perf: change underlying value from string to interface{}

[muktihari]

• Change underlying value from string to interface value, reducing unnecessary string conversion overhead.
• Use sync.Pool to handle Visitor creation to reduce allocs, instead of creating new Visitor each Visit, we can use existing Visitors. Look at this code example, after Visitor X finish visiting, Visitor Y might reuse Visitors created during previous Visitor X's Visit. This will happen recursively. And also, next time user trigger any expr function, chances are they can ruse existing Visitor objects from previous invocation, only if it is still being kept alive by sync.Pool.

func (v *Visitor) visitBinary(binaryExpr *ast.BinaryExpr) ast.Visitor {
    vx := visitorPool.Get().(*Visitor)
    defer visitorPool.Put(vx)
    vx.reset()
    vx.options = v.options

    ast.Walk(vx, binaryExpr.X)
    if vx.err != nil {
         v.err = vx.err
        return nil
    }

    /* Visitor Y might reuse existing Visitor objects created during Visitor X's Visit 
       (or from previous invocation) */
    vy := visitorPool.Get().(*Visitor)
    defer visitorPool.Put(vy)
    vy.reset()
    vy.options = v.options

    ast.Walk(vy, binaryExpr.Y)
    if vy.err != nil {
        v.err = vy.err
        return nil
    }
...

Here is the benchstat from old Version that use string with this new Version using interface plus sync.Pool.

goos: darwin
goarch: amd64
pkg: github.com/muktihari/expr
cpu: Intel(R) Core(TM) i5-5257U CPU @ 2.70GHz
                  │    old.txt    │               xxx.txt                │
                  │    sec/op     │    sec/op     vs base                │
Any/KindBoolean-4    13.51µ ±  3%   13.28µ ± 31%        ~ (p=0.631 n=10)
Any/KindInt-4        7.053µ ± 60%   5.364µ ± 14%  -23.94% (p=0.000 n=10)
Any/KindFloat-4     11.209µ ±  4%   9.288µ ± 31%  -17.14% (p=0.029 n=10)
Any/KindImag-4       9.600µ ± 72%   5.651µ ± 17%  -41.14% (p=0.000 n=10)
geomean              10.06µ         7.819µ        -22.29%

                  │   old.txt    │               xxx.txt                │
                  │     B/op     │     B/op      vs base                │
Any/KindBoolean-4   5.297Ki ± 0%   2.734Ki ± 0%  -48.38% (p=0.000 n=10)
Any/KindInt-4       2.664Ki ± 0%   1.562Ki ± 0%  -41.35% (p=0.000 n=10)
Any/KindFloat-4     3.844Ki ± 0%   2.047Ki ± 0%  -46.75% (p=0.000 n=10)
Any/KindImag-4      2.719Ki ± 0%   1.641Ki ± 0%  -39.66% (p=0.000 n=10)
geomean             3.485Ki        1.946Ki       -44.15%

                  │   old.txt   │              xxx.txt               │
                  │  allocs/op  │ allocs/op   vs base                │
Any/KindBoolean-4   111.00 ± 0%   72.00 ± 0%  -35.14% (p=0.000 n=10)
Any/KindInt-4        62.00 ± 0%   44.00 ± 0%  -29.03% (p=0.000 n=10)
Any/KindFloat-4      91.00 ± 0%   61.00 ± 0%  -32.97% (p=0.000 n=10)
Any/KindImag-4       77.00 ± 0%   49.00 ± 0%  -36.36% (p=0.000 n=10)
geomean              83.33        55.47       -33.43%

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (287578d) 100.00% compared to head (9e16f36) 100.00%.

Additional details and impacted files

@@            Coverage Diff            @@
##            master       #28   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files           12        12           
  Lines          837       901   +64     
=========================================
+ Hits           837       901   +64     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.