Now instead of just saying how many packages failed to be
reinstalled/reactivated, "pub cache repair" will print a list of those
packages.

This also fixes an issue where "pub cache repair" would incorrectly
return a 0 exit code when packages failed to reactivate.

Closes #1301

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1239623005 .

This is helpful for subprocesses to be able to spawn Dart with
Platform.executableArguments and have that work regardless of the
working directory.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1236973002 .

If a package used two instances of its own transformer on the
transformer's own file (or a file it imported), the dependency computer
would crash.

Closes #1298

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1229303004 .

This causes a substantial amount of pain, and doesn't seem to really
help anyone. It's rare to have the connection with the server go
completely dead without actually closing.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1235013002 .

This will help avoid issues like #1294 in the future.

R=rnystrom@google.com, rnystrom@gmail.com

Review URL: https://codereview.chromium.org//1227913009 .

* forward_signal_posix_test needed to be marked as incompatible with
  Windows.

* Deleting the sandbox directory would sometimes fail due to locking
  issues, and this failure would obscure underlying issues.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1226083002 .

Closes #1296

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1233533002 .

* Normalize paths before they're passed to dart2js. It handles "." and
  ".." path components poorly.

* Ensure that paths emitted by list-package-dirs are always absolute and
  normalized for backwards-compatibility with the existing protocol.

* Make Package.listFiles work for an in-memory package.

* Fix an error message expectation in packages_file_test.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1228683003 .

It can choke on relative paths in surprising ways.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1231893004 .

* The package spec's reference to the entrypoint package was pointing to
  the hosted cache rather than the package's directory. This was caused
  by the package graph's instance of the entrypoint package incorrectly
  pointing to the cache; the graph now re-uses `entrypoint.root`.

* The package spec didn't contain relative paths for relative path
  dependencies. This was being missed by our tests because the package
  spec parser was automatically converting all paths to absolute at
  parse-time.

Closes #1294

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1228093003 .

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1224153002 .

* Some HTTP requests were still being made. These are now avoided, and
  the tests have been modified to assert that no HTTP requests are made
  at all.

* When a specific package version was in a package's lockfile *and*
  available on the server but wasn't in the user's cache, pub would
  crash when run in offline mode.

Closes #1293

Print a nice error for missing offline packages.

Previously, this would print an unhelpful message along with "this is an
unexpected error".

Closes #1293

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1226133003 .

OS X symlinks its /tmp directory, so we need to resolve symlinks before
we can safely compare paths that might include /tmp.

Review URL: https://codereview.chromium.org//1225483003.

Update README.md

When a package both transformed itself and had a dev transformer (that
is, a transformer that only runs on non-public files), *and* that
package's self-transformed code was imported by another transformer, the
dependency computer could return the dev transformer in its results.

Closes #1290
Closes #1291

R=kevmoo@google.com

Review URL: https://codereview.chromium.org//1220223008.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1217943009.

Somehow I didn't notice that the test I added for this wasn't actually
passing.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1211243009.

I forgot to update the --help output here.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1220543006.

Now that pub uses the new test runner, some of its logic for running in
the Dart SDK repository needs to be updated.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1216163002.

When running an executable directly from the filesystem, "pub run" would
sometimes fail to give it the correct package root.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1217923002.

Now that users in general have a reason to disable this, they should
know it exists.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1217013004.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1212843004.

Now that Platform.resolvedExecutable exists, the environment variable
only provided a way to load files from the wrong SDK.

Closes #1288

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1220553002.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1215833002.

The git invocations in the snapshot-generation code became very flaky
when run in massive parallel via the new test runner. Now a snapshot
will be used if it's manually-generated, and otherwise pub will be run
from source.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1216433003.

Fix for pub crashing in checked mode

R=nweiz@google.com

Review URL: https://codereview.chromium.org//1096723002.

As per #1283, this makes it easier to script pub, since it's now
programattically clear whether publishing went through (or, if using
--dry-run, whether validation succeeded).

Closes #1283

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1172803002.

We moved this location again to get custom pub testing

R=nweiz@google.com
BUG=

Review URL: https://codereview.chromium.org//1164203003

If all packages depended on by an executable don't use transformers,
this will avoid loading barback entirely and instead run the VM against
the Dart files on the filesystem.

This shaves about 600ms off the load time for "pub run test" when running from
source.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1166343002.

These tests broke when removing compatibility with the old "pub run"
behavior.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1171783002

Closes #1235.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1161613012

When I originally wrote this code, I hadn't thought about running tests
from within the Dart repo itself, so the snapshotting logic didn't
work for that case.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1166783002

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1166623003

This should fix the analyzer buildbot (see http://build.chromium.org/p/client.dart/builders/analyzer_experimental-linux-release-be/builds/1826/steps/tests%20--host-checked%20--checked/logs/stdio)

R=nweiz@google.com

Review URL: https://codereview.chromium.org//1160013003

These issues were hidden last time I was testing because I couldn't
actually bring in pub via DEPS.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1162603009

I hadn't considered the case of running tests from within the Dart repo,
but now the package-loading and SDK-detection logic works in that
scenario as well.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1153733003