Refactor reactivity system with doubly-linked list and push-pull model

[wangshunnn]

Refactor

• 响应式系统重构：
  • Deps 和 Subs 由之前的 Array 和 Set 数据结构改为双向链表结构（Doubly-linked list），受 Preact Signals 启发。
  • 基于类似图着色传播算法实现 Push-pull hybrid model，受 @reactively、alien-signals (Vue3.6) 启发。之前是类似 Vue2 的 Push-based Model。提升性能，解决“相等性问题”。
• 抽离响应式核心部分到新的子包 —— @mpxjs/reactivity，使用 TS 重写：
  • 抽离新包是考虑架构解耦，健壮 Mpx 生态基建。
  • TS 重写是考虑到后续维护友好、类型友好、开发友好。
• 新增响应式单测、benchmark。
• @mpxjs/utils 新增类型声明文件（packages/utils/@types），包含之前 @mpxjs/webpack-plugin 内部单独声明的 declare module '@mpxjs/utils' 部分，方便 @mpxjs/reactivity 等 TS 包引入使用。
• 非侵入式/破坏性重构，向后兼容，不影响上层 API 设计以及上层库 @mpxjs/core 对外调用，用户层使用无感知。
• CI：根目录采用 lerna run tsc:build 统一管理需要 TS 构建的子包，方便后续扩展。

Performance

和 v2.10.6 版本的性能对比 benchmark 测试如下图所示，性能普遍全面提升，平均为之前的 2x~4x，最为明显的是 create computed（~9x），和 computed/effect 依赖大量 ref 的场景（~4x)，以及大量 computed/effect 场景（~3x）。

前者明显提升的另一个原因是之前 computed 创建需要同时实例化一个 RefImpl 和一个 ReactiveEffect，现在仅需实例化一个 ComputedRefImpl。

