Skip to content

parser/derive: replaces usages of Expr/Node with Box<Expr>/Box<Node>#542

Merged
Kijewski merged 1 commit intoaskama-rs:masterfrom
Kijewski:pr-box
Jul 25, 2025
Merged

parser/derive: replaces usages of Expr/Node with Box<Expr>/Box<Node>#542
Kijewski merged 1 commit intoaskama-rs:masterfrom
Kijewski:pr-box

Conversation

@Kijewski
Copy link
Copy Markdown
Member

This means that there is much less copying of data done at runtime.

Two new unsafe transmutations had to be added, both in the same function, because even though &T and Box<T> have the same layout, rust currently rejects any Boxes in a const value. Miri is happy with askama_derive's tests.

askama_derive benchmark
synthetic/hello_world   time:   [35.108 µs 35.271 µs 35.412 µs]
                        thrpt:  [1.3735 MiB/s 1.3790 MiB/s 1.3854 MiB/s]
                 change:
                        time:   [−6.2826% −5.1769% −4.1151%] (p = 0.00 < 0.05)
                        thrpt:  [+4.2917% +5.4595% +6.7038%]
                        Performance has improved.

librustdoc/item_info.html
                        time:   [39.687 µs 39.957 µs 40.186 µs]
                        thrpt:  [3.9157 MiB/s 3.9381 MiB/s 3.9649 MiB/s]
                 change:
                        time:   [−3.5603% −2.5102% −1.3793%] (p = 0.00 < 0.05)
                        thrpt:  [+1.3986% +2.5749% +3.6917%]
                        Performance has improved.

librustdoc/item_union.html
                        time:   [101.64 µs 101.95 µs 102.27 µs]
                        thrpt:  [9.6516 MiB/s 9.6818 MiB/s 9.7111 MiB/s]
                 change:
                        time:   [−4.0638% −3.5966% −3.1652%] (p = 0.00 < 0.05)
                        thrpt:  [+3.2686% +3.7307% +4.2359%]
                        Performance has improved.

librustdoc/page.html    time:   [379.80 µs 381.74 µs 383.65 µs]
                        thrpt:  [16.140 MiB/s 16.221 MiB/s 16.304 MiB/s]
                 change:
                        time:   [−6.6429% −5.2611% −4.3384%] (p = 0.00 < 0.05)
                        thrpt:  [+4.5351% +5.5533% +7.1156%]
                        Performance has improved.

librustdoc/print_item.html
                        time:   [88.483 µs 88.849 µs 89.313 µs]
                        thrpt:  [10.571 MiB/s 10.626 MiB/s 10.670 MiB/s]
                 change:
                        time:   [−2.5658% −2.0639% −1.5517%] (p = 0.00 < 0.05)
                        thrpt:  [+1.5761% +2.1074% +2.6333%]
                        Performance has improved.

librustdoc/short_item_info.html
                        time:   [90.028 µs 90.157 µs 90.281 µs]
                        thrpt:  [10.035 MiB/s 10.049 MiB/s 10.063 MiB/s]
                 change:
                        time:   [−2.4814% −1.9435% −1.3640%] (p = 0.00 < 0.05)
                        thrpt:  [+1.3829% +1.9821% +2.5445%]
                        Performance has improved.

librustdoc/sidebar.html time:   [106.91 µs 108.17 µs 109.57 µs]
                        thrpt:  [11.263 MiB/s 11.408 MiB/s 11.543 MiB/s]
                 change:
                        time:   [−4.2433% −3.3985% −2.6066%] (p = 0.00 < 0.05)
                        thrpt:  [+2.6763% +3.5181% +4.4314%]
                        Performance has improved.

librustdoc/source.html  time:   [71.149 µs 71.416 µs 71.652 µs]
                        thrpt:  [10.289 MiB/s 10.322 MiB/s 10.361 MiB/s]
                 change:
                        time:   [−2.7381% −2.1144% −1.5588%] (p = 0.00 < 0.05)
                        thrpt:  [+1.5835% +2.1601% +2.8152%]
                        Performance has improved.

librustdoc/type_layout.html
                        time:   [88.405 µs 88.920 µs 89.595 µs]
                        thrpt:  [30.049 MiB/s 30.277 MiB/s 30.453 MiB/s]
                 change:
                        time:   [+0.6665% +1.8855% +3.0267%] (p = 0.00 < 0.05)
                        thrpt:  [−2.9378% −1.8506% −0.6621%]
                        Change within noise threshold.

librustdoc/type_layout_size.html
                        time:   [44.048 µs 44.246 µs 44.398 µs]
                        thrpt:  [6.1004 MiB/s 6.1213 MiB/s 6.1488 MiB/s]
                 change:
                        time:   [−5.8376% −4.9477% −4.0335%] (p = 0.00 < 0.05)
                        thrpt:  [+4.2030% +5.2052% +6.1996%]
                        Performance has improved.

…<Node>`

This means that there is much less copying of data done at runtime.

Two new `unsafe` transmutations had to be added, both in the same
function, because even though `&T` and `Box<T>` have the same layout,
rust currently rejects any `Box`es in a `const` value. Miri is happy
with `askama_derive`'s tests.

