docs(ivy): add a note about exporting top level variables (#22234)

PR Close #22234
This commit is contained in:
Victor Berchet 2018-02-14 17:14:49 -08:00
parent f755db78dc
commit d241532488
2 changed files with 53 additions and 19 deletions

View File

@ -0,0 +1,53 @@
## General Notes
Each Array costs 70 bytes and is composed of `Array` and `(array)` object
* `Array` javascript visible object: 32 bytes
* `(array)` VM object where the array is actually stored in: 38 bytes
Each Object cost is 24 bytes plus 8 bytes per property.
For small arrays, it is more efficient to store the data as a linked list
of items rather than small arrays. However, the array access is faster as
shown here: https://jsperf.com/small-arrays-vs-linked-objects
## Monomorphic vs Megamorphic code
Great read: [What's up with monomorphism?](http://mrale.ph/blog/2015/01/11/whats-up-with-monomorphism.html)
1) Monomorphic prop access is 100 times faster then megamorphic.
2) Monomorphic call is 4 times faster the megamorphic call.
See benchmark [here](https://jsperf.com/mono-vs-megamorphic-property-access).
## Exporting top level variables
Exporting top level variables should be avoided where possible where performance
and code size matters:
```
// Typescript
export let exported = 0;
let notExported = 0;
notExported = exported;
// Would be compiled to
exports.exported = 0;
var notExported = 0;
notExported = exports.exported;
```
Most minifiers do not rename properties (closure is an exception here).
What could be done instead is:
```
let exported = 0;
export function getExported() { return exported; }
export function setExported(v) { exported = v; }
```
Also writing to a property of `exports` might change its hidden class resulting in megamorphic access.

View File

@ -1,19 +0,0 @@
## General Notes
Each Array costs 70 bytes and is composed of `Array` and `(array)` object
* `Array` javascript visible object: 32 bytes
* `(array)` VM object where the array is actually stored in: 38 bytes
Each Object cost is 24 bytes plus 8 bytes per property.
For small arrays, it is more efficient to store the data as a linked list
of items rather than small arrays. However, the array access is faster as
shown here: https://jsperf.com/small-arrays-vs-linked-objects
## Monomorphic vs Megamorphic code
1) Monomophic prop access is 100 times faster then megamorphic.
2) Monomorphic call is 4 times faster the megamorphic call.
See benchmark [here](https://jsperf.com/mono-vs-megamorphic-property-access).