Update pnacl/DEPS to bring in stricter ABI checks for metadata

Update pnacl/DEPS to bring in stricter ABI checks for metadata.

f8b761e Check for metadata in PNaCl ABI checker.

Use -pnaclabi-allow-debug-metadata when checking the ABI
in pnacl-ld, since we do not strip debug metadata until
pnacl-finalize is run.

On the other hand pnaclabi-check will not have that flag
by default and do the full check.  Also run the
-strip-metadata pass after optimizations in pnacl-ld to get
through the non-debug-metadata checks.

Bump DEPS at the same time, since we need to have the
-allow-debug-metadata to pass validation.


BUG= http://code.google.com/p/nativeclient/issues/detail?id=3348
R=mseaborn@chromium.org

Committed: https://src.chromium.org/viewvc/native_client?view=rev&revision=11252

[Mark Seaborn, 2013-04-25 18:03:35]

https://codereview.chromium.org/14358048/diff/2001/pnacl/driver/pnacl-ld.py
File pnacl/driver/pnacl-ld.py (right):

https://codereview.chromium.org/14358048/diff/2001/pnacl/driver/pnacl-ld.py#n...
pnacl/driver/pnacl-ld.py:403: RunDriver('abicheck', [output,
'-allow-debug-metadata'])
Why a separate invocation?  This will be slower.  It does make it easier to
debug when the ABI verifier rejects the code, but maybe we should handle that
sort of debugging via something like -save-temps?

[jvoung (off chromium), 2013-04-26 21:24:48]

https://codereview.chromium.org/14358048/diff/2001/pnacl/driver/pnacl-ld.py
File pnacl/driver/pnacl-ld.py (right):

https://codereview.chromium.org/14358048/diff/2001/pnacl/driver/pnacl-ld.py#n...
pnacl/driver/pnacl-ld.py:403: RunDriver('abicheck', [output,
'-allow-debug-metadata'])
On 2013/04/25 18:03:35, Mark Seaborn wrote:
> Why a separate invocation?  This will be slower.  It does make it easier to
> debug when the ABI verifier rejects the code, but maybe we should handle that
> sort of debugging via something like -save-temps?

Merged the check back into the other run of opt.

[Mark Seaborn, 2013-04-26 21:36:45]

Maybe the description should be "Update pnacl/DEPS to bring in stricter ABI
checks for metadata"?  That's the main change.  Adding
"-pnaclabi-allow-debug-metadata" is a secondary thing to keep the checks
working.

LGTM

https://codereview.chromium.org/14358048/diff/9001/pnacl/DEPS
File pnacl/DEPS (right):

https://codereview.chromium.org/14358048/diff/9001/pnacl/DEPS#newcode3
pnacl/DEPS:3: "pnacl_llvm_rev": "f8b761e8e49cf46f796510e5ef946b2a261dd45b",
Maybe list the changes you're rolling in in the commit message?  (I've tended to
keep the rolls separate for clarity.)

https://codereview.chromium.org/14358048/diff/9001/pnacl/driver/pnacl-ld.py
File pnacl/driver/pnacl-ld.py (right):

https://codereview.chromium.org/14358048/diff/9001/pnacl/driver/pnacl-ld.py#n...
pnacl/driver/pnacl-ld.py:381: # We place '-strip-metadata' after optimization
passes are run
It would be clearer to omit "are run"

https://codereview.chromium.org/14358048/diff/9001/pnacl/driver/pnacl-ld.py#n...
pnacl/driver/pnacl-ld.py:391: '-pnaclabi-allow-debug-metadata']
Maybe comment "This option modifies the behaviour of -verify-pnaclabi-*".  It's
odd having it in a list called "passes" otherwise.

[jvoung (off chromium), 2013-04-26 22:54:51]

https://codereview.chromium.org/14358048/diff/9001/pnacl/DEPS
File pnacl/DEPS (right):

https://codereview.chromium.org/14358048/diff/9001/pnacl/DEPS#newcode3
pnacl/DEPS:3: "pnacl_llvm_rev": "f8b761e8e49cf46f796510e5ef946b2a261dd45b",
On 2013/04/26 21:36:45, Mark Seaborn wrote:
> Maybe list the changes you're rolling in in the commit message?  (I've tended
to
> keep the rolls separate for clarity.)

Listed.

https://codereview.chromium.org/14358048/diff/9001/pnacl/driver/pnacl-ld.py
File pnacl/driver/pnacl-ld.py (right):

https://codereview.chromium.org/14358048/diff/9001/pnacl/driver/pnacl-ld.py#n...
pnacl/driver/pnacl-ld.py:381: # We place '-strip-metadata' after optimization
passes are run
On 2013/04/26 21:36:45, Mark Seaborn wrote:
> It would be clearer to omit "are run"

Done.

https://codereview.chromium.org/14358048/diff/9001/pnacl/driver/pnacl-ld.py#n...
pnacl/driver/pnacl-ld.py:391: '-pnaclabi-allow-debug-metadata']
On 2013/04/26 21:36:45, Mark Seaborn wrote:
> Maybe comment "This option modifies the behaviour of -verify-pnaclabi-*". 
It's
> odd having it in a list called "passes" otherwise.

Done.

[jvoung (off chromium), 2013-04-29 16:35:00]

Committed patchset #5 manually as r11252 (presubmit successful).