[NaCl SDK] Fix create_nmf on dynamic bionic nexes

[NaCl SDK] Fix create_nmf on dynamic bionic nexes

create_nmf runs objdump on its shared objects to determine what
dependencies it has. libc.so is a dependency of bionic, and it is an ELF
file. glibc has libc.so as well, but it is not an ELF file, but a linker
script.

There was a check in get_shared_deps.py that prevented running objdump on
libc.so, if "bionic" is not in the path. Since we've added a package
called "bionic_canary", "bionic" is always in the path, so objdump was
being run on glibc's "libc.so", which fails.

The (hack) fix is to check if a relative path from the root of the SDK has
"bionic". This should only be true if the path is
"toolchain/linux_arm_bionic/...".

BUG=none
R=sbc@chromium.org, noelallen@chromium.org
TEST=test_all.py

Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=278133

[Sam Clegg, 2014-06-18 18:21:30]

https://codereview.chromium.org/341983002/diff/1/native_client_sdk/src/tools/...
File native_client_sdk/src/tools/lib/get_shared_deps.py (right):

https://codereview.chromium.org/341983002/diff/1/native_client_sdk/src/tools/...
native_client_sdk/src/tools/lib/get_shared_deps.py:218: if name == 'libc.so' and
'bionic' not in rel_dirname:
Why not just update the check to "_arm_bionic"?

Then this CL is just a single line.

https://codereview.chromium.org/341983002/diff/1/native_client_sdk/src/tools/...
File native_client_sdk/src/tools/lib/tests/get_shared_deps_test.py (right):

https://codereview.chromium.org/341983002/diff/1/native_client_sdk/src/tools/...
native_client_sdk/src/tools/lib/tests/get_shared_deps_test.py:22:
sys.path.append(BUILD_TOOLS_DIR)
These changes are not needed are they?

[binji, 2014-06-18 18:29:36]

https://codereview.chromium.org/341983002/diff/1/native_client_sdk/src/tools/...
File native_client_sdk/src/tools/lib/get_shared_deps.py (right):

https://codereview.chromium.org/341983002/diff/1/native_client_sdk/src/tools/...
native_client_sdk/src/tools/lib/get_shared_deps.py:218: if name == 'libc.so' and
'bionic' not in rel_dirname:
On 2014/06/18 18:21:30, Sam Clegg wrote:
> Why not just update the check to "_arm_bionic"?

Because the problem would pop up again if the user renames their directory.
Unlikely, but why not just fix it the right way?

> 
> Then this CL is just a single line.

https://codereview.chromium.org/341983002/diff/1/native_client_sdk/src/tools/...
File native_client_sdk/src/tools/lib/tests/get_shared_deps_test.py (right):

https://codereview.chromium.org/341983002/diff/1/native_client_sdk/src/tools/...
native_client_sdk/src/tools/lib/tests/get_shared_deps_test.py:22:
sys.path.append(BUILD_TOOLS_DIR)
On 2014/06/18 18:21:30, Sam Clegg wrote:
> These changes are not needed are they?

This test script doesn't run from the source directory without this change, only
from test_all.py. I could put it in another CL, but it seems related enough.

[Sam Clegg, 2014-06-18 18:36:23]

lgtm

https://codereview.chromium.org/341983002/diff/1/native_client_sdk/src/tools/...
File native_client_sdk/src/tools/lib/get_shared_deps.py (right):

https://codereview.chromium.org/341983002/diff/1/native_client_sdk/src/tools/...
native_client_sdk/src/tools/lib/get_shared_deps.py:218: if name == 'libc.so' and
'bionic' not in rel_dirname:
On 2014/06/18 18:29:35, binji wrote:
> On 2014/06/18 18:21:30, Sam Clegg wrote:
> > Why not just update the check to "_arm_bionic"?
> 
> Because the problem would pop up again if the user renames their directory.
> Unlikely, but why not just fix it the right way?

Ok.  I guess I was thinking that this check horrible fragile anyway, but I guess
it is actually less fragile your way.
> 
> > 
> > Then this CL is just a single line.

[binji, 2014-06-18 18:39:51]

https://codereview.chromium.org/341983002/diff/1/native_client_sdk/src/tools/...
File native_client_sdk/src/tools/lib/get_shared_deps.py (right):

https://codereview.chromium.org/341983002/diff/1/native_client_sdk/src/tools/...
native_client_sdk/src/tools/lib/get_shared_deps.py:218: if name == 'libc.so' and
'bionic' not in rel_dirname:
You're right, it's a hack either way. :) The real fix is specified in the TODO.

[binji, 2014-06-18 18:42:52]

Committed patchset #1 manually as r278133 (presubmit successful).