<details>
<summary>askama_derive benchmark</summary>

```text
synthetic/hello_world   time:   [35.108 µs 35.271 µs 35.412 µs]
                        thrpt:  [1.3735 MiB/s 1.3790 MiB/s 1.3854 MiB/s]
                 change:
                        time:   [−6.2826% −5.1769% −4.1151%] (p = 0.00 < 0.05)
                        thrpt:  [+4.2917% +5.4595% +6.7038%]
                        Performance has improved.

librustdoc/item_info.html
                        time:   [39.687 µs 39.957 µs 40.186 µs]
                        thrpt:  [3.9157 MiB/s 3.9381 MiB/s 3.9649 MiB/s]
                 change:
                        time:   [−3.5603% −2.5102% −1.3793%] (p = 0.00 < 0.05)
                        thrpt:  [+1.3986% +2.5749% +3.6917%]
                        Performance has improved.

librustdoc/item_union.html
                        time:   [101.64 µs 101.95 µs 102.27 µs]
                        thrpt:  [9.6516 MiB/s 9.6818 MiB/s 9.7111 MiB/s]
                 change:
                        time:   [−4.0638% −3.5966% −3.1652%] (p = 0.00 < 0.05)
                        thrpt:  [+3.2686% +3.7307% +4.2359%]
                        Performance has improved.

librustdoc/page.html    time:   [379.80 µs 381.74 µs 383.65 µs]
                        thrpt:  [16.140 MiB/s 16.221 MiB/s 16.304 MiB/s]
                 change:
                        time:   [−6.6429% −5.2611% −4.3384%] (p = 0.00 < 0.05)
                        thrpt:  [+4.5351% +5.5533% +7.1156%]
                        Performance has improved.

librustdoc/print_item.html
                        time:   [88.483 µs 88.849 µs 89.313 µs]
                        thrpt:  [10.571 MiB/s 10.626 MiB/s 10.670 MiB/s]
                 change:
                        time:   [−2.5658% −2.0639% −1.5517%] (p = 0.00 < 0.05)
                        thrpt:  [+1.5761% +2.1074% +2.6333%]
                        Performance has improved.

librustdoc/short_item_info.html
                        time:   [90.028 µs 90.157 µs 90.281 µs]
                        thrpt:  [10.035 MiB/s 10.049 MiB/s 10.063 MiB/s]
                 change:
                        time:   [−2.4814% −1.9435% −1.3640%] (p = 0.00 < 0.05)
                        thrpt:  [+1.3829% +1.9821% +2.5445%]
                        Performance has improved.

librustdoc/sidebar.html time:   [106.91 µs 108.17 µs 109.57 µs]
                        thrpt:  [11.263 MiB/s 11.408 MiB/s 11.543 MiB/s]
                 change:
                        time:   [−4.2433% −3.3985% −2.6066%] (p = 0.00 < 0.05)
                        thrpt:  [+2.6763% +3.5181% +4.4314%]
                        Performance has improved.

librustdoc/source.html  time:   [71.149 µs 71.416 µs 71.652 µs]
                        thrpt:  [10.289 MiB/s 10.322 MiB/s 10.361 MiB/s]
                 change:
                        time:   [−2.7381% −2.1144% −1.5588%] (p = 0.00 < 0.05)
                        thrpt:  [+1.5835% +2.1601% +2.8152%]
                        Performance has improved.

librustdoc/type_layout.html
                        time:   [88.405 µs 88.920 µs 89.595 µs]
                        thrpt:  [30.049 MiB/s 30.277 MiB/s 30.453 MiB/s]
                 change:
                        time:   [+0.6665% +1.8855% +3.0267%] (p = 0.00 < 0.05)
                        thrpt:  [−2.9378% −1.8506% −0.6621%]
                        Change within noise threshold.

librustdoc/type_layout_size.html
                        time:   [44.048 µs 44.246 µs 44.398 µs]
                        thrpt:  [6.1004 MiB/s 6.1213 MiB/s 6.1488 MiB/s]
                 change:
                        time:   [−5.8376% −4.9477% −4.0335%] (p = 0.00 < 0.05)
                        thrpt:  [+4.2030% +5.2052% +6.1996%]
                        Performance has improved.
```
</details>
@GuillaumeGomez
Copy link
Copy Markdown
Collaborator

Thanks!

When I'll need to update my WIP for token streams, it's gonna one hell of a ride. XD

@Kijewski
Copy link
Copy Markdown
Member Author

Oops, sorry. :D Most of my changes are simple replacements, though: %s/WithSpan<'_, Expr<'_>/WithSpan<'_, Box<Expr<'_>>/g, then afterwards add an asterisks or &* to all lines that contain errors. :) Most of the function bodies in the generated are mostly unchanged.

@GuillaumeGomez
Copy link
Copy Markdown
Collaborator

Nah don't worry. I don't expect this PR to take much to rebase.

@Kijewski Kijewski merged commit b8e11e8 into askama-rs:master Jul 25, 2025
42 checks passed
@Kijewski Kijewski deleted the pr-box branch July 25, 2025 21:44
@Kijewski Kijewski mentioned this pull request Jul 23, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants