`a_opt: a?;` is not equivalent to `a_opt: a | EMPTY;` and causes an error `First set empty for grammar symbol "a_opt". An infinite recursion on the grammar symbol.` when `a` is a sequence. about parglare

[igordejanovic]

? is syntax sugar which unrolls to rules with _opt suffix. Please see here.

Using rule gyro_specifier_opt: gyro_specifier?; you end-up with gyro_specifier_opt: gyro_specifier_opt | EMPTY;. The first alternative is invalid self-reference. A workaround is to use different name for gyro_specifier_opt, or gyro_specifier? directly in other rules RHSs (which is IMHO better for readability as it makes apparent that something is optional).

While this is not a bug, i.e. it works as designed, I think it would be nice to have better error report in this case where user rules clashes with rules produced by syntax sugar unroll.

[KOLANICH]

Thank you.

A workaround is to use different name for gyro_specifier_opt, or gyro_specifier? directly in other rules RHSs (which is IMHO better for readability as it makes apparent that something is optional).

I just resorted to doing desugaring in this case myself. It may be better for readability to have a? than a_opt: a | EMPTY; ... a_opt, but for me it is not an option (for now) because for some parser generators one has to have each thing being captured in a separate rule and having the same tree structure for all the grammars for all the backends allows me use the same code for postprocessing.

While this is not a bug, i.e. it works as designed, I think it would be nice to have better error report in this case where user rules clashes with rules produced by syntax sugar unroll.

Could we detect the cases when a: b; and optimize them out by renaming b everywhere into a, when b is a generated name?

[igordejanovic]

Manual desugaring is perfectly fine as the desugaring process will introduce new rules only if they don't already exist.

Could we detect the cases when a: b; and optimize them out by renaming b everywhere into a, when b is a generated name?

I don't think that such transformation should be implemented in parglare.

[KOLANICH]

Thanks for making it clear.