More benchmarks for this PR vs. v2.10.6

 ✓ __benchmarks__/reactivity.bench.ts > reactivity benchmark 13607ms
     name                                                                     hz     min     max    mean     p75     p99    p995    p999     rme   samples
   · create computed                                               21,770,565.22  0.0000  0.1245  0.0000  0.0000  0.0001  0.0001  0.0002  ±0.09%  10885283  [9.94x] ⇑
     create computed                                                2,189,229.86  0.0002  8.2277  0.0005  0.0003  0.0006  0.0007  0.0023  ±6.21%   1094615  (baseline)
   · write ref, don't read computed (without effect)               10,807,893.91  0.0000  0.1243  0.0001  0.0001  0.0002  0.0003  0.0003  ±0.15%   5403947  [1.17x] ⇑
     write ref, don't read computed (without effect)                9,264,878.89  0.0000  0.1885  0.0001  0.0001  0.0002  0.0002  0.0003  ±0.14%   4632440  (baseline)
   · write ref, don't read computed (with effect)                   4,342,400.99  0.0001  0.7008  0.0002  0.0003  0.0005  0.0005  0.0006  ±0.51%   2171201  [1.83x] ⇑
     write ref, don't read computed (with effect)                   2,369,266.04  0.0003  1.3953  0.0004  0.0004  0.0011  0.0013  0.0039  ±1.01%   1184634  (baseline)
   · write ref, read computed (without effect)                      5,759,068.69  0.0001  0.0712  0.0002  0.0002  0.0003  0.0003  0.0003  ±0.06%   2879535  [1.82x] ⇑
     write ref, read computed (without effect)                      3,165,287.80  0.0002  5.7341  0.0003  0.0003  0.0004  0.0010  0.0015  ±2.54%   1582644  (baseline)
   · write ref, read computed (with effect)                         4,477,675.74  0.0001  0.0592  0.0002  0.0003  0.0003  0.0003  0.0004  ±0.04%   2238838  [1.88x] ⇑
     write ref, read computed (with effect)                         2,375,472.60  0.0003  0.1391  0.0004  0.0004  0.0006  0.0008  0.0015  ±0.22%   1187737  (baseline)
   · write ref, don't read 1000 computeds (without effect)          7,685,740.88  0.0000  0.0546  0.0001  0.0001  0.0002  0.0002  0.0003  ±0.04%   3842871  [1.17x] ⇑
     write ref, don't read 1000 computeds (without effect)          6,585,490.41  0.0001  0.1553  0.0002  0.0002  0.0002  0.0003  0.0003  ±0.07%   3292746  (baseline)
   · write ref, don't read 1000 computeds (with multiple effects)      11,931.50  0.0817  0.1946  0.0838  0.0837  0.0894  0.0905  0.0976  ±0.06%      5966  [3.24x] ⇑
     write ref, don't read 1000 computeds (with multiple effects)       3,687.57  0.2326  2.0430  0.2712  0.2623  0.4875  0.6559  1.3192  ±1.36%      1847  (baseline)
   · write ref, don't read 1000 computeds (with single effect)         12,382.85  0.0797  0.3807  0.0808  0.0803  0.0856  0.0880  0.1892  ±0.22%      6192  [2.15x] ⇑
     write ref, don't read 1000 computeds (with single effect)          5,756.84  0.1455  3.2653  0.1737  0.1658  0.3537  0.8557  1.1373  ±1.85%      2881  (baseline)
   · write ref, read 1000 computeds (no effect)                        18,489.98  0.0510  0.0721  0.0541  0.0557  0.0591  0.0605  0.0645  ±0.08%      9245  [2.79x] ⇑
     write ref, read 1000 computeds (no effect)                         6,635.03  0.1307  1.1408  0.1507  0.1478  0.2449  0.2706  1.0137  ±0.90%      3318  (baseline)
   · write ref, read 1000 computeds (with multiple effects)            11,142.21  0.0788  0.4256  0.0897  0.0884  0.1257  0.1479  0.2161  ±0.30%      5572  [3.26x] ⇑
     write ref, read 1000 computeds (with multiple effects)             3,421.93  0.2545  2.6755  0.2922  0.2862  0.4819  0.5655  1.3722  ±1.39%      1712  (baseline)
   · write ref, read 1000 computeds (with single effect)                8,354.15  0.1035  0.7691  0.1197  0.1191  0.1523  0.1632  0.1925  ±0.33%      4178  [2.53x] ⇑
     write ref, read 1000 computeds (with single effect)                3,301.05  0.2784  1.4085  0.3029  0.2983  0.4586  0.5016  1.3667  ±0.87%      1651  (baseline)
   · 1000 refs, read 1 computed (without effect)                       22,408.97  0.0335  1.2976  0.0446  0.0426  0.2199  0.2906  0.5021  ±1.61%     11205  [2.09x] ⇑
     1000 refs, read 1 computed (without effect)                       10,720.22  0.0798  1.0688  0.0933  0.0933  0.1207  0.1503  0.3137  ±0.63%      5361  (baseline)
   · 1000 refs, read 1 computed (with effect)                          34,174.15  0.0275  0.2401  0.0293  0.0292  0.0338  0.0352  0.0814  ±0.18%     17088  [4.31x] ⇑
     1000 refs, read 1 computed (with effect)                           7,919.91  0.1067  3.2648  0.1263  0.1201  0.3023  0.3950  1.3248  ±2.13%      3960  (baseline)

