feat(static): add static package as a DCE-friendly alternative

[vincentbernat]

The package is a drop-in replacement with two features removed:

• functions in environment
• Eval() method

The changes are quite minimal and I am using go generate to copy expr.go to v2/expr.go. The full user interface is therefore not present in expr.go anymore. We could keep Eval() and make it panic instead.

Fix #863.

[vincentbernat]

Dunno why the coverage test fails.

Some benchmarks:

goos: linux
goarch: amd64
pkg: github.com/expr-lang/expr
cpu: AMD Ryzen 5 5600X 6-Core Processor
                            │ before.txt  │              after.txt              │
                            │   sec/op    │    sec/op     vs base               │
_expr-12                      85.63n ± 1%   79.22n ±  4%   -7.49% (p=0.002 n=6)
_expr_eval-12                 5.781µ ± 1%   5.994µ ±  2%   +3.68% (p=0.002 n=6)
_expr_reuseVm-12              60.66n ± 3%   53.55n ±  2%  -11.72% (p=0.002 n=6)
_len-12                       48.46n ± 4%   47.82n ±  1%   -1.32% (p=0.004 n=6)
_filter-12                    42.43µ ± 2%   43.10µ ±  1%   +1.58% (p=0.041 n=6)
_filterLen-12                 37.82µ ± 2%   37.79µ ±  2%        ~ (p=0.699 n=6)
_filterFirst-12               380.0n ± 1%   383.9n ± 10%   +1.03% (p=0.039 n=6)
_filterLast-12                716.7n ± 2%   684.9n ±  2%   -4.44% (p=0.002 n=6)
_filterMap-12                 3.861µ ± 3%   3.829µ ±  1%        ~ (p=0.258 n=6)
_arrayIndex-12                84.54n ± 2%   86.55n ±  2%        ~ (p=0.065 n=6)
_envStruct-12                 53.07n ± 1%   55.24n ±  4%   +4.10% (p=0.002 n=6)
_envStruct_noEnv-12           140.5n ± 2%   141.3n ±  1%        ~ (p=0.394 n=6)
_envMap-12                    64.05n ± 3%   60.57n ±  1%   -5.43% (p=0.002 n=6)
_callFunc-12                  348.0n ± 6%   358.9n ±  1%        ~ (p=0.058 n=6)
_callMethod-12                365.4n ± 1%   382.8n ±  1%   +4.75% (p=0.002 n=6)
_callField-12                 75.81n ± 2%   74.95n ±  0%   -1.13% (p=0.004 n=6)
_callFast-12                  76.38n ± 3%   76.66n ±  2%        ~ (p=0.937 n=6)
_callConstExpr-12             35.02n ± 1%   35.37n ±  2%        ~ (p=0.240 n=6)
_largeStructAccess-12         138.4n ± 1%   144.2n ±  5%   +4.26% (p=0.002 n=6)
_largeNestedStructAccess-12   148.0n ± 1%   160.8n ±  1%   +8.65% (p=0.002 n=6)
_largeNestedArrayAccess-12    1.073m ± 8%   1.064m ±  9%        ~ (p=0.937 n=6)
_sort-12                      3.948µ ± 1%   4.493µ ±  1%  +13.80% (p=0.002 n=6)
_sortBy-12                    11.33µ ± 3%   11.24µ ±  1%        ~ (p=0.132 n=6)
_groupBy-12                   8.259µ ± 1%   7.964µ ±  2%   -3.57% (p=0.002 n=6)
_reduce-12                    34.64n ± 1%   34.59n ±  2%        ~ (p=0.818 n=6)
_min-12                       500.6n ± 1%   496.3n ±  1%   -0.86% (p=0.026 n=6)
_max-12                       506.7n ± 1%   504.8n ±  1%   -0.37% (p=0.013 n=6)
_mean-12                      218.1n ± 2%   220.3n ±  2%        ~ (p=0.221 n=6)
_median-12                    1.599µ ± 1%   1.631µ ±  1%   +2.03% (p=0.002 n=6)
geomean                       564.7n        566.7n         +0.36%

pkg: github.com/expr-lang/expr/checker
                                            │ before.txt  │             after.txt              │
                                            │   sec/op    │   sec/op     vs base               │
Checker/name=function_calls-12                6.248µ ± 5%   6.098µ ± 3%        ~ (p=0.093 n=6)
Checker/name=unary_and_binary_operations-12   6.167µ ± 3%   6.309µ ± 3%   +2.30% (p=0.026 n=6)
Checker/name=deep_struct_access-12            16.48µ ± 2%   21.50µ ± 1%  +30.48% (p=0.002 n=6)
geomean                                       8.594µ        9.387µ        +9.22%

