The analyzer makes a copy of `lineStarts` without the last element,
and copying it in to a `Uint16List` or an `Uint32List` depending on
what's needed (to avoid using more ram than necessary).
Because the (actual) scanner (`AbstractScanner`) does the same trick
with `Uint16List`/`Uint32List` the linestarts output from the scanner,
while a `List<int>`, is actually a `LineStarts` class, and copying that
into a `Uint16List` (or `Uint32List`) goes into
`_TypedIntListMixin._SlowSetRange` which as the name suggests is slower
than it has to be.
This CL puts the copying into the `LineStarts` class where it actually
has a `Uint16List`/`Uint32List` already, and doing the copy from that
instead avoids the slowdown.
Numbers from `perf stat`:
Before:
```
39,552,191,101 instructions:u
39,552,058,132 instructions:u
39,552,065,649 instructions:u
```
After:
```
39,419,426,977 instructions:u
39,419,429,914 instructions:u
39,419,441,613 instructions:u
```
Saving: about 132 million instructions.
Change-Id: Ibf6c7f6cc9324a54d2bd99c37058103af106164c
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/508580
Reviewed-by: Johnni Winther <johnniwinther@google.com>
Jens Johansen
0c2d7decd7