1278754 - add Vector::initLengthUninitialized

Status: RESOLVED FIXED

(Core :: MFBT, defect)

Description

[Luke Wagner [:luke]]

Attachment: add-init-length

This patch introduces a new method initLengthUninitialized() which is equivalent to, but nicer to read than:

  initCapacity(length)
  infallibleGrowByUninitialized(length)

In particular, this is useful for Vectors that are grown once to a target size and thus do not need the over-allocation that resize() provides.  Using this method instead of plain resize() saves 10mb on the wasm Unity demo.

Comment 1

[Luke Wagner [:luke]]

Attachment: add-init-length

Oops, curlies.

Comment 2

[Jeff Walden [:Waldo]]

Comment on attachment 8761064 [details] [diff] [review]
add-init-length

Review of attachment 8761064 [details] [diff] [review]:
-----------------------------------------------------------------

Possibly worth adding docs along the lines of, "If you don't subsequently initialize the elements you've allocated before the vector is destroyed, behavior is undefined." or somesuch.

::: mfbt/Vector.h
@@ +981,5 @@
> +Vector<T, N, AP>::initLengthUninitialized(size_t aRequest)
> +{
> +  if (!initCapacity(aRequest)) {
> +    return false;
> +  }

Bah, braces.  So stoopid.

Comment 3

[Luke Wagner [:luke]]

Well, technically I think it's totally well-defined if you're destroying POD types (and non-virtual classes with trivial or boring destructors etc etc), so I don't think I need to sound quite that scary.  And it's not just destruction either; there's also resize (copy) etc.  I'll update the first sentence to be more clear that its totally uninitialized memory (with all the caveats that comes with), though.

Comment 4

[Luke Wagner [:luke]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/03765412b2a8

Comment 5

[Guilherme Lima]

https://hg.mozilla.org/mozilla-central/rev/03765412b2a8 - landed this morning.