derive: escape strings at compile-time when possible

[Kijewski]

Todo:

• look if there is a proper string parsing function
• benchmark derive

[Kijewski]

derive benchmark

hello_world             time:   [50.551 µs 50.673 µs 50.795 µs]
                        change: [+0.2145% +0.5681% +0.9316%] (p = 0.00 < 0.05)
                        Change within noise threshold.

item_info.html          time:   [60.100 µs 60.221 µs 60.354 µs]
                        change: [+0.7116% +0.9772% +1.2335%] (p = 0.00 < 0.05)
                        Change within noise threshold.

item_union.html         time:   [187.10 µs 187.49 µs 187.92 µs]
                        change: [+3.3728% +3.8026% +4.2713%] (p = 0.00 < 0.05)
                        Performance has regressed.

page.html               time:   [777.26 µs 777.97 µs 778.76 µs]
                        change: [-0.5346% -0.0855% +0.5532%] (p = 0.83 > 0.05)
                        No change in performance detected.

print_item.html         time:   [154.32 µs 154.46 µs 154.62 µs]
                        change: [-0.6921% -0.4250% -0.1376%] (p = 0.00 < 0.05)
                        Change within noise threshold.

short_item_info.html    time:   [129.97 µs 130.19 µs 130.43 µs]
                        change: [+1.5539% +1.9741% +2.4175%] (p = 0.00 < 0.05)
                        Performance has regressed.

sidebar.html            time:   [200.02 µs 200.28 µs 200.56 µs]
                        change: [+2.1864% +2.4044% +2.6125%] (p = 0.00 < 0.05)
                        Performance has regressed.

source.html             time:   [116.60 µs 116.76 µs 116.92 µs]
                        change: [+3.0496% +3.3321% +3.6158%] (p = 0.00 < 0.05)
                        Performance has regressed.

type_layout.html        time:   [167.39 µs 167.57 µs 167.78 µs]
                        change: [+3.4081% +4.8191% +5.7045%] (p = 0.00 < 0.05)
                        Performance has regressed.

type_layout_size.html   time:   [67.252 µs 67.361 µs 67.470 µs]
                        change: [+1.2518% +1.6146% +1.9632%] (p = 0.00 < 0.05)
                        Performance has regressed.

Acceptable.

[GuillaumeGomez]

It's a great idea! I'm surprised the impact on compile-time is this little, but it makes it even better.

Thanks!

[Kijewski]

I'm surprised the impact on compile-time is this little, but it makes it even better.

The case is not actually used in our tests. In rustdoc's code which we use in out benchmarks, there is no {{ "…" }}. So we just messure the overhead of inspecting the Exprs before pushing them into the write queue, and fast dropping a Cow::Borrowed(…) is.

The "red and black suits" example comes from out of our tests, though.