pkg: github.com/expr-lang/expr/internal/difflib
                   │ before.txt  │             after.txt             │
                   │   sec/op    │   sec/op     vs base              │
SplitLines100-12     905.9n ± 1%   841.1n ± 1%  -7.15% (p=0.002 n=6)
SplitLines10000-12   81.46µ ± 1%   79.75µ ± 3%       ~ (p=0.065 n=6)
geomean              8.590µ        8.190µ       -4.66%

pkg: github.com/expr-lang/expr/internal/testify/assert
              │ before.txt  │             after.txt             │
              │   sec/op    │   sec/op     vs base              │
BytesEqual-12   144.9n ± 5%   143.6n ± 3%       ~ (p=0.506 n=6)
NotNil-12       4.416n ± 1%   4.379n ± 2%       ~ (p=0.485 n=6)
geomean         25.29n        25.08n       -0.85%

pkg: github.com/expr-lang/expr/optimizer
                                 │ before.txt  │             after.txt              │
                                 │   sec/op    │   sec/op     vs base               │
CountGtZero-12                     1.438µ ± 3%   1.343µ ± 1%   -6.61% (p=0.002 n=6)
CountGtZeroLargeEarlyMatch-12      9.107µ ± 7%   9.317µ ± 6%        ~ (p=0.310 n=6)
CountGtZeroNoMatch-12              29.32µ ± 4%   29.55µ ± 1%        ~ (p=0.485 n=6)
CountGteOneEarlyMatch-12           1.452µ ± 2%   1.339µ ± 5%   -7.82% (p=0.002 n=6)
CountGteOneNoMatch-12              29.22µ ± 5%   28.98µ ± 1%        ~ (p=0.065 n=6)
CountThresholdEarlyMatch-12        15.00µ ± 3%   13.19µ ± 1%  -12.06% (p=0.002 n=6)
CountThresholdGteEarlyMatch-12     12.20µ ± 2%   10.84µ ± 3%  -11.12% (p=0.002 n=6)
CountThresholdNoEarlyExit-12       342.4µ ± 2%   333.6µ ± 2%   -2.59% (p=0.009 n=6)
CountThresholdLargeEarlyMatch-12   89.66µ ± 2%   83.75µ ± 5%   -6.59% (p=0.004 n=6)
CountThresholdLtEarlyExit-12       15.56µ ± 9%   13.16µ ± 8%  -15.42% (p=0.002 n=6)
CountThresholdLteEarlyExit-12      12.45µ ± 3%   11.50µ ± 7%   -7.70% (p=0.002 n=6)
CountThresholdLtNoEarlyExit-12     340.8µ ± 3%   349.7µ ± 2%        ~ (p=0.093 n=6)
CountThresholdLteNoEarlyExit-12    353.2µ ± 3%   348.9µ ± 5%        ~ (p=0.394 n=6)
SumArray-12                        187.1n ± 4%   149.0n ± 5%  -20.39% (p=0.002 n=6)
SumRange_Optimized-12              35.64n ± 7%   36.22n ± 2%        ~ (p=0.701 n=6)
ReduceRangeSum_Optimized-12        35.45n ± 6%   35.03n ± 5%        ~ (p=0.485 n=6)
SumRange_Unoptimized-12            3.267µ ± 4%   3.369µ ± 2%        ~ (p=0.065 n=6)
geomean                            7.854µ        7.451µ        -5.13%

pkg: github.com/expr-lang/expr/parser
          │  before.txt  │             after.txt             │
          │    sec/op    │   sec/op     vs base              │
Parser-12   618.2n ± 13%   588.3n ± 2%  -4.84% (p=0.002 n=6)

          │ before.txt │           after.txt           │
          │    B/op    │    B/op     vs base           │
Parser-12   256.0 ± 0%   256.0 ± 0%  ~ (p=1.000 n=6) ¹
¹ all samples are equal

