JSObject wastes too much memory on unused property slots

• Component: JavaScriptCore
• Status: RESOLVED FIXED

[Filip Pizlo]

1) Non-final objects should not have inline property storage, since it is used rarely and allocating out-of-line storage is cheap.

2) The initial size of out-of-line storage should not be a whopping 16 slots, or 128 bytes.  It should be some sensible amount, like maybe 4 (32 bytes), or 2x the inline storage, whichever is bigger.

[Filip Pizlo]

Created attachment 150108 [details]
the patch

[Mark Hahnenberg]

Comment on attachment 150108 [details]
the patch

View in context: https://bugs.webkit.org/attachment.cgi?id=150108&action=review

Overall, looks good. r=me with suggestions :-)

> Source/JavaScriptCore/runtime/JSObject.h:351
> +#if JSNonFinalObject_inlineStorageCapacity

I guess this is to make it easier if we want to change back to non-zero inline storage for JSNonFinalObjects in the future?

> Source/JavaScriptCore/runtime/JSObject.h:361
> +                globalData, structure,

Why on a separate line?

> Source/JavaScriptCore/runtime/JSObject.h:574
> +    if (propertyStorageCapacity() == inlineStorageCapacity())

You could assert that propertyStorageCapacity() >= inlineStorageCapacity() at the beginning of the function and then return propertyStorageCapacity() == inlineStorageCapacity(). Two less lines!

[Filip Pizlo]

Comment on attachment 150108 [details]
the patch

Found some bugs.  New patch on the way.

[Filip Pizlo]

Created attachment 150276 [details]
the patch

[Mark Hahnenberg]

Comment on attachment 150276 [details]
the patch

r=me

[Filip Pizlo]

Landed in http://trac.webkit.org/changeset/121605

[Zan Dobersek]

Re-opened since this is blocked by 90336

[Filip Pizlo]

Relanded with fixes in http://trac.webkit.org/changeset/121633.