This folds the distinction between "resolved" and "unresolved"
descriptions into the distinction between PackageIds and PackageRefs. In
other words, all PackageIds now have resolved descriptions and all
PackageRefs now have unresolved descriptions. This makes it easier to
track which descriptions are resolved and express resolution constraints
via the type system.

This also restricts the creation of PackageRefs and PackageIds to source
classes, which helps ensure that the description format—which is an
implementation detail—doesn't leak into surrounding code.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1528523003 .

Previously these names were only disallowed when publishing, but now the
package config format requires that we be more stringent about them.

Closes #1361

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1493463002 .

This allows getVersion() to return a list of package IDs rather than a
list of pubspecs, which in turn will allow the Git source to return
resolved IDs in #1335.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1459733002 .

Previously, "pub global run" would happily try to run an executable with
an SDK constraint that doesn't match the current SDK, even though it
probably wouldn't work (and could fail in confusing ways). Now it errors
out with a user-friendly message instead.

Closes #1353

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1427523005 .

Closes #1351

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1406063005 .

When we tried to run a global script that didn't exist, we would try to
load an AssetEnvironment, which would try to compute the transformers
for it. Because the script doesn't exist, the computation would
conservatively assume that all transformers had to run in case one of
them generated it.

To find all the scripts, it iterated through the package's dependencies.
Since it's notionally the entrypoint, it tried to include its dev
dependencies, but this broke because dev dependencies aren't installed
for globally-activated packages.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1413713010 .

We didn't properly check for no sources in a dependency, which caused a
crash for some invalid pubspecs.

Closes #1348

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1417993002 .

This is useful for visualizing what transitive dependencies a package
will add for its users. This is especially useful now that the test
runner has so many dependencies.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1409043004 .

This includes several fixes:

* It upgrades to the latest compiler_unsupported which uses
  dart:developer rather than dart:profiler,

* It fixes the test runner detection to match the new spawning
  mechanism.

* It works around dart-lang/sdk#24653 in a dart2js test.

* It upgrades an invocation of `new oauth2.Credentials()` to match the
  new oauth2 API.

Closes #1346

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1408423006 .

We used to do this, but we stopped without really meaning to when the
dart:io default changed.

Closes #1332

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1340043004 .

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1344633003 .

Instead of relying only on modification times, this does a more complex
compatibility check if the modification times of the pubspec, lockfile,
and .packages file aren't ordered properly. If they are up-to-date, it
updates the modification times accordingly.

This also fixes a case where "pub run" wasn't properly checking that the
dependencies were up-to-date.

Closes #1322

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1307853004 .

If the root directory of a package is .gitignored, it will no longer be
considered to be a "git package", since that just means that it'll be
seen to contain no files at all.

Closes #1321

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1317043002 .

Pub won't upload gitignored files, so its validation shouldn't consider
them to exist either.

See #1321

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1310253007 .

Closes #1319

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1293383006 .

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1295803003 .

We were assuming that all globally-activated packages have a .packages
file, but this isn't true for packages that were activated by earlier
versions of pub.

Closes #1316

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1292003005 .

Since there's no way to load resources without using the packages/
directory, this flag will introduce a lot of confusing unfixable
breakage. We should un-hide it once there's sufficient SDK support for
locating/loading resources (dart-lang/sdk#23990).

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1292933003 .

Previously, if a dependency depended on the entrypoint package, the
entrypoint would have all its direct dependencies re-printed. This fixes
that.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1274763010 .

Previously, several commands automatically ran "pub get" implicitly before doing
their work. We've determined that this behavior is confusing, and for "pub run"
in particular, makes the executables seem much slower than they actually are.

This also allows us to make access to the entrypoint's package graph
synchronous.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1282533003 .

The spec doesn't actually need all the data loaded from the package
graph; all it needs are the locations of packages which can be looked up
using their IDs and a SourceRegistry.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1282083002 .

This also has it store its SourceRegistry, so that various methods don't
need to re-supply it.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1281043004 .

It never really made sense for this to be asynchronous; most sources
didn't even do any IO to determine the directory. The only exception was
the git source, which would actually *clone a repository* if one didn't
exist yet. This heavyweight behavior wasn't a good fit for what looks
like a lightweight method anyway.

Making this synchronous has a ripple effect of making a bunch of other
methods synchronous, too. Hooray for synchrony!

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1276673006 .

Closes pub#22

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1277773002 .

Fixes #1230.

R=nweiz@google.com

Review URL: https://codereview.chromium.org//1272813003 .

This reverts commit 7fbc1040, which
reverted commit dc626af3.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1272903002 .

Currently we don't auto-run "pub get" if there's no lockfile and also no
dependencies. We should, in order to create a self-link.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1277633004 .

Makes pub serve less spammy when used with Dartium where you have a
slew of requests for each source file.

Review URL: https://codereview.chromium.org//1267023002 .

This reverts commit dc626af3.

This broke due to dart-lang/sdk#23991.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1272833003 .

Previously this was using the packages/ directory, but that's no longer
reliable since there's now an option to not generate it.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1269393003 .

Fixes #1109.

R=nweiz@google.com

Review URL: https://codereview.chromium.org//1267853002 .

Closes #1304

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1248483003 .

It turns out it's not safe to assume that pub is checked out from Git
when running from the SDK repo.

R=rnystrom@google.com

Review URL: https://codereview.chromium.org//1246713005 .

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 .

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 .