Replace uniseg with uax29

[clipperhouse]

Hi! I experimented with my graphemes package with your StringWidth and Truncate methods, and got some good results — considerably faster perf and lower allocs.

goos: darwin
goarch: arm64
pkg: github.com/mattn/go-runewidth
cpu: Apple M2
                                   │   old.txt    │              new.txt               │
                                   │    sec/op    │   sec/op     vs base               │
String1WidthAll/regular-8            108.63m ± 1%   41.51m ± 1%  -61.78% (p=0.002 n=6)
String1WidthAll/lut-8                 93.61m ± 1%   26.00m ± 0%  -72.23% (p=0.002 n=6)
String1Width768/regular-8             60.46µ ± 1%   15.80µ ± 0%  -73.87% (p=0.002 n=6)
String1Width768/lut-8                 60.69µ ± 0%   15.55µ ± 0%  -74.38% (p=0.002 n=6)
String1WidthAllEastAsian/regular-8   115.24m ± 1%   47.06m ± 0%  -59.16% (p=0.002 n=6)
String1WidthAllEastAsian/lut-8        93.62m ± 0%   26.02m ± 1%  -72.21% (p=0.002 n=6)
String1Width768EastAsian/regular-8    75.25µ ± 0%   27.99µ ± 1%  -62.81% (p=0.002 n=6)
String1Width768EastAsian/lut-8        60.79µ ± 0%   15.58µ ± 0%  -74.37% (p=0.002 n=6)
geomean                               2.559m        783.2µ       -69.40%

                                   │    old.txt     │               new.txt               │
                                   │      B/op      │     B/op      vs base               │
String1WidthAll/regular-8            106.251Mi ± 0%   4.250Mi ± 0%  -96.00% (p=0.002 n=6)
String1WidthAll/lut-8                106.251Mi ± 0%   4.250Mi ± 0%  -96.00% (p=0.002 n=6)
String1Width768/regular-8             75.000Ki ± 0%   3.000Ki ± 0%  -96.00% (p=0.002 n=6)
String1Width768/lut-8                 75.000Ki ± 0%   3.000Ki ± 0%  -96.00% (p=0.002 n=6)
String1WidthAllEastAsian/regular-8   106.251Mi ± 0%   4.250Mi ± 0%  -96.00% (p=0.002 n=6)
String1WidthAllEastAsian/lut-8       106.251Mi ± 0%   4.250Mi ± 0%  -96.00% (p=0.002 n=6)
String1Width768EastAsian/regular-8    75.000Ki ± 0%   3.000Ki ± 0%  -96.00% (p=0.002 n=6)
String1Width768EastAsian/lut-8        75.000Ki ± 0%   3.000Ki ± 0%  -96.00% (p=0.002 n=6)
geomean                                2.790Mi        114.3Ki       -96.00%

                                   │   old.txt   │              new.txt               │
                                   │  allocs/op  │  allocs/op   vs base               │
String1WidthAll/regular-8            3.342M ± 0%   1.114M ± 0%  -66.67% (p=0.002 n=6)
String1WidthAll/lut-8                3.342M ± 0%   1.114M ± 0%  -66.67% (p=0.002 n=6)
String1Width768/regular-8            2304.0 ± 0%    768.0 ± 0%  -66.67% (p=0.002 n=6)
String1Width768/lut-8                2304.0 ± 0%    768.0 ± 0%  -66.67% (p=0.002 n=6)
String1WidthAllEastAsian/regular-8   3.342M ± 0%   1.114M ± 0%  -66.67% (p=0.002 n=6)
String1WidthAllEastAsian/lut-8       3.342M ± 0%   1.114M ± 0%  -66.67% (p=0.002 n=6)
String1Width768EastAsian/regular-8   2304.0 ± 0%    768.0 ± 0%  -66.67% (p=0.002 n=6)
String1Width768EastAsian/lut-8       2304.0 ± 0%    768.0 ± 0%  -66.67% (p=0.002 n=6)
geomean                              87.75k        29.25k       -66.67%

Hopefully interesting!

[mattn]

Thank you. I'll look into it in later.

[clipperhouse]

Cheers. I just noticed this earlier conversation, with concern for compatibility with older Go versions.

