experience trying out lambda-8-b74

[Arne Siegel]

Hi lambda nation,

was wondering if I could lambda-ize this little file system crawler:

    public void forAllFilesDo(Block<File> fileConsumer, File baseFile) {
        if (baseFile.isHidden()) {
            // do nothing
        } else if (baseFile.isFile()) {
            fileConsumer.accept(baseFile);
        } else if (baseFile.isDirectory()) {
            File[] files = baseFile.listFiles();
            Arrays.stream(files).forEach(f -> forAllFilesDo(fileConsumer, f));
        }
    }

Using b74, the following reformulation produces the identical behaviour:

    public void forAllFilesDo3(Block<File> fileConsumer, File baseFile) {
        BiBlock<Stream.Downstream<File>,File> fileExploder = (ds, f) -> {
            if (f.isHidden()) {
                // do nothing
            } else if (f.isFile()) {
                ds.send(f);
            } else if (f.isDirectory()) {
                Arrays.stream(f.listFiles())
                      .forEach(fi -> fileExploder(ds, fi));   /* recursive call ! */
            }
        };
        Arrays.stream(new File[] {baseFile})
              .explode(fileExploder)
              .forEach(fileConsumer::accept);
    }

As my next step I was looking if I could parallelize the last step. But simply inserting .parallel() 
seems not to be working - presumably because the initial one-element Spliterator is 
governing split behaviour.

Inserting .collect(Collectors.<File>toList()).parallelStream() instead works fine:

        Arrays.stream(new File[] {baseFile})
              .explode(fileExploder)
              .collect(Collectors.<File>toList())
              .parallelStream()
              .forEach(fileConsumer::accept);

Main observations:

(*) Exploding a single element requires a lot of programming overhead:
1st step - creation of a one-element array or collection
2nd step - stream the result of the 1st step
3rd step - explode the result of the 2nd step
Not sure if anything can be done about this.

(*) .parallel() after .explode() behaves unexpectedly. Maybe the downstream should be based 
on a completely fresh unknown-size spliterator.

Regards
Arne Siegel