960350 - Merged B2G profiles should display untranslated addresses correctly

Status: RESOLVED FIXED

(Core :: Gecko Profiler, defect)

Description

[Jed Davis [:jld] ⟨⏰|UTC-7⟩ ⟦he/him⟧]

If some of the addresses in a B2G profiles aren't translated, the merging script will prepend the pid (formatted with %05d) but not include a translation back to the original non-disambiguated name, so it will show up in Cleopatra as "031760x403108e4" or similar, which is confusing.

Comment 1

[Jed Davis [:jld] ⟨⏰|UTC-7⟩ ⟦he/him⟧]

Attachment: bug960350-profile-merge-unxlat-hg0.diff

I also changed the address disambiguation to set off the pid with a colon.  (Originally I was going to do just that, but then I realized I could hide the (redundant) pid entirely.)  It probably doesn't matter now that the result isn't shown to the user, but it also shouldn't hurt.  (And the field with of 5 might be obsolete someday.)

Comment 2

[Benoit Girard (:BenWa)]

I wonder if the merge script should just convert into some lib+offset format instead and then everything can be simplier. Something like libxul.so+0x1234. I believe the profiler uses to do that but we change it and I feel like its a big mistake now. We went away from that and stored absolute address + the shared library map because with that we can include some version hash (used only on windows ATM). May now is the right time to backup on this mistake. What do you think?

Comment 3

[Jed Davis [:jld] ⟨⏰|UTC-7⟩ ⟦he/him⟧]

(In reply to Benoit Girard (:BenWa) from comment #2)
> I wonder if the merge script should just convert into some lib+offset format
> instead and then everything can be simplier. Something like
> libxul.so+0x1234. I believe the profiler uses to do that but we change it
> and I feel like its a big mistake now. We went away from that and stored
> absolute address + the shared library map because with that we can include
> some version hash (used only on windows ATM). May now is the right time to
> backup on this mistake. What do you think?

The problem there is that we're currently running the merge script after symbolication, so addresses that are within a library have already been mapped to names.  Redoing the conversion into lib+offset at this point seems not very helpful.

Comment 4

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/integration/b2g-inbound/rev/dbaedf87a3cf

Comment 5

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/dbaedf87a3cf