# ⎕FEATURE - Feature Control

 Z←⎕FEATURE returns the current value of the Feature Control vector
Z is an integer vector whose elements each control a separate experimental Feature of the system.

 ⎕FEATURE←R sets all of the values of the Feature Control Vector to R. ⎕FEAURE←⍬ sets all of the values of the Feature Control vector to their system default state. ⎕FEATURE[L]←R sets the value(s) of the Feature Control vector indexed by L to R — a 1 in R enables the feature, 0 disables it.

This system variable allows the user to enable or disable certain experimental features — the initial setting is for all such features to be disabled. The features controlled are as follows:

 [1] Allow negative indices [2] Allow ¯0 [3] User Hurwitz's Floor function instead of McDonnell's [4] Withdrawn (formerly used to allow NaNs) [5] Use APL2's definition of Inner Product

For example,

• With the first feature enabled, in origin-1, the index 1 refers to the first element in a vector and 0 refers to the last element:
```      ⎕FEATURE[1]←1
a←'abcdef'
a[0 1 2]
fab
a[¯2 ¯1 0 1 2]
defab
⍳¯3
¯2 ¯1 0
```
When enabled, the first feature affects all contexts sensitive to the Index Origin, such as the L in   R[L],   R[L]←A,   R[L]f←A,   L⌷R,   L⍉R,   and   L⊃R.

• With the second feature disabled and then enabled:
```      ⎕FEATURE[2]←0
÷∞ ¯∞
0 0
÷÷∞ ¯∞
∞ ∞
⎕FEATURE[2]←1
÷∞ ¯∞
0 ¯0
÷÷∞ ¯∞
∞ ¯∞
```

Because ¯0 can not be stored as a 64-bit Integer, enabling this feature has a drawback of occasionally forcing a type promotion from Integer to Floating Point. For example, if R is an Integer array with 0∊R, then an expression such as -R can no longer be stored as an Integer array, and the entire array must be promoted to Floating Point, possibly losing precision because 64-bit Integers have 63 bits of precision and 64-bit Floating Point numbers have only 53.

For example,
```      ⎕FEATURE[2]←0
0 ⎕DR ⎕←-¯1 0 1,⌊2*62
1 0 ¯1 ¯4611686018427387904
Integer (6412):  64 bits per element
⎕FEATURE[2]←1
0 ⎕DR ⎕←-¯1 0 1,⌊2*62
1 ¯0 ¯1 ¯4.611686018427388E18
Floating Point (6413):  64 bits per element```
• With the third feature disabled and then enabled:
```      a←0.5J¯0.5
⎕FEATURE[3]←0 ⋄ ⌊a
1J¯1
⎕FEATURE[3]←1 ⋄ ⌊a
0J0
```

The industry standard definition of the Floor function on Complex numbers is McDonnell's. However, McDonnell's Floor function isn't defined on Quaternions, but Hurwitz's is:

```      a←<1 3 5 7÷2
⎕FEATURE[3]←0 ⋄ ⌊a
DOMAIN ERROR
⎕FEATURE[3]←0 ⋄ ⌊a
∧
⎕FEATURE[3]←1 ⋄ ⌊a
0i1j2k3```

The two Floor functions return the same result for all Real numbers; they differ only on certain Complex numbers.

The setting of this feature affects not only the result of the Floor function but also all other primitive functions directly or indirectly sensitive to the Floor function including Ceiling (⌈R), Residue (L|R), Greatest Common Divisor (L∨R), and Least Common Multiple (L∧R).

For more details on this topic, see Hypercomplex GCD in APL.

• The fourth feature has been withdrawn.
• With the fifth feature disabled and then enabled:
```      a←3 6⍴'Queue EschewAchoo '
a
Queue
Eschew
Achoo
⎕FEATURE[5]←0
a+.∊'aeiou'
LENGTH ERROR
a+.∊'aeiou'
∧
⎕FEATURE[5]←1
a+.∊'aeiou'
4 1 2
```

System Variables (A value may be assigned to these except for ⎕DM)
ALX CT DM DT ELX FC FEATURE FPC IC IO
LR LX PP PR PW RL SA WSID
Niladic System Functions (a value cannot be assigned to these)
A AV EM ET LC NNAMES NNUMS SI SYSID SYSVER
T TC TCBEL TCBS TCESC TCFF TCHT TCLF TCNL TCNUL
TS WA
Monadic or dyadic system functions (a value cannot be assigned to these)
AT CR DC DFT DL DR EA EC ERROR ES
EX FMT FX MF NAPPEND NC NCREATE NERASE NINFO NL
NLOCK NREAD NRENAME NREPLACE NRESIZE NSIZE NTIE NUNTIE STOP TF
TRACE UCS VR
Note that quad functions and variables (except for the ⎕A family of functions) are case insensitive