…s for string bitwise ops
… improvements; don't need the nqp::want rigmarole any more.
…ng stuff the inliner can do well with.
Does not catch unused non-$ variables in double quoted strings. This also moves $Inf and $NaN to an earlier setting file, because they were used before the declaration. It also contains a very ugly workaound around a problem with the nqp %markhash being global, and reused for multiple evals. This bug was exposed by other parts of the patch.
…_REDUCE; it's a binary op already. Just use it.
…gs now runs in under half the time, and we shave a little bit off the spectest runtime too.
…d to NaN. However, Num.new() seems to be expected to give back zero, so add an explicit new that makes this happen. Means we regress no spectests now after reprapi2.
…e types properly. This gets us a bit further into the build.
…We now get to trying to build the setting...BOOM SEGFAULT!
[Coke]++ for complaining; I too found the old behavior inconvenient, but never enough to take action.