 ✓ __benchmarks__/reactivity.bench.ts > reactivity benchmark 13589ms
     name                                                                     hz     min     max    mean     p75     p99    p995    p999     rme   samples
   · create computed                                               21,709,952.96  0.0000  0.0865  0.0000  0.0000  0.0001  0.0001  0.0002  ±0.08%  10854978  [9.92x] ⇑
     create computed                                                2,189,229.86  0.0002  8.2277  0.0005  0.0003  0.0006  0.0007  0.0023  ±6.21%   1094615  (baseline)
   · write ref, don't read computed (without effect)               11,075,474.56  0.0000  0.0525  0.0001  0.0001  0.0002  0.0002  0.0002  ±0.03%   5537738  [1.20x] ⇑
     write ref, don't read computed (without effect)                9,264,878.89  0.0000  0.1885  0.0001  0.0001  0.0002  0.0002  0.0003  ±0.14%   4632440  (baseline)
   · write ref, don't read computed (with effect)                   4,493,788.91  0.0001  0.0553  0.0002  0.0003  0.0003  0.0003  0.0004  ±0.03%   2246895  [1.90x] ⇑
     write ref, don't read computed (with effect)                   2,369,266.04  0.0003  1.3953  0.0004  0.0004  0.0011  0.0013  0.0039  ±1.01%   1184634  (baseline)
   · write ref, read computed (without effect)                      5,401,377.64  0.0001  0.1857  0.0002  0.0002  0.0003  0.0003  0.0003  ±0.11%   2700689  [1.71x] ⇑
     write ref, read computed (without effect)                      3,165,287.80  0.0002  5.7341  0.0003  0.0003  0.0004  0.0010  0.0015  ±2.54%   1582644  (baseline)
   · write ref, read computed (with effect)                         4,520,859.38  0.0001  0.0934  0.0002  0.0002  0.0003  0.0003  0.0004  ±0.06%   2260430  [1.90x] ⇑
     write ref, read computed (with effect)                         2,375,472.60  0.0003  0.1391  0.0004  0.0004  0.0006  0.0008  0.0015  ±0.22%   1187737  (baseline)
   · write ref, don't read 1000 computeds (without effect)          6,844,584.48  0.0001  0.1929  0.0001  0.0002  0.0002  0.0003  0.0003  ±0.10%   3422293  [1.04x] ⇑
     write ref, don't read 1000 computeds (without effect)          6,585,490.41  0.0001  0.1553  0.0002  0.0002  0.0002  0.0003  0.0003  ±0.07%   3292746  (baseline)
   · write ref, don't read 1000 computeds (with multiple effects)      12,135.56  0.0793  0.1039  0.0824  0.0829  0.0876  0.0885  0.0925  ±0.04%      6068  [3.29x] ⇑
     write ref, don't read 1000 computeds (with multiple effects)       3,687.57  0.2326  2.0430  0.2712  0.2623  0.4875  0.6559  1.3192  ±1.36%      1847  (baseline)
   · write ref, don't read 1000 computeds (with single effect)         12,897.96  0.0763  0.1465  0.0775  0.0775  0.0821  0.0830  0.0897  ±0.04%      6449  [2.24x] ⇑
     write ref, don't read 1000 computeds (with single effect)          5,756.84  0.1455  3.2653  0.1737  0.1658  0.3537  0.8557  1.1373  ±1.85%      2881  (baseline)
   · write ref, read 1000 computeds (no effect)                        19,956.12  0.0478  0.0733  0.0501  0.0513  0.0547  0.0564  0.0591  ±0.07%      9979  [3.01x] ⇑
     write ref, read 1000 computeds (no effect)                         6,635.03  0.1307  1.1408  0.1507  0.1478  0.2449  0.2706  1.0137  ±0.90%      3318  (baseline)
   · write ref, read 1000 computeds (with multiple effects)            11,155.07  0.0879  0.1085  0.0896  0.0897  0.0945  0.0952  0.0972  ±0.02%      5578  [3.26x] ⇑
     write ref, read 1000 computeds (with multiple effects)             3,421.93  0.2545  2.6755  0.2922  0.2862  0.4819  0.5655  1.3722  ±1.39%      1712  (baseline)
   · write ref, read 1000 computeds (with single effect)                8,587.46  0.1148  0.1260  0.1164  0.1167  0.1209  0.1219  0.1239  ±0.02%      4294  [2.60x] ⇑
     write ref, read 1000 computeds (with single effect)                3,301.05  0.2784  1.4085  0.3029  0.2983  0.4586  0.5016  1.3667  ±0.87%      1651  (baseline)
   · 1000 refs, read 1 computed (without effect)                       25,091.32  0.0316  0.2643  0.0399  0.0411  0.0636  0.0745  0.0977  ±0.29%     12546  [2.34x] ⇑
     1000 refs, read 1 computed (without effect)                       10,720.22  0.0798  1.0688  0.0933  0.0933  0.1207  0.1503  0.3137  ±0.63%      5361  (baseline)
   · 1000 refs, read 1 computed (with effect)                          34,442.11  0.0252  0.2235  0.0290  0.0291  0.0376  0.0430  0.0560  ±0.15%     17222  [4.35x] ⇑
     1000 refs, read 1 computed (with effect)                           7,919.91  0.1067  3.2648  0.1263  0.1201  0.3023  0.3950  1.3248  ±2.13%      3960  (baseline)

