Rename is_chromeos in sqlite.

Rename is_chromeos in sqlite.

is_chromeos does a poor job of distinguishing developer builds of
chrome for chromeos for your desktop, and chrome for chromeos on
a device (where the runtimes are non local, for example)

Split this into is_chromeos_ui and is_chromeos_os.

BUG=519943
TEST=ChromiumOS compiles equivalently.

[Peter Mayo, 2015-08-13 01:21:26]

Dependent on https://codereview.chromium.org/1283313004/

upload suggested shess@ as a reviewer

[Scott Hess - ex-Googler, 2015-08-13 18:00:48]

shess@chromium.org changed reviewers:
+ shess@chromium.org

[Scott Hess - ex-Googler, 2015-08-13 18:00:49]

https://codereview.chromium.org/1286273004/diff/1/third_party/sqlite/BUILD.gn
File third_party/sqlite/BUILD.gn (right):

https://codereview.chromium.org/1286273004/diff/1/third_party/sqlite/BUILD.gn...
third_party/sqlite/BUILD.gn:34: if (is_chromeos_ui) {
I read the other CL, and I don't entirely understand this.  The following define
is predicated on running on the kind of filesystem the underlying ChromeOS
system provides, it has nothing to do with the UI layer.  So I would think this
should be is_chromeos_os.

If the goal is to be able to test a greater variety of things in more generic
environment, then maybe you need another flag to handle that, rather than
constraining by a goal-type constraint.

[Peter Mayo, 2015-08-13 18:20:00]

Address review optimization

[Peter Mayo, 2015-08-13 18:20:35]

https://codereview.chromium.org/1286273004/diff/1/third_party/sqlite/BUILD.gn
File third_party/sqlite/BUILD.gn (right):

https://codereview.chromium.org/1286273004/diff/1/third_party/sqlite/BUILD.gn...
third_party/sqlite/BUILD.gn:34: if (is_chromeos_ui) {
On 2015/08/13 18:00:49, Scott Hess wrote:
> I read the other CL, and I don't entirely understand this.  The following
define
> is predicated on running on the kind of filesystem the underlying ChromeOS
> system provides, it has nothing to do with the UI layer.  So I would think
this
> should be is_chromeos_os.

I applaud the correction.  Patch coming.

> 
> If the goal is to be able to test a greater variety of things in more generic
> environment, then maybe you need another flag to handle that, rather than
> constraining by a goal-type constraint.

At some point we start with goals, get mapped through build systems to flags,
and then through source code to implementations.  How many of the intermediate
steps get labels is a matter of judgement.  If the owners of SQLite want to add
an is_journalled_file_system flag here, whose default is set so that chromeos
enables it where most others don't,  I'd be happy to write or review it in a
separate CL.

The actual point is in trying to make the compiles work correctly on both the
devices and developers I ran into the problem that there was no separation in
the is_chromeos flag.  I other words: today this is set for
non-journalled-filesystems too when compiling to test for chromeos.  That could
int heory be a contributor to test flake.

[Peter Mayo, 2015-08-13 18:21:07]

https://codereview.chromium.org/1286273004/diff/1/third_party/sqlite/BUILD.gn
File third_party/sqlite/BUILD.gn (right):

https://codereview.chromium.org/1286273004/diff/1/third_party/sqlite/BUILD.gn...
third_party/sqlite/BUILD.gn:34: if (is_chromeos_ui) {
On 2015/08/13 18:00:49, Scott Hess wrote:
> I read the other CL, and I don't entirely understand this.  The following
define
> is predicated on running on the kind of filesystem the underlying ChromeOS
> system provides, it has nothing to do with the UI layer.  So I would think
this
> should be is_chromeos_os.
> 
> If the goal is to be able to test a greater variety of things in more generic
> environment, then maybe you need another flag to handle that, rather than
> constraining by a goal-type constraint.

Done.

[Scott Hess - ex-Googler, 2015-08-13 20:55:19]

lgtm

https://codereview.chromium.org/1286273004/diff/1/third_party/sqlite/BUILD.gn
File third_party/sqlite/BUILD.gn (right):

https://codereview.chromium.org/1286273004/diff/1/third_party/sqlite/BUILD.gn...
third_party/sqlite/BUILD.gn:34: if (is_chromeos_ui) {
On 2015/08/13 18:20:35, Peter Mayo wrote:
> At some point we start with goals, get mapped through build systems to flags,
> and then through source code to implementations.  How many of the intermediate
> steps get labels is a matter of judgement.  If the owners of SQLite want to
add
> an is_journalled_file_system flag here, whose default is set so that chromeos
> enables it where most others don't,  I'd be happy to write or review it in a
> separate CL.

I don't think that would make sense, because the comment below isn't true.  Just
because the filesystem is journaled doesn't mean that it is providing the same
guarantees that SQLite is attempting to maintain using fsync.  As best I
understand things, ext4 journaling with data=journal _may_ provide those
guarantees, but data=ordered doesn't, but whenever I try to dig deeply I find
myself tangled in special cases which make even that uncertain.

[Scott Hess - ex-Googler, 2015-08-17 20:04:38]

On 2015/08/13 20:55:19, Scott Hess wrote:
> lgtm
> 
> https://codereview.chromium.org/1286273004/diff/1/third_party/sqlite/BUILD.gn
> File third_party/sqlite/BUILD.gn (right):
> 
>
https://codereview.chromium.org/1286273004/diff/1/third_party/sqlite/BUILD.gn...
> third_party/sqlite/BUILD.gn:34: if (is_chromeos_ui) {
> On 2015/08/13 18:20:35, Peter Mayo wrote:
> > At some point we start with goals, get mapped through build systems to
flags,
> > and then through source code to implementations.  How many of the
intermediate
> > steps get labels is a matter of judgement.  If the owners of SQLite want to
> add
> > an is_journalled_file_system flag here, whose default is set so that
chromeos
> > enables it where most others don't,  I'd be happy to write or review it in a
> > separate CL.
> 
> I don't think that would make sense, because the comment below isn't true. 
Just
> because the filesystem is journaled doesn't mean that it is providing the same
> guarantees that SQLite is attempting to maintain using fsync.  As best I
> understand things, ext4 journaling with data=journal _may_ provide those
> guarantees, but data=ordered doesn't, but whenever I try to dig deeply I find
> myself tangled in special cases which make even that uncertain.

To be clear, the above wasn't a comment intended to drive work right now in this
CL.  I just don't want to add build-config code implying that anyone has proven
that using SQLITE_NO_SYNC works right in SQLite on a journaled filesystem. 
SQLite doesn't make that guarantee, so I'm certainly not willing to.