feat: Add stacksafe feature

[osiewicz]

Objective

This PR adds stacksafe feature. We ran taffy super_deep benchmark with different display kinds and annotated all stack overflow-inducing call-sites with stacksafe::stacksafe.

Changes that will affect external library users must update RELEASES.md before they will be merged.

Context

There was a prior discussion about stacksafe use in #853.

[nicoburns]

This should probably also be added to https://github.com/DioxusLabs/taffy/blob/main/src/compute/grid/mod.rs#L43

[osiewicz]

Oddly enough I could not trigger an overflow with grid layout easily, but yeah, let's stay consistent.

[osiewicz]

I ran benchmarks against taffy without stacksafe enabled:
Grid:

Grid (tl;dr: 8-10% perf regression)

grid/wide/31x31/961 time: [656.85 µs 664.80 µs 668.10 µs] change: [−3.3794% −0.2172% +3.4173%] (p = 0.91 > 0.05) No change in performance detected. grid/wide/100x100/10000 time: [9.0808 ms 9.1444 ms 9.1716 ms] change: [−1.4588% +1.1055% +3.6078%] (p = 0.40 > 0.05) No change in performance detected. Benchmarking grid/wide/316x316/99856: Warming up for 3.0000 s Warning: Unable to complete 10 samples in 5.0s. You may wish to increase target time to 9.2s or enable flat sampling. grid/wide/316x316/99856 time: [154.95 ms 155.43 ms 156.40 ms] change: [+2.7604% +3.2565% +3.8323%] (p = 0.00 < 0.05) Performance has regressed. grid/deep/2x2/1024 time: [1.6690 ms 1.6836 ms 1.6905 ms] change: [+7.0525% +8.3548% +9.6254%] (p = 0.00 < 0.05) Performance has regressed. grid/deep/3x3/6561 time: [9.1765 ms 9.2361 ms 9.2734 ms] change: [+6.2477% +7.4270% +8.6247%] (p = 0.00 < 0.05) Performance has regressed. grid/deep/2x2/16384 time: [34.196 ms 34.301 ms 34.448 ms] change: [+6.3845% +7.2430% +8.0459%] (p = 0.00 < 0.05) Performance has regressed. Found 1 outliers among 10 measurements (10.00%) 1 (10.00%) high mild grid/superdeep/1x1/100 time: [175.93 µs 179.93 µs 182.26 µs] change: [+10.696% +12.494% +14.303%] (p = 0.00 < 0.05) Performance has regressed. grid/superdeep/1x1/1000 time: [1.8662 ms 1.8758 ms 1.8812 ms] change: [+9.9307% +10.608% +11.334%] (p = 0.00 < 0.05) Performance has regressed.

Flexbox (tl;dr: >10% perf regression)

yoga 'huge nested'/Taffy 0.7 /10000 time: [2.4040 ms 2.4058 ms 2.4077 ms] change: [+11.913% +12.043% +12.166%] (p = 0.00 < 0.05) Performance has regressed. Found 6 outliers among 100 measurements (6.00%) 3 (3.00%) low mild 1 (1.00%) high mild 2 (2.00%) high severe Wide tree/Taffy 0.7 (2-level hierarchy)/10000 time: [3.9141 ms 3.9206 ms 3.9266 ms] change: [+9.9228% +11.393% +12.883%] (p = 0.00 < 0.05) Performance has regressed. Found 2 outliers among 10 measurements (20.00%) 2 (20.00%) low severe Deep tree (auto size)/Taffy 0.7 (12-level hierarchy)/4000 time: [2.2935 ms 2.3227 ms 2.3361 ms] change: [+14.057% +16.311% +18.752%] (p = 0.00 < 0.05) Performance has regressed. Deep tree (auto size)/Taffy 0.7 (14-level hierarchy)/10000 time: [5.9560 ms 5.9848 ms 5.9957 ms] change: [+13.704% +15.171% +16.676%] (p = 0.00 < 0.05) Performance has regressed. Found 2 outliers among 10 measurements (20.00%) 2 (20.00%) low mild Benchmarking Deep tree (random size)/Taffy 0.7 (12-level hierarchy)/4000: Collecting 10 samples in estimated 5.0668 s (2530 iterationDeep tree (random size)/Taffy 0.7 (12-level hierarchy)/4000 time: [1.5994 ms 1.6166 ms 1.6241 ms] change: [+3.1367% +7.1762% +11.125%] (p = 0.00 < 0.05) Performance has regressed. Benchmarking Deep tree (random size)/Taffy 0.7 (14-level hierarchy)/10000: Collecting 10 samples in estimated 5.2958 s (935 iterationDeep tree (random size)/Taffy 0.7 (14-level hierarchy)/10000 time: [4.3597 ms 4.3754 ms 4.3847 ms] change: [+8.5868% +9.6264% +10.751%] (p = 0.00 < 0.05) Performance has regressed. Found 1 outliers among 10 measurements (10.00%) 1 (10.00%) low mild super deep/Taffy 0.7 /100 time: [196.86 µs 198.73 µs 199.47 µs] change: [+15.924% +18.632% +21.418%] (p = 0.00 < 0.05) Performance has regressed.

That's a bit yikes (even though we at Zed are likely to roll with it). I wonder if all of the callsites need the stacksafe annotation and if not, whether reducing them would reduce the perf impact of stacksafe on taffy's performance.

[nicoburns]

Some principled reasoning for where to include the annotations:

As a starting point, we want to consider the functions here: https://docs.rs/taffy/latest/taffy/compute/index.html. Looking at those:

• compute_flexbox_layout
• compute_grid_layout
• compute_block_layout
• compute_leaf_layout

Should definitely have the annotation. One of these functions gets called each time we recurse into a child node.

• compute_root_layout doesn't need it. That only gets called once for the very root node of the tree.
• compute_cached_layout doesn't need it because it is redundant with the layout-mode specific functions above. And we prefer to put it on the other methods to speed up cache access.
• compute_hidden_layout maybe needs it? This also recurses down the tree but with a very minimal function that doesn't use a lot of stack space: https://docs.rs/taffy/latest/src/taffy/compute/mod.rs.html#266-278
• round_layout - The inner round_layout_inner should probably have it (you already have this).
• print_tree - The inner print_node should probably have it

[nicoburns]

Removing from here might help perf, because this function gets called just prior to Taffy reading from it's cache. It's also probably not necessary here, because:

• If there's a cache hit then we will immediately return back up the stack
• If there's a cache miss then we will recurse into another function which will be annotated

[nicoburns]

I'd also remove here for same reason as above. This is just prior to hitting the cache, and we'd like to only run this check if we get a cache miss to keep cache access fast.

[nicoburns]

I expect this and perform_final_layout_on_in_flow_children aren't necessary, but are probably harmless.