pgx/pgtype
Evan Jones 07670dddca do not share the original input string
This allows the original input string to be garbage collected, so it
should not change the memory footprint. This is a slower than the
version that shares a string, but only a small amount. It is still
faster than binary parsing (until that is optimized).

benchstat difference of original versus this version:

                               │  orig.txt   │     new-do-not-share-string.txt     │
                               │   sec/op    │   sec/op     vs base                │
HstoreScan/databasesql.Scan-10   82.11µ ± 1%   14.24µ ± 2%  -82.66% (p=0.000 n=10)
HstoreScan/text-10               83.30µ ± 1%   14.97µ ± 1%  -82.03% (p=0.000 n=10)
HstoreScan/binary-10             15.99µ ± 2%   15.80µ ± 0%   -1.16% (p=0.024 n=10)
geomean                          47.82µ        14.99µ       -68.66%

                               │   orig.txt   │     new-do-not-share-string.txt      │
                               │     B/op     │     B/op      vs base                │
HstoreScan/databasesql.Scan-10   56.23Ki ± 0%   20.11Ki ± 0%  -64.24% (p=0.000 n=10)
HstoreScan/text-10               65.12Ki ± 0%   29.00Ki ± 0%  -55.47% (p=0.000 n=10)
HstoreScan/binary-10             21.09Ki ± 0%   21.09Ki ± 0%        ~ (p=0.722 n=10)
geomean                          42.58Ki        23.08Ki       -45.80%

                               │  orig.txt  │     new-do-not-share-string.txt      │
                               │ allocs/op  │ allocs/op   vs base                  │
HstoreScan/databasesql.Scan-10   744.0 ± 0%   340.0 ± 0%  -54.30% (p=0.000 n=10)
HstoreScan/text-10               743.0 ± 0%   339.0 ± 0%  -54.37% (p=0.000 n=10)
HstoreScan/binary-10             464.0 ± 0%   464.0 ± 0%        ~ (p=1.000 n=10) ¹
geomean                          635.4        376.8       -40.70%
¹ all samples are equal

benchstat difference of the shared string versus not:

                               │ new-share-string.txt │     new-do-not-share-string.txt     │
                               │        sec/op        │   sec/op     vs base                │
HstoreScan/databasesql.Scan-10            10.57µ ± 2%   14.24µ ± 2%  +34.69% (p=0.000 n=10)
HstoreScan/text-10                        11.60µ ± 2%   14.97µ ± 1%  +29.03% (p=0.000 n=10)
HstoreScan/binary-10                      15.87µ ± 2%   15.80µ ± 0%        ~ (p=0.280 n=10)
geomean                                   12.48µ        14.99µ       +20.07%

                               │ new-share-string.txt │     new-do-not-share-string.txt      │
                               │         B/op         │     B/op      vs base                │
HstoreScan/databasesql.Scan-10           11.68Ki ± 0%   20.11Ki ± 0%  +72.17% (p=0.000 n=10)
HstoreScan/text-10                       20.58Ki ± 0%   29.00Ki ± 0%  +40.93% (p=0.000 n=10)
HstoreScan/binary-10                     21.08Ki ± 0%   21.09Ki ± 0%        ~ (p=0.427 n=10)
geomean                                  17.17Ki        23.08Ki       +34.39%

                               │ new-share-string.txt │      new-do-not-share-string.txt       │
                               │      allocs/op       │  allocs/op   vs base                   │
HstoreScan/databasesql.Scan-10             44.00 ± 0%   340.00 ± 0%  +672.73% (p=0.000 n=10)
HstoreScan/text-10                         44.00 ± 0%   339.00 ± 0%  +670.45% (p=0.000 n=10)
HstoreScan/binary-10                       464.0 ± 0%    464.0 ± 0%         ~ (p=1.000 n=10) ¹
geomean                                    96.49         376.8       +290.47%
2023-06-16 15:30:54 -05:00
..
2023-02-10 20:21:25 -06:00
2023-05-20 08:53:23 -05:00
2022-11-12 07:06:54 -06:00
2022-07-11 08:07:23 -05:00
2022-12-17 09:10:02 -06:00
2023-03-25 10:22:11 -05:00
2023-03-25 10:22:11 -05:00
2023-03-25 10:22:11 -05:00
2022-09-24 10:30:12 -05:00
2023-01-24 07:55:00 -06:00