Workaround #6986.

Workaround #6986.

This avoids trying to read a non-existent lockfile. On some
non-English locales, this fails when trying to parse the error
message coming from Windows.
BUG=6986
Committed: https://code.google.com/p/dart/source/detail?r=15992

[Bob Nystrom, 2012-12-11 18:31:59]

This should avoid the issue in almost all cases, and I think the code is still
pretty clean, so it doesn't seem like a bad change to me.

[Mads Ager (google), 2012-12-11 18:39:20]

LGTM

Thanks Bob! This looks cleaner to me too. I'll get the issue with localized
error messages fixed first thing tomorrow morning.

[nweiz, 2012-12-11 20:22:50]

Please add a comment indicating that this is a workaround of issue 6986 -- maybe
with a brief explanation of what that issue is, since it's a bit hard to tell by
just glancing at the bug.

[Bob Nystrom, 2012-12-11 20:35:17]

On 2012/12/11 20:22:50, nweiz wrote:
> Please add a comment indicating that this is a workaround of issue 6986 --
maybe
> with a brief explanation of what that issue is, since it's a bit hard to tell
by
> just glancing at the bug.

I actually think the code is a bit cleaner after the patch. We were still doing
a second exists() check in the original code, so the "try read and handle
failure" pattern doesn't actually avoid the potential race condition. Given
that, I think the new code is just fine to keep even after that bug is fixed.

[nweiz, 2012-12-11 20:39:35]

On 2012/12/11 20:35:17, Bob Nystrom wrote:
> On 2012/12/11 20:22:50, nweiz wrote:
> > Please add a comment indicating that this is a workaround of issue 6986 --
> maybe
> > with a brief explanation of what that issue is, since it's a bit hard to
tell
> by
> > just glancing at the bug.
> 
> I actually think the code is a bit cleaner after the patch. We were still
doing
> a second exists() check in the original code, so the "try read and handle
> failure" pattern doesn't actually avoid the potential race condition. Given
> that, I think the new code is just fine to keep even after that bug is fixed.

We should have been checking the osError field of the FileIOException to see if
it failed because the file didn't exist -- that would prevent the race condition
there.

This code also recovers less gracefully if a lockfile exists but is unreadable.