Make the update installer fail when attempting to install a 64-bit update on a 32-bit-only system

Make the update installer fail when attempting to install a 64-bit-only update
on a 32-bit-only system.

The server infrastructure should not deliver 64-bit-only updates to
32-bit-only systems, but this depends on the new -32bit tag suffix support
already being on the system prior to the attempted delivery of a 64-bit-only
update. The update installer will now sanity-check this, and if the system
will not be capable of running the updated version, it will exit before
attempting to install it. Additionally, it will set the -32bit tag suffix so
that during a subsequent update check, the server should deliver the proper
32-bit-only update (likely the final 32-bit version), or no update at all.

This also serves as a failsafe against breaking users on 32-bit-only systems
if the server-side -32bit tag suffix detection isn't working properly.

BUG=18323, 303280
R=thakis@chromium.org, thakis

Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=243088

[Mark Mentovai, 2014-01-06 01:44:54]

This addresses your use case of “I didn’t open that old laptop in 3 months.”

[Nico, 2014-01-06 05:32:13]

lgtm

Once this is in, you will want to work with vclarke to get this case tested by
qa.

https://codereview.chromium.org/124813002/diff/30001/chrome/installer/mac/key...
File chrome/installer/mac/keystone_install.sh (right):

https://codereview.chromium.org/124813002/diff/30001/chrome/installer/mac/key...
chrome/installer/mac/keystone_install.sh:34: # 14  The update is incompatible
with the system
Since that other patch mentioned something in chrome checks these update codes:
will this exit code 14 be handled by old chromes?

[Mark Mentovai, 2014-01-06 14:34:58]

Nothing in Chrome checks these codes. They only show up in diagnostic logs.

The 12 I mentioned in another patch is a different 12. That one is
kAutoupdateInstallFailed from chrome/browser/mac/keystone_glue.h. It’s very
generic.

There’s an open Keystone bug about plumbing the script’s exit status through
KeystoneRegistration to Chrome so that it can provide more meaningful error
messages. No traction in several years, though.

[Mark Mentovai, 2014-01-06 14:35:35]

I’ve already talked with Victoria. She doesn’t have access to any 32-bit-only
Macs. I do, so I have to do the testing myself…

[Mark Mentovai, 2014-01-06 14:37:02]

Committed patchset #1 manually as r243088 (presubmit successful).