Check in a binary of gold from binutils 2.21.1.

Check in a binary of gold from binutils 2.21.1.

This will be used by the Chrome build to make gold the default.
Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=119932

[Evan Martin, 2012-01-30 23:32:45]

(note the base URL on the left -- this is under trunk/deps so I'll need to pull
it in with DEPS in a follow-up)

[Lei Zhang, 2012-01-30 23:42:55]

You should also check in a copy of ld.bfd in the same directory to make NaCl
happy. I think you weren't here when that issue came up.

[tony, 2012-01-30 23:48:58]

On 2012/01/30 23:42:55, Lei Zhang wrote:
> You should also check in a copy of ld.bfd in the same directory to make NaCl
> happy. I think you weren't here when that issue came up.

Specifically, src/native_client/src/trusted/service_runtime/linux/ld_bfd.py
appears to look for a file named ld.bfd.  Maybe we can fix that instead.

[Evan Martin, 2012-01-30 23:50:42]

On 2012/01/30 23:48:58, tony wrote:
> On 2012/01/30 23:42:55, Lei Zhang wrote:
> > You should also check in a copy of ld.bfd in the same directory to make NaCl
> > happy. I think you weren't here when that issue came up.
> 
> Specifically, src/native_client/src/trusted/service_runtime/linux/ld_bfd.py
> appears to look for a file named ld.bfd.  Maybe we can fix that instead.

I think I'm ok with them falling back on the system ld, using the mechanism they
currently do.  Is that not ok?

[Lei Zhang, 2012-01-30 23:55:12]

I can't remember all the details exactly, but in my head I have this nothing
that we need to put a ld.bfd next to ld (.gold)

Bradchen probably knows what the deal is.

[tony, 2012-01-30 23:58:00]

On 2012/01/30 23:55:12, Lei Zhang wrote:
> I can't remember all the details exactly, but in my head I have this nothing
> that we need to put a ld.bfd next to ld (.gold)
> 
> Bradchen probably knows what the deal is.

Looking at the script in detail, it looks like it falls back on g++
-print-prog-name=ld, which should be safe.

LGTM.

Brad, feel free to chime in if you think this will break the build.

[Evan Martin, 2012-01-31 00:03:03]

http://code.google.com/codesearch#OAMlx_jo-ck/src/native_client/src/trusted/s...

[Evan Martin, 2012-01-31 00:09:14]

Having read this three times, I think I convinced myself it'll be fine as long
as there exists an ld.bfd "somewhere"

[Lei Zhang, 2012-01-31 00:18:41]

On 2012/01/31 00:09:14, Evan Martin wrote:
> Having read this three times, I think I convinced myself it'll be fine as long
> as there exists an ld.bfd "somewhere"

So the way I read it, is that the script figures out where 'ld' is first and
append .bfd to it. E.g. if ld is /path/to/ld, then it looks for /path/to/ld.bfd.
If /path/to/ld.bfd DNE, then it uses /path/to/ld.

IIRC, there's issues with NaCl + gold, thus they don't want to fall back to
/path/to/ld, where ld is gold.

[Evan Martin, 2012-01-31 00:23:18]

Code looks like:

  ld = GccPrintName('ld', '-print-prog-name=ld', 'ld')
# ld == "ld", using all variations we could find

  ld_bfd = PathTo(ld + ".bfd")
# this now searches the path for the string "ld.bfd"

So it constructs the string ld.bfd first, then searches the path for that.

[Lei Zhang, 2012-01-31 00:32:59]

On 2012/01/31 00:23:18, Evan Martin wrote:
> Code looks like:
> 
>   ld = GccPrintName('ld', '-print-prog-name=ld', 'ld')
> # ld == "ld", using all variations we could find
> 
>   ld_bfd = PathTo(ld + ".bfd")
> # this now searches the path for the string "ld.bfd"
> 
> So it constructs the string ld.bfd first, then searches the path for that.

This would depend on how you pass -B to g++. If you are like me and you use g++
-B/home/thestig/bin, then PathTo() would look for /home/thestig/bin/ld.bfd. In
which case, PathTo() will join some path with /home/thestig/bin/ld.bfd and
return /home/thestig/bin/ld.bfd. If /home/thestig/bin/ld.bfd DNE, then it will
fail to find /home/thestig/bin/ld.bfd in the path all together, and still return
/home/thestig/bin/ld.bfd.

[Evan Martin, 2012-01-31 17:30:52]

Ah, I understand now.  Thanks for explaining.

What do you think of me fixing this ld_bfd.py script to just run "ld.bfd" out of
your $PATH?

(I am reluctant to follow your advice to just check in an ld.bfd because I don't
feel like it's my place to a replacement for the "system" linker when I'm just
trying to provide gold.)

[Evan Martin, 2012-01-31 17:44:20]

Another idea: check in a shell script called ld.bfd that just runs
/usr/bin/ld.bfd and falls back to /usr/bin/ld.

[Evan Martin, 2012-01-31 18:42:34]

Brad, can you help?  My long review round trips with Lei are killing me.  :)

[Evan Martin, 2012-01-31 18:48:59]

On 2012/01/31 18:42:34, Evan Martin wrote:
> Brad, can you help?  My long review round trips with Lei are killing me.  :)

https://chromiumcodereview.appspot.com/9316002 as well

[bradn, 2012-01-31 18:57:25]

Otherwise LGTM.

https://chromiumcodereview.appspot.com/9302017/diff/11001/README.chromium
File README.chromium (right):

https://chromiumcodereview.appspot.com/9302017/diff/11001/README.chromium#new...
README.chromium:3: Version: 2.21.1
Pawel's license script is likely to scream if this doesn't have a license.

https://chromiumcodereview.appspot.com/9302017/diff/11001/README.chromium#new...
README.chromium:9: 
Trailing CR

https://chromiumcodereview.appspot.com/9302017/diff/11001/ld.bfd
File ld.bfd (right):

https://chromiumcodereview.appspot.com/9302017/diff/11001/ld.bfd#newcode3
ld.bfd:3: # This script forwards requests for a program named "ld.bfd", to
NaCl

[Evan Martin, 2012-01-31 19:04:37]

https://chromiumcodereview.appspot.com/9302017/diff/11001/README.chromium
File README.chromium (right):

https://chromiumcodereview.appspot.com/9302017/diff/11001/README.chromium#new...
README.chromium:3: Version: 2.21.1
On 2012/01/31 18:57:26, bradn wrote:
> Pawel's license script is likely to scream if this doesn't have a license.

(I fixed this in the other change, to filter this out of the licenses scan.)

[Lei Zhang, 2012-01-31 19:48:11]

Not a morning person. :)
This LGTM.