There are three "in-place" operations for the "direct" datatypes:
xyz=: aa i} xyz xyz=: aa (i}) xyz xyz=: xyz i}~ aa NB. J6.03 or later xyz=: xyz,aa xyz=: b}xyz,:y xyz=: b}y,:xyz
=. instead of =: also works; aa or i may be parenthesized. Boxed, extended integer, and rationals are indirect datatypes; other datatypes are direct.
In-place means that an existing array is modified as opposed to creating a new array, resulting in large savings in time and space. The following examples illustrate the point:
xyz=: 1e6 ?@$ 2e9 i=: 100 ?@$ # xyz aa=: ($i) ?@$ 1e3 ts=: 6!:2 , 7!:2@] NB. time and space ts 'xyz=: aa i} xyz' NB. in-place 2.12317e_5 1600 ts 'xyz=: aa i} (xyz)' NB. not in-place 0.0117526 4.19597e6 ts 'aa i}xyz' NB. not in-place 0.0112489 4.19571e6 pqr=: 1e6 ?@$ 2e9 b=: ($xyz) ?@$ 2 (b}xyz,:pqr) -: (xyz*-.b)+pqr*b 1 ts 'xyz=: b}xyz,:pqr' NB. in-place 0.0170625 1280 ts 'b}xyz,:pqr' NB. not in-place 0.0693116 2.09725e7 ts 'xyz=: (xyz*-.b)+pqr*b' NB. not in-place 0.0245808 1.36326e7
The patterns are very specific and are recognized at word formation time. Therefore, the patterns accommodate variations immaterial with respect to word formation (such as non-essential spaces), but are thwarted by syntactic variations (such as in the second benchmark above, with the extra parens).
In-place operations impose a small cost on all sentences, even those that have nothing to do with in-place operations, because it takes time to detect (or reject) the patterns. For explicit definitions, that cost is paid just once at the time of definition.
Contributed by RogerHui.