Split Node type into more appropriate types

[Swatinem]

I recently did some heap profiling and noticed that v8 makes all the Node objects the same size.
Even though TS internally differentiates between Node / Identifier / Token, v8 still treats them the same because they come from the same constructor function.

Splitting those up can lead give up to ~8% reduction in memory usage, as shown in my PR here: #33431

I also noticed that the Token type is not really a good fit, since also literals like numbers, strings, boolean, etc are of type Token, but they do participate in type-checking and thus get additional properties later on, which is not ideal for performance (although my PR did not show any significant change).

So my questions here are:

• Should we micro-optimize this? I would argue yes, since my very simple change already shows great memory usage savings.
• What is a more appropriate grouping of Node types? I would argue putting syntax-only nodes, such as true Tokens (punctuation, operators, etc) and keywords into their own category, since they have no influence on checker performance.
• Depending on properly split Node types, what properties should each type have? For example, syntax-only Nodes do not need an id, but they still have now because of literals.

So far, I only looked at typecker performance when run through tsc. So far, I have no idea how tsserver works with regards to incremental parsing and checking. Is the type of a node immutable on incremental parsing / checking?

[RyanCavanaugh]

@rbuckton this looks like a promising angle of analysis - please follow up on the associated PR

[Swatinem]

BTW, here is a more detailed list of all the Nodes (SyntaxKind) that get created when compiling the ts compiler:

Oh github, Y U NO fenced code blocks inside of <details>?¿?

  1:     141
  8:  30,201
 10:  11,383
 13:      65
 14:      25
 15:     151
 16:      88
 17:     151
 24:  70,971
 25:     425
 27:      60
 28:     996
 30:     268
 31:     126
 32:     230
 35:   4,280
 36:   1,358
 37:   1,148
 38:     904
 39:     324
 40:      19
 41:       4
 42:       8
 43:       5
 46:     453
 47:      16
 48:       4
 49:   2,089
 50:   1,118
 51:       6
 52:      11
 54:   4,046
 55:   3,421
 56:   8,234
 57:   2,428
 60:  10,895
 61:     245
 62:       2
 63:       1
 67:       1
 70:      53
 71:     342
 72:       7
 73: 323,907
 78:     344
 86:   7,210
 88:   1,402
 94:       4
 95:       4
 97:     742
101:     138
103:   1,588
107:   3,293
114:      27
115:       4
116:       2
117:      43
121:   1,075
124:   3,429
126:     526
133:      73
134:     616
136:   5,177
137:      49
139:   6,392
140:     169
142:   3,878
144:      34
147:     105
149:   1,988
150:      87
151:   4,514
152:  49,256
154:  10,118
155:     295
156:   4,132
157:   1,990
158:      88
159:      11
160:       2
161:      44
162:     100
163:      64
164:     756
165:  42,527
166:   3,903
167:      72
168:     121
169:     641
170:   3,977
171:     157
173:       2
174:   6,856
175:     171
176:      41
177:      13
178:     262
179:   1,311
180:   1,780
181:     166
182:      26
183:   2,784
184:       3
185:     249
186:      34
187:     590
188:   1,190
189:   1,538
190:  60,681
191:   5,095
192:  37,215
193:      86
194:       2
195:   3,034
196:   4,413
197:   1,553
198:   4,423
199:       6
200:      91
201:     242
203:   4,270
204:     959
205:  33,534
206:   2,861
207:     159
208:       9
209:      91
210:       3
211:       2
212:   1,142
213:   1,011
214:     975
216:      30
217:     248
219:  28,823
220:  22,830
221:       1
222:  17,531
223:  13,239
224:      13
225:     368
226:     998
227:      60
228:     623
229:     193
230:     622
231:  13,644
233:     726
234:      22
235:      19
236:      49
237:       1
238:  16,191
239:  24,181
240:  24,794
241:     114
242:   2,927
243:     716
244:     361
245:     414
246:     398
247:     723
250:      51
251:      51
252:      34
253:      17
254:      28
255:       6
256:      29
257:      29
258:       2
272:   5,727
273:     229
274:     956
275:      33
276:   5,801
277:     703
278:      21
279:   5,265
280:       1
282:       1
285:     350
286:      26
287:       1
288:       2
289:       4
292:       1
293:       1
298:   4,309
301:     253
307:   2,540
308:      69
315:     877
316:   6,924
317:       1
318:      15
319:     199

[Swatinem]

I think this was tried and didn’t lead to any more wins. In any case, I won’t be working on this anymore.