112470 - Make MOZ_TIMELINE and MOZ_TRACE_MALLOC default

Status: VERIFIED FIXED

(SeaMonkey :: Build Config, defect, P1)

Description

[Suresh Duddi (gone)]

Making these default will eliminate barrier to perf and footprint work.

a debug build (default build) should turn these ON by default. These should be
individually ON/OFF able.

(for linux look at leak-webshell-count to see how to do it right - dbaron)

Comment 1

[Suresh Duddi (gone)]

Things to check / fix:

- ensure trace_malloc doesn't cost unless enabled
- MOZ_TIMELINE should not print by default. NS_TIMELINE_ENABLE should be
required for it to print.

Comment 2

[Suresh Duddi (gone)]

Attachment: Windows: TIMELINE and TRACE_MALLOC on by default

TIMELINE outputs wont be printed by default. NS_TIMELINE_ENABLE would be
required. To check this in mcafee needs to set that for tinderboxes.

Comment 3

[Suresh Duddi (gone)]

Attachment: Windows: MOZ_TIMELINE and MOZ_TRACE_MALLOC on for debug builds

xpcom/base/makefile.win was the only makefile using MOZ_TRACE_MALLOC and I
didnt of any other way for making it pick up the trace malloc setting.
rules.mak was included after setting all of EXPORTS, C_OBJS and CPP_OBJS and
that was when we needed the setting.

If this is too ugly, I could include all the files that for specific to trace
malloc unconditionally and ifdef the files toally out in their source. But then
that would still get this one file nsTraceMalloc.h into include which I didnt'
like. Hence the hack.

Comment 4

[hacker formerly known as [e-mail]]

Umm, attachment 59955 [details] [diff] [review] looks like an updated version of attachment 59878 [details] [diff] [review].  I
don't see the xpcom/base/makefile.win changes that you're talking about.  What's
the ETA on the unix & mac changes?

Comment 5

[Suresh Duddi (gone)]

Comment on attachment 59955 [details] [diff] [review]
Windows: MOZ_TIMELINE and MOZ_TRACE_MALLOC on for debug builds

Seawood, the xpcom/base/makefile.win patch is at the top of this attachment.
Maybe your cache was playing funny games with you.

Unix is monday.

mac - I need simons help. I hope to try get it in by next weekend.

Comment 6

[Suresh Duddi (gone)]

Attachment: unix: timeline and trace_malloc enabled default for debug builds

- should I checkin configure or configure.in I vaguely remember I should change
configure.in and checking it in will cause configure to change automatically.
For now the patch contains configure changed.

- makes sure --disable-timeline or --disable-trace-malloc is obeyed as override


- only for debug timeline and tracemalloc turned ON

Comment 7

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

> - should I checkin configure or configure.in I vaguely remember I should change
> configure.in and checking it in will cause configure to change automatically.
> For now the patch contains configure changed.

Check in configure.in, and configure will be updated automatically within 15
minutes.

Comment 8

[Suresh Duddi (gone)]

Attachment: windows: timeline and trace malloc enabled default for debug builds

same as patch 59955. With diff -u

Comment 9

[Suresh Duddi (gone)]

Ccing Garrett as he is also touching this area to get trace-malloc working on
optimized builds on windows.

Comment 10

[Cathleen]

Comment on attachment 60222 [details] [diff] [review]
windows: timeline and trace malloc enabled default for debug builds

r=cathleen

Comment 11

[Garrett Arch Blythe]

Comment on attachment 60068 [details] [diff] [review]
unix: timeline and trace_malloc enabled default for debug builds

r=blythe

Comment 12

[Suresh Duddi (gone)]

As part of this, should we also pull/build space-trave
mozilla/tools/trace-malloc by default ? I think it would make developers life
easy. What do others think ?

Comment 13

[Garrett Arch Blythe]

I agree that they will need the tools build in order to use the default
functionality.  they build really quick from a time standpoint.

Comment 14

[Cathleen]

I agree to pull/build mozilla/tools/trace-malloc.  otherwise, once you generate
the data from running the build, you would still need to pull/build the tool to
analyze that big chuck of data.

Comment 15

[Alec Flett]

Comment on attachment 60222 [details] [diff] [review]
windows: timeline and trace malloc enabled default for debug builds

wait, there is a FAR easier way to fix xpcom/base/makefile.win
I'll attach a patch momentarily

however, sr=alecf on the rest

Comment 16

[Alec Flett]

Attachment: include config.mak early

Comment 17

[Suresh Duddi (gone)]

Alec, excellent. That works. I will adopt that in instead of my hack. Thanks.

Comment 18

[Suresh Duddi (gone)]

Attachment: Unix: MOZ_TIMELINE and TRACE_MALLOC ON by default for DEBUG builds

Here is the configure.in patch.

Comment 19

[hacker formerly known as [e-mail]]

Comment on attachment 60411 [details] [diff] [review]
Unix: MOZ_TIMELINE and TRACE_MALLOC ON by default for DEBUG builds

sr=cls

Comment 20

[Darin Fisher]

this is causing redness on shrike linux (among others)... i'm not sure what
needs to be done to fix it.

Comment 21

[Suresh Duddi (gone)]

Here is why : mozilla/Makefile.in has

< ifdef NS_TRACE_MALLOC
< tier_9_dirs   += tools/trace-malloc
< endif
---
> #ifdef NS_TRACE_MALLOC
> #tier_9_dirs  += tools/trace-malloc
> #endif

This is the patch from my tree (that I didn't include with my configure.in
patch!!!).

But then, this probably would break some of the tinderboxes building trace
malloc tools and getting bloatblame numbers. Hence I decided to back my change out.

Solution would be to pull tools/trace-malloc by default. We wanted to do that
anyway. Guess we need to do that along with the configure.in change.

Comment 22

[Suresh Duddi (gone)]

We did this one windows. We didn't do this on unix. I dont belive a lot of folks
are using this. So I am going to call it quits and move on.

Comment 23

[Tracy Walker [:tracy]]

verified