Use the installed/downloaded objdump instead of work-dir one.

Use the installed/downloaded objdump instead of work-dir one.

Sometimes the work dir is empty and only cached results are used.
I couldn't switch the LLVM one from work-dir to install/download-dir
because we prune all the .a file in the download-dir.

BUG=none
R=kschimpf@google.com, stichnot@chromium.org

Committed: https://gerrit.chromium.org/gerrit/gitweb?p=native_client/pnacl-subzero.git;a=commit;h=68a06338e6cb4964212128169a69da52037874ce

[jvoung (off chromium), 2015-03-05 16:51:44]

jvoung@chromium.org changed reviewers:
+ kschimpf@google.com, stichnot@chromium.org

[jvoung (off chromium), 2015-03-05 16:52:25]

Happened to choose the download dir vs the install dir, assuming binutils
doesn't change much.

[Karl, 2015-03-05 17:09:10]

lgtm

[Jim Stichnoth, 2015-03-05 17:19:53]

https://codereview.chromium.org/983533003/diff/1/Makefile.standalone
File Makefile.standalone (right):

https://codereview.chromium.org/983533003/diff/1/Makefile.standalone#newcode6
Makefile.standalone:6: #        LIBCXX_INSTALL_PATH=<path> CLANG_PATH=<path> ...
Maybe add BINUTILS_BIN_PATH here for consistency?

https://codereview.chromium.org/983533003/diff/1/Makefile.standalone#newcode17
Makefile.standalone:17: # process. It should contain the tools like opt, llc and
clang. The default
Can you update the comment to reflect the actual set of tools?  This includes
llvm-as, pnacl-freeze, and llvm-mc (via lit.cfg and run-pnacl-sz.py), and
llvm-config and clang-format (via this makefile).

Interestingly, all of those except clang-format appear to be available in
BINUTILS_BIN_PATH, so maybe the latter could be used in the normal case? 
(Except that the two versions of llvm-config seem to produce slightly different
output, so...)

https://codereview.chromium.org/983533003/diff/1/Makefile.standalone#newcode32
Makefile.standalone:32: BINUTILS_BIN_PATH ?= $(shell readlink -e \
There's a pretty strong and fixed relationship between LIBCXX_INSTALL_PATH and
BINUTILS_BIN_PATH, right?  Could we drop one of them or combine into a new one?

[jvoung (off chromium), 2015-03-05 19:26:07]

https://codereview.chromium.org/983533003/diff/1/Makefile.standalone
File Makefile.standalone (right):

https://codereview.chromium.org/983533003/diff/1/Makefile.standalone#newcode6
Makefile.standalone:6: #        LIBCXX_INSTALL_PATH=<path> CLANG_PATH=<path> ...
On 2015/03/05 17:19:53, stichnot wrote:
> Maybe add BINUTILS_BIN_PATH here for consistency?

Done.

https://codereview.chromium.org/983533003/diff/1/Makefile.standalone#newcode17
Makefile.standalone:17: # process. It should contain the tools like opt, llc and
clang. The default
On 2015/03/05 17:19:53, stichnot wrote:
> Can you update the comment to reflect the actual set of tools?  This includes
> llvm-as, pnacl-freeze, and llvm-mc (via lit.cfg and run-pnacl-sz.py), and
> llvm-config and clang-format (via this makefile).

My main fear of listing all the tools in the comment, is that it
can go stale without breaking tests/notice, but okay.


> Interestingly, all of those except clang-format appear to be available in
> BINUTILS_BIN_PATH, so maybe the latter could be used in the normal case? 
> (Except that the two versions of llvm-config seem to produce slightly
different
> output, so...)

Yes, that bin path has almost everything. It would be nice if there's one bin
directory, but as you noticed there are problems with clang-format and
llvm-config. The llvm-config --ldflags says:


-L/usr/local/google/home/nacl3/native_client/toolchain/linux_x86/pnacl_newlib/lib

but that directory has pruned out the .a files since normal developers don't
really need them, so we can't use that directory. I guess one alternative is to
instead have an LLVM_LIBS_PATH to replace the `llvm-config --ldflags` output.

