0: