You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
83 lines
2.9 KiB
83 lines
2.9 KiB
3 years ago
|
# flatstr
|
||
|
|
|
||
|
|
Flattens the underlying C structures of a concatenated JavaScript string
|
||
|
|
|
||
|
|
## About
|
||
|
|
|
||
|
|
If you're doing lots of string concatenation and then writing that
|
||
|
|
string somewhere, you may find that passing your string through
|
||
|
|
`flatstr` vastly improves performance.
|
||
|
|
|
||
|
|
## Usage
|
||
|
|
|
||
|
|
```js
|
||
|
|
var flatstr = require('flatstr')
|
||
|
|
flatstr(someHeavilyConcatenatedString)
|
||
|
|
```
|
||
|
|
|
||
|
|
## Benchmarks
|
||
|
|
|
||
|
|
Benchmarks test flat vs non-flat strings being written to
|
||
|
|
an `fs.WriteStream`.
|
||
|
|
|
||
|
|
```
|
||
|
|
unflattenedManySmallConcats*10000: 147.540ms
|
||
|
|
flattenedManySmallConcats*10000: 105.994ms
|
||
|
|
unflattenedSeveralLargeConcats*10000: 287.901ms
|
||
|
|
flattenedSeveralLargeConcats*10000: 226.121ms
|
||
|
|
unflattenedExponentialSmallConcats*10000: 410.533ms
|
||
|
|
flattenedExponentialSmallConcats*10000: 219.973ms
|
||
|
|
unflattenedExponentialLargeConcats*10000: 2774.230ms
|
||
|
|
flattenedExponentialLargeConcats*10000: 1862.815ms
|
||
|
|
```
|
||
|
|
|
||
|
|
In each case, flattened strings win,
|
||
|
|
here's the performance gains from using `flatstr`
|
||
|
|
|
||
|
|
```
|
||
|
|
ManySmallConcats: 28%
|
||
|
|
SeveralLargeConcats: 21%
|
||
|
|
ExponentialSmallConcats: 46%
|
||
|
|
ExponentialLargeConcats: 33%
|
||
|
|
```
|
||
|
|
|
||
|
|
## How does it work
|
||
|
|
|
||
|
|
In the v8 C++ layer, JavaScript strings can be represented in two ways.
|
||
|
|
|
||
|
|
1. As an array
|
||
|
|
2. As a tree
|
||
|
|
|
||
|
|
When JavaScript strings are concatenated, tree structures are used
|
||
|
|
to represent them. For the concat operation, this is cheaper than
|
||
|
|
reallocating a larger array. However, performing other operations
|
||
|
|
on the tree structures can become costly (particularly where lots of
|
||
|
|
concatenation has occurred).
|
||
|
|
|
||
|
|
V8 has a a method called `String::Flatten`which converts the tree into a C array. This method is typically called before operations that walk through the bytes of the string (for instance, when testing against a regular expression). It may also be called if a string is accessed many times over,
|
||
|
|
as an optimization on the string. However, strings aren't always flattened. One example is when we pass a string into a `WriteStream`, at some point the string will be converted to a buffer, and this may be expensive if the underlying representation is a tree.
|
||
|
|
|
||
|
|
`String::Flatten` is not exposed as a JavaScript function, but it can be triggered as a side effect.
|
||
|
|
|
||
|
|
There are several ways to indirectly call `String::Flatten` (see `alt-benchmark.js`),
|
||
|
|
but coercion to a number appears to be (one of) the cheapest.
|
||
|
|
|
||
|
|
However since Node 10 the V8 version has stopped using Flatten in all
|
||
|
|
places identified. Thus the code has been updated to seamlessly
|
||
|
|
use the native runtime function `%FlattenString` without having to use
|
||
|
|
the `--allow-natives-syntax` flag directly.
|
||
|
|
|
||
|
|
One final note: calling flatstr too much can in fact negatively effect performance. For instance, don't call it every time you concat (if that
|
||
|
|
was performant, v8 wouldn't be using trees in the first place). The best
|
||
|
|
place to use flatstr is just prior to passing it to an API that eventually
|
||
|
|
runs non-v8 code (such as `fs.WriteStream`, or perhaps `xhr` or DOM apis in the browser).
|
||
|
|
|
||
|
|
|
||
|
|
## Acknowledgements
|
||
|
|
|
||
|
|
* Sponsored by nearForm
|
||
|
|
|
||
|
|
## License
|
||
|
|
|
||
|
|
MIT
|