Memory Usage

内存测试 benchmark 中创建了 1 ref + 10000 computeds (100 chained pairs) + 10000 effects (1 computed with an effect)，按树形拓扑结构连接（详见代码或下面折叠内容）。

• Before (v2.10.6): 19125.24 KB
• Now (this PR): 5443.71 KB (-72%)

Memory benchmark 拓扑结构

  /**
   * 测试用例中的响应式树形拓扑结构如下:
   *
   *            src(ref)
   *          /    |     \
   * computed1  computed1  computed1 ...  (w columns)
   *    |         |         |
   * computed2  computed2  computed2 ...
   *    |         |         |
   *    .         .         .
   *    .         .         . (h rows)
   *    |         |         |
   * computedH  computedH  computedH ...
   *
   * w = h = 100
   * 每个 computed 还对应一个 effect
   * 所有节点包括: 1 ref + 100 * 100 computed + 100 * 100 effects
   */

Analyse

空间复杂度方面，一个明显优化在依赖收集阶段：

• Before: 一个 Dep 需要一个 Sub[] 数组，而一个 Sub 则需要记录新旧依赖的两个 Dep[] 数组，以及两个记录新旧依赖 id 编号的 Set()，用于去重和后续查询清理。
• Now: 一个 Dep 和 一个 Sub 仅靠一个中间节点 link 通过双向链表关联，也无需额外 id 编号，各自仅需头尾节点即可实现链式查询和遍历。

时间复杂度方面，一个明显的优化在于 computed/effect 每次运行后的依赖清理阶段：

• Before: 需要遍历 deps[] 数组然后对比旧数组来筛选移除，并且 remove 涉及数组 index.of 和 splice 操作，所以当不需要移除任何依赖时时间复杂度最优为 $O(N)$，需要移除所有依赖时时间复杂度最坏为 $O(N*M)$，其中 $N$ 是 Sub 的 deps[] 数组长度， $M$ 是 Dep 的 subs[] 数组长度。
• Now: 结合双向链表，依赖收集结束后 Sub 尾节点 depsTail 之后的节点便是需要移除的，而且双向链表 remove 单个节点仅需 $O(1)$，所以时间复杂度最优仅为 $O(1)$，最坏为 $O(N)$，其中 $N$ 是 Sub 的 deps 数量。相比之前平均降低了一个量级。

Fix and others

修复一些小程序和 Web (Vue2) 表现不一致等历史遗留问题：

• diff（triggerRef）：triggerRef 之前仅对 shallowRef 生效，现在对普通 Ref 也生效，与 Web(Vue) 端对齐。
• diff（toRef）：修复 toRef 缺少的第三个可选默认值参数，与当前类型声明以及 Web 端对齐。
• diff（toRefs）：增加对数组类型的支持，与 Web(Vue) 端对齐。
• diff（effectScope）： 修复嵌套的 scope.off() 后 getCurrentScope 失效问题 - refer Vue fix。

[wangshunnn]

内部讨论后续 Mpx3 倾向于接入 @vue/reactivity 核心包，减少未来维护成本，也能提高跨端输出一致性。
本 PR 作为前期实践暂时转为 Draft 存档。