1464537 - 20k symbols are "<name omitted>" in XUL.sym for macOS and android builds

Status: RESOLVED FIXED

(Socorro :: General, task)

Description

[Mike Hommey [:glandium]]

The reason for the "<name omitted>"s is that those symbols have an undemanglable DW_AT_MIPS_linkage_name, and no DW_AT_name. That's not a problem on Linux/GCC where the symbols have a DW_AT_name. This is apparently a "feature" of clang when targetting osx (it doesn't happen when clang is targetting Linux).

Comment 1

[Mike Hommey [:glandium]]

Actually, that's the explanation for static initializers, but it's not for the others.

Comment 2

[Mike Hommey [:glandium]]

The others are due to "errors" like:
  the DIE at offset 0x5606 has a DW_AT_abstract_origin attribute referring to the die at offset 0x57b9, which either was not marked as an inline, or comes later in the file

That is, the breakpad processor doesn't know how to handle abstract origins that occur later in the dwarf data, and many symbols have no name but an abstract origin that occur later in the dwarf data.

Comment 3

[Mike Hommey [:glandium]]

On Android, another reason is that the dwarf info uses DWARF 4 DW_AT_linkage_name instead of the (old) extension DW_AT_MIPS_linkage_name. They have the same meaning (the former is essentially the standardized version of the latter), but the former is not supported by breakpad.

Comment 4

[Mike Hommey [:glandium]]

Attachment: Bug 1464537 - Add support for DWARF 4 DW_AT_linkage_name.

Review commit: https://reviewboard.mozilla.org/r/247378/diff/#index_header
See other reviews: https://reviewboard.mozilla.org/r/247378/

Comment 5

[Mike Hommey [:glandium]]

Attachment: Bug 1464537 - Fall back to the raw symbol name from DW_AT_MIPS_linkage_name when there is nothing else.

When DW_AT_MIPS_linkage_name doesn't demangle, breakpad currently throws
the symbol completely, but in some cases, there is no DW_AT_name or
DW_AT_abstract_origin to figure out a name, and the raw value from
DW_AT_MIPS_linkage_name is still better than nothing. Fall back to that
in when there is nothing else.

Review commit: https://reviewboard.mozilla.org/r/247380/diff/#index_header
See other reviews: https://reviewboard.mozilla.org/r/247380/

Comment 6

[Mike Hommey [:glandium]]

Note that the attached patches don't address comment 2, because that's much more work than I'm ready to put in. That's still a problem we should probably address somehow, though.

Comment 7

[(not currently active) Ted Mielczarek]

Comment on attachment 8981279 [details]
Bug 1464537 - Fall back to the raw symbol name from DW_AT_MIPS_linkage_name when there is nothing else.

https://reviewboard.mozilla.org/r/247380/#review254040

You should submit this upstream as well.

Comment 8

[(not currently active) Ted Mielczarek]

Comment on attachment 8981278 [details]
Bug 1464537 - Add support for DWARF 4 DW_AT_linkage_name.

https://reviewboard.mozilla.org/r/247378/#review254038

This should go upstream as well. If you submit it for upstream review I can land it there.

Comment 9

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/fbb3d4a2d904
Add support for DWARF 4 DW_AT_linkage_name. r=ted
https://hg.mozilla.org/integration/autoland/rev/6effe4ad8ac3
Fall back to the raw symbol name from DW_AT_MIPS_linkage_name when there is nothing else. r=ted

Comment 10

[Andreea Pavel [:apavel]]

https://hg.mozilla.org/mozilla-central/rev/fbb3d4a2d904
https://hg.mozilla.org/mozilla-central/rev/6effe4ad8ac3

Comment 11

[Mike Hommey [:glandium]]

(In reply to Ted Mielczarek [:ted.mielczarek] from comment #8)
> Comment on attachment 8981278 [details]
> Bug 1464537 - Add support for DWARF 4 DW_AT_linkage_name.
> 
> https://reviewboard.mozilla.org/r/247378/#review254038
> 
> This should go upstream as well. If you submit it for upstream review I can
> land it there.

That one is already supported upstream:
https://chromium.googlesource.com/breakpad/breakpad/+/7398ce15b79daf038cd07fbae4e37e183e99788d

almost 2 years...

Comment 12

[Mike Hommey [:glandium]]

https://chromium-review.googlesource.com/c/breakpad/breakpad/+/1082176

Comment 13

[Ionuț Goldan [:igoldan]]

Perf metrics are now pouring in. Looks like the test is back online.

== Change summary for alert #13557 (as of Wed, 30 May 2018 19:01:31 GMT) ==

Regressions:

  0%  compiler_metrics num_static_constructors osx-cross opt      0.00 -> 96.08
  0%  compiler_metrics num_static_constructors android-4-0-armv7-api16 opt 0.00 -> 87.08
  0%  compiler_metrics num_static_constructors android-4-2-x86 opt 0.00 -> 88.08
  0%  compiler_metrics num_static_constructors osx-cross debug    0.00 -> 198.00
  0%  compiler_metrics num_static_constructors android-5-0-aarch64 opt 0.00 -> 87.08
  0%  compiler_metrics num_static_constructors android-4-0-armv7-api16 debug 0.00 -> 116.08

For up to date results, see: https://treeherder.mozilla.org/perf.html#/alerts?id=13557