Improve performance and cleanup code

[psadac]

• remove leading and trailing identical runes before comparing
• use builtin min() function
• use supported Go versions, update actions to latest version in CI

Long strings with differences at the beginning (long_lead), in the middle (long_middle) or at the end (long_trail) show significant improvements in processing time and memory allocations. When the optimization is ineffective due to different leading and trailing characters (long_diff) there is no change in processing time or memory allocation.

goos: linux
goarch: amd64
pkg: github.com/agnivade/levenshtein
cpu: AMD Ryzen 7 7840U w/ Radeon  780M Graphics
                      │  before.txt  │              after.txt              │
                      │    sec/op    │   sec/op     vs base                │
Simple/ASCII-16         134.20n ± 0%   79.03n ± 0%  -41.11% (p=0.000 n=20)
Simple/French-16         254.8n ± 0%   129.7n ± 0%  -49.09% (p=0.000 n=20)
Simple/Nordic-16         500.6n ± 1%   208.0n ± 0%  -58.45% (p=0.000 n=20)
Simple/Long_lead-16     1862.0n ± 0%   209.6n ± 1%  -88.75% (p=0.000 n=20)
Simple/Long_middle-16   3613.0n ± 0%   325.0n ± 0%  -91.00% (p=0.000 n=20)
Simple/Long_trail-16    3911.0n ± 0%   399.0n ± 1%  -89.80% (p=0.000 n=20)
Simple/Long_diff-16      4.030µ ± 0%   4.029µ ± 1%        ~ (p=0.899 n=20)
Simple/Tibetan-16        413.0n ± 0%   277.3n ± 0%  -32.86% (p=0.000 n=20)
geomean                  964.6n        299.5n       -68.95%

                      │  before.txt  │              after.txt               │
                      │     B/op     │    B/op     vs base                  │
Simple/ASCII-16         0.000 ± 0%     0.000 ± 0%        ~ (p=1.000 n=20) ¹
Simple/French-16        0.000 ± 0%     0.000 ± 0%        ~ (p=1.000 n=20) ¹
Simple/Nordic-16        0.000 ± 0%     0.000 ± 0%        ~ (p=1.000 n=20) ¹
Simple/Long_lead-16     464.0 ± 0%     368.0 ± 0%  -20.69% (p=0.000 n=20)
Simple/Long_middle-16   672.0 ± 0%     544.0 ± 0%  -19.05% (p=0.000 n=20)
Simple/Long_trail-16    720.0 ± 0%     576.0 ± 0%  -20.00% (p=0.000 n=20)
Simple/Long_diff-16     720.0 ± 0%     720.0 ± 0%        ~ (p=1.000 n=20) ¹
Simple/Tibetan-16       0.000 ± 0%     0.000 ± 0%        ~ (p=1.000 n=20) ¹
geomean                            ²                -7.99%                ²
¹ all samples are equal
² summaries must be >0 to compute geomean

                      │  before.txt  │              after.txt               │
                      │  allocs/op   │ allocs/op   vs base                  │
Simple/ASCII-16         0.000 ± 0%     0.000 ± 0%        ~ (p=1.000 n=20) ¹
Simple/French-16        0.000 ± 0%     0.000 ± 0%        ~ (p=1.000 n=20) ¹
Simple/Nordic-16        0.000 ± 0%     0.000 ± 0%        ~ (p=1.000 n=20) ¹
Simple/Long_lead-16     3.000 ± 0%     2.000 ± 0%  -33.33% (p=0.000 n=20)
Simple/Long_middle-16   3.000 ± 0%     2.000 ± 0%  -33.33% (p=0.000 n=20)
Simple/Long_trail-16    3.000 ± 0%     2.000 ± 0%  -33.33% (p=0.000 n=20)
Simple/Long_diff-16     3.000 ± 0%     3.000 ± 0%        ~ (p=1.000 n=20) ¹
Simple/Tibetan-16       0.000 ± 0%     0.000 ± 0%        ~ (p=1.000 n=20) ¹
geomean                            ²               -14.11%                ²
¹ all samples are equal
² summaries must be >0 to compute geomean

[agnivade]

Thanks for this. Appreciate the care!

[agnivade]

Sorry, I had removed Travis, but somehow it still seems to linger.

Anyways, just one little thing: Could you please update the PR message with the full benchmark results from all the tests? Just for posterity. Thank you.

[psadac]

I updated the message for "posterity". For your Travis problem it may be in the webhooks.

[agnivade]

There are no webhooks anymore. I deleted everything. Weird. Anyways, thanks for the PR!