Split installing a package locally from symlinking to it.

Split installing a package locally from symlinking to it.

This is prep work for global packages, which need to ensure their
dependencies are local, but do not want to create symlinks for them.

R=nweiz@google.com

Committed: https://code.google.com/p/dart/source/detail?r=37672

[nweiz, 2014-06-24 19:57:19]

https://codereview.chromium.org/357483004/diff/1/sdk/lib/_internal/pub/lib/sr...
File sdk/lib/_internal/pub/lib/src/entrypoint.dart (right):

https://codereview.chromium.org/357483004/diff/1/sdk/lib/_internal/pub/lib/sr...
sdk/lib/_internal/pub/lib/src/entrypoint.dart:104:
source.createSymlink(resolved, packageDir);
The Future this returns needs to be hooked into the Future chain.

https://codereview.chromium.org/357483004/diff/1/sdk/lib/_internal/pub/lib/sr...
sdk/lib/_internal/pub/lib/src/entrypoint.dart:119: /// using the `package:`
scheme. Returns the resolved [PackageId].
I think this description has become confusing, since the "package:" scheme will
only work via pub build/serve/run after this runs.

https://codereview.chromium.org/357483004/diff/1/sdk/lib/_internal/pub/lib/sr...
File sdk/lib/_internal/pub/lib/src/source.dart (right):

https://codereview.chromium.org/357483004/diff/1/sdk/lib/_internal/pub/lib/sr...
sdk/lib/_internal/pub/lib/src/source.dart:107: Future createSymlink(PackageId
id, String symlink);
If this requires [ensureLocal] to be called first and [ensureLocal] is basically
a no-op if the package already exists locally, why doesn't this just call
[ensureLocal] itself?

https://codereview.chromium.org/357483004/diff/1/sdk/lib/_internal/pub/lib/sr...
File sdk/lib/_internal/pub/lib/src/source/git.dart (right):

https://codereview.chromium.org/357483004/diff/1/sdk/lib/_internal/pub/lib/sr...
sdk/lib/_internal/pub/lib/src/source/git.dart:221: return
git.runSync(["rev-parse", _getEffectiveRef(id)],
I feel a little weird about making this synchronous. I intentionally didn't do
so when I added sync support to git. Source stuff is all async anyway so we
don't get much of a code-quality benefit and this seems like the sort of thing
where parallelism could be valuable.

[Bob Nystrom, 2014-06-24 21:22:18]

Thanks!

https://codereview.chromium.org/357483004/diff/1/sdk/lib/_internal/pub/lib/sr...
File sdk/lib/_internal/pub/lib/src/entrypoint.dart (right):

https://codereview.chromium.org/357483004/diff/1/sdk/lib/_internal/pub/lib/sr...
sdk/lib/_internal/pub/lib/src/entrypoint.dart:104:
source.createSymlink(resolved, packageDir);
On 2014/06/24 19:57:18, nweiz wrote:
> The Future this returns needs to be hooked into the Future chain.

Done.

https://codereview.chromium.org/357483004/diff/1/sdk/lib/_internal/pub/lib/sr...
sdk/lib/_internal/pub/lib/src/entrypoint.dart:119: /// using the `package:`
scheme. Returns the resolved [PackageId].
On 2014/06/24 19:57:18, nweiz wrote:
> I think this description has become confusing, since the "package:" scheme
will
> only work via pub build/serve/run after this runs.

Done.

https://codereview.chromium.org/357483004/diff/1/sdk/lib/_internal/pub/lib/sr...
File sdk/lib/_internal/pub/lib/src/source.dart (right):

https://codereview.chromium.org/357483004/diff/1/sdk/lib/_internal/pub/lib/sr...
sdk/lib/_internal/pub/lib/src/source.dart:107: Future createSymlink(PackageId
id, String symlink);
On 2014/06/24 19:57:18, nweiz wrote:
> If this requires [ensureLocal] to be called first and [ensureLocal] is
basically
> a no-op if the package already exists locally, why doesn't this just call
> [ensureLocal] itself?

Done.

https://codereview.chromium.org/357483004/diff/1/sdk/lib/_internal/pub/lib/sr...
File sdk/lib/_internal/pub/lib/src/source/git.dart (right):

https://codereview.chromium.org/357483004/diff/1/sdk/lib/_internal/pub/lib/sr...
sdk/lib/_internal/pub/lib/src/source/git.dart:221: return
git.runSync(["rev-parse", _getEffectiveRef(id)],
On 2014/06/24 19:57:18, nweiz wrote:
> I feel a little weird about making this synchronous. I intentionally didn't do
> so when I added sync support to git. Source stuff is all async anyway so we
> don't get much of a code-quality benefit and this seems like the sort of thing
> where parallelism could be valuable.

Done.

I didn't want to make this sync, but doing so avoided a really bizarre bug I ran
into in dart:io. With my other changes, that doesn't seem to be needed anymore.
<shrug>

[nweiz, 2014-06-24 21:43:14]

lgtm

[Bob Nystrom, 2014-06-24 21:44:31]

Committed patchset #2 manually as r37672 (presubmit successful).