This is certainly pragmatic, although it breaks the math
q type size q literal forms underlying integer value (encoding)
----------------------------------------------------------------------------------------------------------
short (h) 16-bit 0Nh / -0Wh / 0Wh null = -32768; -inf = -32767; +inf = 32767
int (i) 32-bit 0Ni / -0Wi / 0Wi null = -2147483648; -inf = -2147483647; +inf = 2147483647
long (j) 64-bit 0N (or 0Nj) / -0W (or -0Wj) / 0W (or 0Wj) null = -9223372036854775808; -inf = -9223372036854775807; +inf = 9223372036854775807