gracefully degrade if package map input file cannot be watched

gracefully degrade if package map input file cannot be watched

BUG=

Committed: https://github.com/dart-lang/sdk/commit/679de372f43b7afd7ca7a25169b1c53842581596

[danrubel, 2015-06-22 21:07:17]

Committed patchset #3 (id:40001) manually as
679de372f43b7afd7ca7a25169b1c53842581596 (presubmit successful).

[danrubel, 2015-06-22 21:08:04]

danrubel@google.com changed reviewers:
+ paulberry@google.com

[Paul Berry, 2015-06-22 21:19:31]

https://codereview.chromium.org/1198423002/diff/40001/pkg/analysis_server/lib...
File pkg/analysis_server/lib/src/context_manager.dart (right):

https://codereview.chromium.org/1198423002/diff/40001/pkg/analysis_server/lib...
pkg/analysis_server/lib/src/context_manager.dart:441: if (resource is File &&
resource.exists) {
This introduces a race condition, because it's possible that the file will be
deleted between the call to .exists here and the existence check that presumably
happens inside the watcher.

It would be better to simply handle the error condition from the file watcher. 
Not only would that eliminate the race condition, it would avoid performing a
redundant existence check in the common scenario where we *aren't* racing with
some other process.

[danrubel, 2015-06-22 21:54:46]

https://codereview.chromium.org/1198423002/diff/40001/pkg/analysis_server/lib...
File pkg/analysis_server/lib/src/context_manager.dart (right):

https://codereview.chromium.org/1198423002/diff/40001/pkg/analysis_server/lib...
pkg/analysis_server/lib/src/context_manager.dart:441: if (resource is File &&
resource.exists) {
On 2015/06/22 21:19:31, Paul Berry wrote:
> This introduces a race condition, because it's possible that the file will be
> deleted between the call to .exists here and the existence check that
presumably
> happens inside the watcher.
> 
> It would be better to simply handle the error condition from the file watcher.

> Not only would that eliminate the race condition, it would avoid performing a
> redundant existence check in the common scenario where we *aren't* racing with
> some other process.

I tried doing that in an earlier iteration of this CL, but could not convince
myself that it would properly handle file system errors without masking problems
in _recomputePackageUriResolver. Based on our conversation I reworked this in
another CL : https://codereview.chromium.org/1199083002