This PR requires Go 1.20. I have another version in the works which can support Go 1.18, if it’s important.

[clipperhouse]

Updated benchmarks, see benchstat.txt:

goos: darwin
goarch: arm64
pkg: github.com/mattn/go-runewidth
cpu: Apple M2
                                   │   old.txt    │              new.txt               │
                                   │    sec/op    │   sec/op     vs base               │
String1WidthAll/regular-8            108.92m ± 0%   35.09m ± 3%  -67.78% (p=0.002 n=6)
String1WidthAll/lut-8                 93.97m ± 0%   18.70m ± 0%  -80.10% (p=0.002 n=6)
String1Width768/regular-8             60.62µ ± 1%   11.54µ ± 0%  -80.97% (p=0.002 n=6)
String1Width768/lut-8                 60.66µ ± 1%   11.43µ ± 0%  -81.16% (p=0.002 n=6)
String1WidthAllEastAsian/regular-8   115.13m ± 1%   40.79m ± 8%  -64.57% (p=0.002 n=6)
String1WidthAllEastAsian/lut-8        93.65m ± 0%   18.70m ± 2%  -80.03% (p=0.002 n=6)
String1Width768EastAsian/regular-8    75.32µ ± 0%   23.49µ ± 0%  -68.82% (p=0.002 n=6)
String1Width768EastAsian/lut-8        60.76µ ± 0%   11.50µ ± 0%  -81.07% (p=0.002 n=6)
geomean                               2.562m        604.5µ       -76.41%

                                   │   old.txt    │                 new.txt                 │
                                   │     B/op     │    B/op      vs base                    │
String1WidthAll/regular-8            106.3Mi ± 0%    0.0Mi ± 0%  -100.00% (p=0.002 n=6)
String1WidthAll/lut-8                106.3Mi ± 0%    0.0Mi ± 0%  -100.00% (p=0.002 n=6)
String1Width768/regular-8            75.00Ki ± 0%   0.00Ki ± 0%  -100.00% (p=0.002 n=6)
String1Width768/lut-8                75.00Ki ± 0%   0.00Ki ± 0%  -100.00% (p=0.002 n=6)
String1WidthAllEastAsian/regular-8   106.3Mi ± 0%    0.0Mi ± 0%  -100.00% (p=0.002 n=6)
String1WidthAllEastAsian/lut-8       106.3Mi ± 0%    0.0Mi ± 0%  -100.00% (p=0.002 n=6)
String1Width768EastAsian/regular-8   75.00Ki ± 0%   0.00Ki ± 0%  -100.00% (p=0.002 n=6)
String1Width768EastAsian/lut-8       75.00Ki ± 0%   0.00Ki ± 0%  -100.00% (p=0.002 n=6)
geomean                              2.790Mi                     ?                      ¹ ²
¹ summaries must be >0 to compute geomean
² ratios must be >0 to compute geomean

                                   │   old.txt   │                 new.txt                 │
                                   │  allocs/op  │  allocs/op   vs base                    │
String1WidthAll/regular-8            3.342M ± 0%   0.000M ± 0%  -100.00% (p=0.002 n=6)
String1WidthAll/lut-8                3.342M ± 0%   0.000M ± 0%  -100.00% (p=0.002 n=6)
String1Width768/regular-8            2.304k ± 0%   0.000k ± 0%  -100.00% (p=0.002 n=6)
String1Width768/lut-8                2.304k ± 0%   0.000k ± 0%  -100.00% (p=0.002 n=6)
String1WidthAllEastAsian/regular-8   3.342M ± 0%   0.000M ± 0%  -100.00% (p=0.002 n=6)
String1WidthAllEastAsian/lut-8       3.342M ± 0%   0.000M ± 0%  -100.00% (p=0.002 n=6)
String1Width768EastAsian/regular-8   2.304k ± 0%   0.000k ± 0%  -100.00% (p=0.002 n=6)
String1Width768EastAsian/lut-8       2.304k ± 0%   0.000k ± 0%  -100.00% (p=0.002 n=6)
geomean                              87.75k                     ?                      ¹ ²
¹ summaries must be >0 to compute geomean
² ratios must be >0 to compute geomean

[mattn]

I decided to merge this in later.

[mattn]

Thank you!