This could be really neat for optimized builds. If using a PHF produces more efficient code for Debug representations, then even at the expense of compile times (only in optimized builds) it might be a big win for libraries using huge error enums.
I wonder if this can be extrapolated even further for error enums that have nested errors in it (with no dynamic fields except the nested error enums), using a PHF to map all the potentially hundreds of variant/subvariant combinations to unique strings, of course assuming this results in a performance win. I'm sure there are even more places where it could be useful to use a PHF too.