Java Stream: is there a way to iterate taking two elements a time instead of one?

Things would be easier if your input is located in the random-access list. This way you can utilize good old List.subList method like this:

List<String> list = Arrays.asList("a", "b", "err1", "c", "d", "err2", "e", 
     "f", "g", "h", "err3", "i", "j");

Map<String, String> map = IntStream.range(0, list.size()-1)
    .mapToObj(i -> list.subList(i, i+2))
    .filter(l -> l.get(0).startsWith("err"))
    .collect(Collectors.toMap(l -> l.get(0), l -> l.get(1)));

The same thing could be done with already mentioned StreamEx library (written by me) in a little bit shorter manner:

List<String> list = Arrays.asList("a", "b", "err1", "c", "d", "err2", "e", 
     "f", "g", "h", "err3", "i", "j");

Map<String, String> map = StreamEx.ofSubLists(list, 2, 1)
    .mapToEntry(l -> l.get(0), l -> l.get(1))
    .filterKeys(key -> key.startsWith("err"))
    .toMap();

Though if you don't want third-party dependency, the poor Stream API solution looks also not very bad.

You can write a custom collector, or use the much simpler approach of streaming over the list's indexes:

Map<String, String> result = IntStream.range(0, data.size() - 1)
        .filter(i -> data.get(i).startsWith("err"))
        .boxed()
        .collect(toMap(data::get, i -> data.get(i+1)));

This assumes that your data is in a random access friendly list or that you can temporarily dump it into one.

If you cannot randomly access the data or load it into a list or array for processing, you can always make a custom pairing collector so you can write

Map<String, String> result = data.stream()
        .collect(pairing(
                (a, b) -> a.startsWith("err"), 
                AbstractMap.SimpleImmutableEntry::new,
                toMap(Map.Entry::getKey, Map.Entry::getValue)
        ));

Here's the source for the collector. It's parallel-friendly and might come in handy in other situations:

public static <T, V, A, R> Collector<T, ?, R> pairing(BiPredicate<T, T> filter, BiFunction<T, T, V> map, Collector<? super V, A, R> downstream) {

    class Pairing {
        T left, right;
        A middle = downstream.supplier().get();
        boolean empty = true;

        void add(T t) {
            if (empty) {
                left = t;
                empty = false;
            } else if (filter.test(right, t)) {
                downstream.accumulator().accept(middle, map.apply(right, t));
            }
            right = t;
        }

        Pairing combine(Pairing other) {
            if (!other.empty) {
                this.add(other.left);
                this.middle = downstream.combiner().apply(this.middle, other.middle);
                this.right = other.right;
            }
            return this;
        }

        R finish() {
            return downstream.finisher().apply(middle);
        }
    }

    return Collector.of(Pairing::new, Pairing::add, Pairing::combine, Pairing::finish);
}

You can build a custom Collector for this task.

Map<String, String> map = 
    Stream.of("a", "b", "err1", "c", "d", "err2", "e", "f", "g", "h", "err3", "i", "j")
          .collect(MappingErrors.collector());

with:

private static final class MappingErrors {

    private Map<String, String> map = new HashMap<>();

    private String first, second;

    public void accept(String str) {
        first = second;
        second = str;
        if (first != null && first.startsWith("err")) {
            map.put(first, second);
        }
    }

    public MappingErrors combine(MappingErrors other) {
        throw new UnsupportedOperationException("Parallel Stream not supported");
    }

    public Map<String, String> finish() {
        return map;
    }

    public static Collector<String, ?, Map<String, String>> collector() {
        return Collector.of(MappingErrors::new, MappingErrors::accept, MappingErrors::combine, MappingErrors::finish);
    }

}

In this collector, two running elements are kept. Each time a String is accepted, they are updated and if the first starts with "err", the two elements are added to a map.

---

Another solution is to use the StreamEx library which provides a pairMap method that applies a given function to the every adjacent pair of elements of this stream. In the following code, the operation returns a String array consisting of the first and second element of the pair if the first element starts with "err", null otherwise. null elements are then filtered out and the Stream is collected into a map.

Map<String, String> map = 
    StreamEx.of("a", "b", "err1", "c", "d", "err2", "e", "f", "g", "h", "err3", "i", "j")
            .pairMap((s1, s2) -> s1.startsWith("err") ? new String[] { s1, s2 } : null)
            .nonNull()
            .toMap(a -> a[0], a -> a[1]);

System.out.println(map);