java.nio.FileSystem support #859
Conversation
It just skips the build and uses the cache, and creates a new tag for the same image.
This is to work around the server closing the connections.
This allows the FileSystem to be set-up, ahead of time using newFilesystem, then to be used through Paths#get(URI) without a reference to the Filesystem itself.
This is instead of a mock.
This is needed to support an empty non-root SmbPath.
And fix a couple of small issues with said integration around directory creation.
|
@dgodfrey I think it would help if you squash all commit into one here. @hierynomus any chance to make progress here? I also using a private fork of this code since a while and like to see this in the main repo and contribute some additional fixes and enhancements to this. |
|
I don't really like squash commits, as it doesn't tell the "story", although I suppose it won't make too much difference here as 95% of the files are new. I don't really know Github - is it possible to squash at merge time? |
|
@dgodfrey yes you can do it in Github UI, while I agree that the history is often valuable for a PR it is often better to have a single commit (or maybe two) maybe done in a separate branch as it is easier to rebase if the repository code changes. In any case @hierynomus has of course to tell what is the best here, it was just a thought as I see 30 commits are here for a single feature :-) |
| @@ -195,9 +195,10 @@ public void read(OutputStream destStream, ProgressListener progressListener) thr | |||
| * @param fileOffset The offset, in bytes, into the file to which the data should be written | |||
| * @return the actual number of bytes that was written to the file | |||
| */ | |||
| public long write(ByteBuffer buffer, long fileOffset) { | |||
| public int write(ByteBuffer buffer, long fileOffset) { | |||
There was a problem hiding this comment.
I think this should not be truncated here, if the caller wants an int it is better cast there.
There was a problem hiding this comment.
I think I changed them as the maximum size of a ByteBuffer is an int, so narrowed it here to match the read version and to do it once, rather than everybody who calls the method having to do it. But pretty easy either int or long.
| @@ -208,7 +209,7 @@ public long write(ByteBuffer buffer, long fileOffset) { | |||
| * @param fileOffset The offset, in bytes, into the file from which the data should be read | |||
| * @return the actual number of bytes that were read; or -1 if the end of the file was reached | |||
| */ | |||
| public long read(ByteBuffer buffer, long fileOffset) { | |||
| public int read(ByteBuffer buffer, long fileOffset) { | |||
There was a problem hiding this comment.
Same here better use a long as long as possible.
There was a problem hiding this comment.
This was getting widened from an int to a long here, then any callers were needing to narrow it back to an int - this just removes that need. Pretty much the whole IO library uses int's for reading/writing bytes (I guess) because that's the maximum size of an array. But again, pretty easy on int or long.
This mainly adds support for reading from SMB shares using a FileSystem implementation. It use the URI format from jcifs, to ease migrations.
It was developed on Windows, so hopefully is ok on Linux, and includes a few "fixes" to SmbStubServer for Windows and a change so it only rebuilds the docker image used for integration testing once, rather than every times it's used.