Skip to content

GitLab

Switch branch/tag
  1. 02 Nov, 2015 4 commits
  3. 30 Oct, 2015 1 commit
  5. 29 Oct, 2015 3 commits
    • Dan Albert's avatar
      Prevent ODR issues with platform libc++. · 00ddc291
      Dan Albert authored 
      Android L and later uses libc++ in platform libraries, and there will
be ODR issues when using libc++ in the NDK on those platforms.

One destructive way this manifests is that the standard streams, which
are initialized by a static constructor, get initialized twice (first
by netd, then by the app), resulting in the app failing to load. This
was resulting in 364 test failures when running on AOSP ToT volantis
(54 remaining afterward).

Change-Id: Ic2af5a570bb0b6677708801fb1aed6cc6f4f15e4
      00ddc291
    • Dan Albert's avatar
      Get LIT working for NDK libc++ tests. · ea5666be
      Dan Albert authored 
      Part of this patch is copying in the libc++ test runner from r244462.
It includes all the work we've done upstream to make testing cross
compiled targets much simpler. The one thing it drops is the use of
`-include nasty_macros.h`, which is just used to make sure we don't
have awful things like _A in our namespace (those are the names used
by some C libraries for the ctype classes).

It then adds all the Android specific testing functionality
(libcxx.android) that is used in the platform. Some day the platform
is actually going to be the upstream for this, but we're going to
suffer through the copy for now (no modifications were needed).

Finally, it adds the NDK specific testing code (libcxx.ndk). This is
thankfully a pretty thin wrapper around the upstream libc++ test
config (albeit with a lot of it disabled/skipped) and the Android
platform test format.

Exception handling doesn't work for ARM because libgcc is getting used
instead of the ARM EHABI unwind library (deja vu). We don't seem to
actually build a separate static library for the ARM unwinder, so
we'll need to fix that separately.

This doesn't link the correct versions of libatomic/libgcc.  We're
never linking thumb, and we're always using arm5.

Current results (run on ~ToT AOSP volantis):

    Expected Passes    : 4322
    Unsupported Tests  : 5
    Unexpected Failures: 418

A lot of those failures are from ODR issues in iostreams (both netd
and the test were trying to run the static constructors for the same
object). This is a separate problem we'll have to solve. That's about
all the diagnosis I've done so far. I'll look closer once this is in
shape and submitted and the noise from ODR has died down.

Bug: http://b/25329002
Change-Id: Ie9858120ac5aa3c872e5cc964e9792df9b3fc8fa
      ea5666be
    • Dan Albert's avatar
      Also copy libatomic with gcclibs. · 88d603ac
      Dan Albert authored 
      Change-Id: I79e10b0a6877572a9fdb8e52d4f81253c0cc0249
      88d603ac
  7. 28 Oct, 2015 1 commit
    • Dan Albert's avatar
      Merge changes I8762ffbe,I19991a97,I35b086e6,Ie38e804b,Ib651f749, ... · 229abaf2
      Dan Albert authored 
      * changes:
  Fix libc++ support lib for standalone toolchains.
  Drop GCC version from the gnustl include path.
  Fix toolchain locations.
  Make a bad error message more useful.
  Add a shebang to make-standalone-toolchain.sh.
  Add a test for make-standalone-toolchain.sh.
  Add a new build test type: test.py.
      229abaf2
  9. 27 Oct, 2015 9 commits
  11. 26 Oct, 2015 9 commits
  13. 24 Oct, 2015 3 commits
  15. 21 Oct, 2015 4 commits
  17. 20 Oct, 2015 4 commits
  19. 16 Oct, 2015 2 commits