          │ before.txt │           after.txt           │
          │ allocs/op  │ allocs/op   vs base           │
Parser-12   8.000 ± 0%   8.000 ± 0%  ~ (p=1.000 n=6) ¹
¹ all samples are equal

pkg: github.com/expr-lang/expr/patcher/value
                    │ before.txt  │             after.txt             │
                    │   sec/op    │   sec/op     vs base              │
_valueAdd-12          86.08n ± 1%   90.62n ± 3%  +5.27% (p=0.002 n=6)
_valueUntypedAdd-12   102.7n ± 2%   110.8n ± 2%  +7.94% (p=0.002 n=6)
_valueTypedAdd-12     84.59n ± 1%   89.58n ± 3%  +5.90% (p=0.002 n=6)
geomean               90.75n        96.53n       +6.36%

pkg: github.com/expr-lang/expr/test/bench
                  │ before.txt  │             after.txt             │
                  │   sec/op    │   sec/op     vs base              │
Call_callTyped-12   8.686µ ± 1%   8.832µ ± 2%  +1.68% (p=0.002 n=6)
Call_eval-12        1.619µ ± 1%   1.684µ ± 2%  +4.01% (p=0.002 n=6)
geomean             3.750µ        3.856µ       +2.84%

pkg: github.com/expr-lang/expr/vm
                          │ before.txt  │          after.txt           │
                          │   sec/op    │   sec/op     vs base         │
VM/name=function_calls-12   1.670µ ± 1%   1.637µ ± 4%  ~ (p=0.058 n=6)

pkg: github.com/expr-lang/expr/vm/runtime
                                          │ before.txt  │              after.txt              │
                                          │   sec/op    │    sec/op     vs base               │
Equal/int_==_int-12                         2.163n ± 1%   2.200n ±  3%   +1.71% (p=0.009 n=6)
Equal/int_!=_int-12                         2.165n ± 0%   2.196n ±  1%   +1.45% (p=0.004 n=6)
Equal/int_==_int8-12                        2.379n ± 0%   2.406n ±  1%   +1.11% (p=0.002 n=6)
Equal/int_==_int16-12                       2.381n ± 0%   2.419n ±  4%   +1.60% (p=0.004 n=6)
Equal/int_==_int32-12                       2.188n ± 2%   2.240n ±  2%   +2.38% (p=0.013 n=6)
Equal/int_==_int64-12                       2.399n ± 1%   2.397n ±  1%        ~ (p=0.699 n=6)
Equal/float_==_float-12                     2.384n ± 1%   2.400n ±  3%   +0.65% (p=0.041 n=6)
Equal/float_!=_float-12                     2.384n ± 0%   2.401n ±  1%   +0.67% (p=0.004 n=6)
Equal/float_==_int-12                       2.606n ± 0%   2.621n ±  1%   +0.59% (p=0.030 n=6)
Equal/float_!=_int-12                       2.607n ± 2%   4.617n ± 43%  +77.10% (p=0.004 n=6)
Equal/string_==_string-12                   3.244n ± 2%   3.329n ± 85%   +2.64% (p=0.015 n=6)
Equal/string_!=_string-12                   4.547n ± 1%   4.547n ±  0%        ~ (p=0.853 n=6)
Equal/bool_==_bool-12                       2.165n ± 0%   2.174n ±  1%   +0.39% (p=0.024 n=6)
Equal/bool_!=_bool-12                       2.167n ± 0%   2.169n ±  0%        ~ (p=0.866 n=6)
Equal/[]any_==_[]int-12                     15.18n ± 6%   15.49n ±  5%        ~ (p=0.981 n=6)
Equal/[]any_!=_[]int-12                     15.50n ± 8%   14.80n ±  6%        ~ (p=0.457 n=6)
Equal/deep_[]any_==_[]any-12                84.08n ± 1%   84.36n ± 11%        ~ (p=0.310 n=6)
Equal/deep_[]any_!=_[]any-12                61.08n ± 2%   61.41n ±  7%        ~ (p=0.132 n=6)
Equal/map[string]any_==_map[string]any-12   253.5n ± 1%   254.4n ±  2%        ~ (p=0.240 n=6)
Equal/map[string]any_!=_map[string]any-12   59.46n ± 1%   59.56n ±  0%        ~ (p=0.848 n=6)
geomean                                     6.188n        6.405n         +3.51%

I didn't try to adapt the benchmark for the "static" version. Maybe we will get better results.

[antonmedv]

I was thinking about creating a build tag: expr_nomethod.

[vincentbernat]

I was thinking about creating a build tag: expr_nomethod.

Yes, it should have 0 performance impact as Go should be able to eliminate the additional tests. Do you want me to do a PR for that? Most of this PR can be reused for this.

Edit: I am working on it.

[antonmedv]

Nice!

Seems like still one test is failing in coverage.

[vincentbernat]

Seems like still one test is failing in coverage.

In this pull request, yes, but not in #957. Let me close this one.