The clang-format issue is that toolchain_build prunes it. We could potentially
change that. I don't know if SDK users would start using that or not.

https://codereview.chromium.org/983533003/diff/1/Makefile.standalone#newcode32
Makefile.standalone:32: BINUTILS_BIN_PATH ?= $(shell readlink -e \
On 2015/03/05 17:19:53, stichnot wrote:
> There's a pretty strong and fixed relationship between LIBCXX_INSTALL_PATH and
> BINUTILS_BIN_PATH, right?  Could we drop one of them or combine into a new
one?

Added a common base, for default initialization. I don't know how necessary the
overrides are, so I kept them separate for now.

I noticed a couple similar paths in other scripts, so I changed build-runtime to
use this root instead. Didn't change crosstests or szbuild (szbuild I assume
would be run without the makefile).

[Jim Stichnoth, 2015-03-05 19:34:18]

LGTM, thanks!

https://codereview.chromium.org/983533003/diff/1/Makefile.standalone
File Makefile.standalone (right):

https://codereview.chromium.org/983533003/diff/1/Makefile.standalone#newcode17
Makefile.standalone:17: # process. It should contain the tools like opt, llc and
clang. The default
On 2015/03/05 19:26:06, jvoung wrote:
> On 2015/03/05 17:19:53, stichnot wrote:
> > Can you update the comment to reflect the actual set of tools?  This
includes
> > llvm-as, pnacl-freeze, and llvm-mc (via lit.cfg and run-pnacl-sz.py), and
> > llvm-config and clang-format (via this makefile).
> 
> My main fear of listing all the tools in the comment, is that it
> can go stale without breaking tests/notice, but okay.

Good point.  If you want to future-proof the test, that's fine, otherwise we can
future-proof it next time we notice it's gone stale. :)

https://codereview.chromium.org/983533003/diff/40001/Makefile.standalone
File Makefile.standalone (right):

https://codereview.chromium.org/983533003/diff/40001/Makefile.standalone#newc...
Makefile.standalone:21: # The default reflects a debug build with autotools
(configure & make).
Sorry, I forgot to note this before.  It's not actually a debug build, right? 
Describe it as a Release+Asserts build?  or just remove this line altogether?

[jvoung (off chromium), 2015-03-05 20:11:03]

https://codereview.chromium.org/983533003/diff/1/Makefile.standalone
File Makefile.standalone (right):

https://codereview.chromium.org/983533003/diff/1/Makefile.standalone#newcode17
Makefile.standalone:17: # process. It should contain the tools like opt, llc and
clang. The default
On 2015/03/05 19:34:18, stichnot wrote:
> On 2015/03/05 19:26:06, jvoung wrote:
> > On 2015/03/05 17:19:53, stichnot wrote:
> > > Can you update the comment to reflect the actual set of tools?  This
> includes
> > > llvm-as, pnacl-freeze, and llvm-mc (via lit.cfg and run-pnacl-sz.py), and
> > > llvm-config and clang-format (via this makefile).
> > 
> > My main fear of listing all the tools in the comment, is that it
> > can go stale without breaking tests/notice, but okay.
> 
> Good point.  If you want to future-proof the test, that's fine, otherwise we
can
> future-proof it next time we notice it's gone stale. :)

Hmm, next time =)

https://codereview.chromium.org/983533003/diff/40001/Makefile.standalone
File Makefile.standalone (right):

https://codereview.chromium.org/983533003/diff/40001/Makefile.standalone#newc...
Makefile.standalone:21: # The default reflects a debug build with autotools
(configure & make).
On 2015/03/05 19:34:18, stichnot wrote:
> Sorry, I forgot to note this before.  It's not actually a debug build, right? 
> Describe it as a Release+Asserts build?  or just remove this line altogether?

Ah right. I'll just leave the bit about autoconf and delete the part about
"debug", since I know the CMake build doesn't work. Besides the difference in
path (Release+Asserts/bin/ vs bin/), the CMake build uses libstdc++, while we
assume libc++.

I tried fixing the CMake build:
https://codereview.chromium.org/978963002/

but that failed... rpaths aren't right or something. The CMake build is much
more shared-library heavy. The tools don't actually run and llvm "check-all"
also breaks on some llvm-go bindings test trying to find shared libraries.

[jvoung (off chromium), 2015-03-05 22:33:44]

Committed patchset #4 (id:50001) manually as
68a06338e6cb4964212128169a69da52037874ce (presubmit successful).