814229 - Measure data hanging off CTypes objects

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Nicholas Nethercote [inactive]]

Attachment 683892 [details] has stacks indicating that we have ~100KiB of ctypes data that's been created by ConstructData().  This bug is about adding memory reporting for that data.

Comment 1

[Nicholas Nethercote [inactive]]

Attachment: (part 1) - Add "objects-extra/ctypes-data" memory report.

This patch adds the "objects-extra/ctypes-data" measurement.
SizeOfDataIfCDataObject() is a bit clumsy but it has the virtue of being the
smallest change to the ctypes interface that I could come up with.

Comment 2

[Nicholas Nethercote [inactive]]

Attachment: (part 2) - Refactor storage of the "objects-extra" numbers.

This patch does some clean-up:

- The six measurements reported under the "objects-extra" sub-tree in
  about:memory are collated into a new struct ObjectsExtraSizes, which is
  similar to the existing TypeInferenceSizes.

- TypeInferenceSizes' constructor is simplified to use memset to zero
  everything.

- CompartmentStats::typeInferenceSizes is renamed as
  CompartmentStats::typeInference, which matches the general rule that names
  of things in CompartmentStats match the strings used in about:memory (i.e.
  "type-inference" in this case).

- TypeObject::sizeOfExcludingThis() is changed to return a value instead of
  updating a field in the TypeInferenceSizes struct.  This makes it like other
  sizeOf*() functions.

diffstat says:

  7 files changed, 86 insertions(+), 112 deletions(-)

Comment 3

[Jason Orendorff [:jorendorff]]

Comment on attachment 684242 [details] [diff] [review]
(part 1) - Add "objects-extra/ctypes-data" memory report.

Review of attachment 684242 [details] [diff] [review]:
-----------------------------------------------------------------

::: js/src/ctypes/CTypes.cpp
@@ +1338,5 @@
> +        return 0;
> +
> +    size_t n = 0;
> +    jsval slot = JS_GetReservedSlot(obj, ctypes::SLOT_OWNS);
> +    if (!JSVAL_IS_VOID(slot)) {

Yeah, the style of the surrounding file is pretty crufty. But feel free to use slot.isUndefined() and so forth if you like.

::: js/xpconnect/src/XPCJSRuntime.cpp
@@ +1586,5 @@
> +
> +    CREPORT_BYTES(cJSPathPrefix + NS_LITERAL_CSTRING("objects-extra/ctypes-data"),
> +                  cStats.objectsExtraCtypesData,
> +                  "Memory allocated for data belonging to ctypes objects.");
> + 

Looks like there's a space character on this blank line.

Comment 4

[Nicholas Nethercote [inactive]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/f3fc7f4f20a0
https://hg.mozilla.org/integration/mozilla-inbound/rev/9b58f6e07b21

Comment 5

[Ed Morley [:emorley]]

https://hg.mozilla.org/mozilla-central/rev/f3fc7f4f20a0
https://hg.mozilla.org/mozilla-central/rev/9b58f6e07b21

Comment 6

[Justin Lebar (not reading bugmail)]

njn mentioned this bug to me on IRC for potential b2g uplift.

I'm tempted to mostly give up on B2G heap-unclassified until we have reporting for workers with ctypes, since that's upwards of 30% of our dark matter.

Also uplifting these guys is often a pain because last time I checked we haven't uplifted /all/ of our changes to these files.

But let me know if you think this might help and I or someone can do some tests on B2G to see how much memory these cover.

Comment 7

[Nicholas Nethercote [inactive]]

It's fine to not uplift this one.  It is probably marginally useful at best.

Comment 8

[David Anderson [:dvander] - inactive, e-mail if emergency]

This broke threadsafe shell builds, with an undefined symbol "js::SizeOfDataIfCDataObject". AWFY can't build anymore.

Comment 9

[Nicholas Nethercote [inactive]]

Attachment: (part 3) - Unbreak JS shell builds that use --enable-threadsafe.

Sorry about the bustage.  Here's a fix.  I tested it on:

- A normal shell build.
- An --enable-threadsafe shell build.
- A normal browser build.

Comment 10

[David Anderson [:dvander] - inactive, e-mail if emergency]

Comment on attachment 694700 [details] [diff] [review]
(part 3) - Unbreak JS shell builds that use --enable-threadsafe.

Review of attachment 694700 [details] [diff] [review]:
-----------------------------------------------------------------

Thanks!

Comment 11

[Nicholas Nethercote [inactive]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/fe8429f81df8

Comment 12

[:Ms2ger (he/him; ⌚ UTC+1/+2)]

https://hg.mozilla.org/mozilla-central/rev/fe8429f81df8