823641 - Enable Dark Matter Detector on Daily builds (memory)

Status: RESOLVED WONTFIX

(Thunderbird :: Build Config, defect)

Description

[Mike Conley (:mconley) (:⚙️)]

According to https://wiki.mozilla.org/Performance/MemShrink/DMD, the Dark Matter Detector allows Gecko apps to drill down and see what the "heap unclassified" memory allocations are all about.

Is this something we want to activate on Daily builds, like we do with profiling?

Comment 1

[Wayne Mery (:wsmwk)]

(In reply to Mike Conley (:mconley) from comment #0)
> Is this something we want to activate on Daily builds, like we do with
> profiling?

Is Mark the right person to answer?
I don't know enough about DMD to know if or how it will help us.

Comment 2

[Mark Banner (:standard8)]

Firefox doesn't seem to activate this by default, even on Daily. So I'd probably start with asking why they don't.

Also, this I'd view this as a community decision to be discussed on tb-planning.

Comment 3

[Wayne Mery (:wsmwk)]

(In reply to Mark Banner (:standard8) from comment #2)
> Firefox doesn't seem to activate this by default, even on Daily. So I'd
> probably start with asking why they don't.

perhaps because of bugs like bug 819839 and bug 827283?

njn ?

> Also, this I'd view this as a community decision to be discussed on
> tb-planning.

good point

Comment 4

[Andrew McCreight [:mccr8]]

Is anybody actively working on adding memory reporters for Thunderbird, or seeing what kinds of things result in high heap-unclassified in about:memory?  If not, I don't think any performance hit from this (I don't know how bad it is, but it is certainly much worse than enabling profiling) would be worthwhile, because nobody is going to make use of the results.

In Firefox, most heap-unclassified problems go like "I get high heap-unclassified on this page", so it is easy enough to just reproduce user-reported issues in a special profiling build.

Comment 5

[Justin Lebar (not reading bugmail)]

You can build with --enable-dmd and then run without DMD enabled (i.e., without setting DMD=1).  In that case, I'd expect the overhead of DMD to be relatively small.

But I agree with everything Andrew said.

Comment 6

[Nicholas Nethercote [inactive]]

The slowdown is small but noticeable, and it's not of wide enough interest and usability for it to be worth turning on for Daily users, IMO.

Comment 7

[Wayne Mery (:wsmwk)]

(In reply to Nicholas Nethercote [:njn] from comment #6)
> The slowdown is small but noticeable, and it's not of wide enough interest
> and usability for it to be worth turning on for Daily users, IMO.

is this an actual assessment of you having used it in Thunderbird?

Comment 8

[Wayne Mery (:wsmwk)]

(In reply to Andrew McCreight [:mccr8] from comment #4)
> Is anybody actively working on adding memory reporters for Thunderbird, 

yes


(In reply to Andrew McCreight [:mccr8] from comment #4)
> In Firefox, most heap-unclassified problems go like "I get high
> heap-unclassified on this page", so it is easy enough to just reproduce
> user-reported issues in a special profiling build.

I'm not sure this reasoning, "reproduction is normally easy", applies to Thunderbird.  Often, only the user can reproduce a crash or other type of problem.

Also, Thunderbird is new to the memory analysis game, and could benefit from a period of making analysis as easy as possible. 

So if the hit of DMD is minimal/not noticeable to the user I'd say it's worth enabling it nightly just so we don't have to hassle of making special builds when we want to test a problem.

Comment 9

[Nicholas Nethercote [inactive]]

> is this an actual assessment of you having used it in Thunderbird?

I haven't tried it on Thunderbird.  Have you?  I'm confident that the effects it has on Firefox will be similar to the effects on Thunderbird, because it slows down malloc/free and both apps, like most apps, use those operations frequently.

DMD slows things down.  It doesn't work properly on Windows (bug 819839) and Android (bug 823354).  If you can live with that and you think the benefit outweighs the costs, then by all means turn it on.

Comment 10

[Wayne Mery (:wsmwk)]

mconley, given the assessments above show we close this wontfix for now?

I'm happy to get what I need for bug 805023 via special builds

Comment 11

[Mike Conley (:mconley) (:⚙️)]

(In reply to Wayne Mery (:wsmwk) from comment #10)
> mconley, given the assessments above show we close this wontfix for now?
> 

Yep.