/// <summary>
/// Partitions the items consumed by the batchable object into buckets, where each bucket contains a set of items that
/// have the same value set on all item metadata consumed by the object.
/// </summary>
/// <remarks>
/// PERF NOTE: Given n items and m batching metadata that produce l buckets, it is usually the case that n > l > m,
/// because a batchable object typically uses one or two item metadata to control batching, and only has a handful of
/// buckets. The number of buckets is typically only large if a batchable object is using single-item batching
/// (where l == n). Any algorithm devised for bucketing therefore, should try to minimize n and l in its complexity
/// equation. The algorithm below has a complexity of O(n*lg(l)*m/2) in its comparisons, and is effectively O(n) when
/// l is small, and O(n*lg(n)) in the worst case as l -> n. However, note that the comparison complexity is not the
/// same as the operational complexity for this algorithm. The operational complexity of this algorithm is actually
/// O(n*m + n*lg(l)*m/2 + n*l/2 + n + l), which is effectively O(n^2) in the worst case. The additional complexity comes
/// from the array and metadata operations that are performed. However, those operations are extremely cheap compared
/// to the comparison operations, which dominate the time spent in this method.
/// </remarks>
/// <returns>ArrayList containing ItemBucket objects (can be empty), each one representing an execution batch.</returns>
private static ArrayList BucketConsumedItems
(
XmlNode parentNode,
Lookup lookup,
Hashtable itemListsToBeBatched,
Dictionary<string, MetadataReference> consumedMetadataReferences
)
{
ErrorUtilities.VerifyThrow(itemListsToBeBatched.Count > 0, "Need item types consumed by the batchable object.");
ErrorUtilities.VerifyThrow(consumedMetadataReferences.Count > 0, "Need item metadata consumed by the batchable object.");
ArrayList buckets = new ArrayList();
// Get and iterate through the list of item names that we're supposed to batch on.
foreach (DictionaryEntry entry in itemListsToBeBatched)
{
string itemName = (string)entry.Key;
// Use the previously-fetched items, if possible
BuildItemGroup items;
if (entry.Value is BuildItemGroup)
{
items = (BuildItemGroup)entry.Value;
}
else
{
items = lookup.GetItems(itemName);
}
if (items != null)
{
foreach (BuildItem item in items)
{
// Get this item's values for all the metadata consumed by the batchable object.
Dictionary<string, string> itemMetadataValues = GetItemMetadataValues(parentNode, item, consumedMetadataReferences);
// put the metadata into a dummy bucket we can use for searching
ItemBucket dummyBucket = ItemBucket.GetDummyBucketForComparisons(itemMetadataValues);
// look through all previously created buckets to find a bucket whose items have the same values as
// this item for all metadata consumed by the batchable object
int matchingBucketIndex = buckets.BinarySearch(dummyBucket);
ItemBucket matchingBucket = (matchingBucketIndex >= 0)
? (ItemBucket)buckets[matchingBucketIndex]
: null;
// If we didn't find a bucket that matches this item, create a new one, adding
// this item to the bucket.
if (null == matchingBucket)
{
matchingBucket = new ItemBucket(itemListsToBeBatched.Keys, itemMetadataValues, lookup, buckets.Count);
// make sure to put the new bucket into the appropriate location
// in the sorted list as indicated by the binary search
// NOTE: observe the ~ operator (bitwise complement) in front of
// the index -- see MSDN for more information on the return value
// from the ArrayList.BinarySearch() method
buckets.Insert(~matchingBucketIndex, matchingBucket);
}
// We already have a bucket for this type of item, so add this item to
// the bucket.
matchingBucket.AddItem(item);
}
}
}
// Put the buckets back in the order in which they were discovered, so that the first
// item declared in the project file ends up in the first batch passed into the target/task.
ArrayList orderedBuckets = ArrayList.Repeat(null, buckets.Count);
foreach (ItemBucket bucket in buckets)
{
orderedBuckets[bucket.BucketSequenceNumber] = bucket;
}
return orderedBuckets;
}
/// <summary>
/// Of all the item lists that are referenced in this batchable object, which ones should we
/// batch on, and which ones should we just pass in wholesale to every invocation of the
/// target/task?
///
/// Rule #1. If the user has referenced any *qualified* item metadata such as %(EmbeddedResource.Culture),
/// then that item list "EmbeddedResource" will definitely get batched.
///
/// Rule #2. For all the unqualified item metadata such as %(Culture), we make sure that
/// every single item in every single item list being passed into the task contains a value
/// for that metadata. If not, it's an error. If so, we batch all of those item lists.
///
/// All other item lists will not be batched, and instead will be passed in wholesale to all buckets.
/// </summary>
/// <returns>Hashtable containing the item names that should be batched.</returns>
private static Hashtable GetItemListsToBeBatched
(
XmlNode parentNode,
Dictionary<string, MetadataReference> consumedMetadataReferences, // Key is [string] potentially qualified metadata name
// Value is [struct MetadataReference]
Hashtable consumedItemReferenceNames, // Key is [string] item name.
// Value is always String.Empty (unused).
Lookup lookup
)
{
// The keys in this hashtable are the names of the items that we will batch on.
// The values are always String.Empty (not used).
Hashtable itemListsToBeBatched = new Hashtable(StringComparer.OrdinalIgnoreCase);
// Loop through all the metadata references and find the ones that are qualified
// with an item name.
foreach (MetadataReference consumedMetadataReference in consumedMetadataReferences.Values)
{
if (consumedMetadataReference.itemName != null)
{
// Rule #1. Qualified metadata reference.
// For metadata references that are qualified with an item name
// (e.g., %(EmbeddedResource.Culture) ), we add that item name to the list of
// consumed item names, even if the item name wasn't otherwise referenced via
// @(...) syntax, and even if every item in the list doesn't necessary contain
// a value for this metadata. This is the special power that you get by qualifying
// the metadata reference with an item name.
itemListsToBeBatched[consumedMetadataReference.itemName] = String.Empty;
// Also add this qualified item to the consumed item references list, because
// %(EmbeddedResource.Culture) effectively means that @(EmbeddedResource) is
// being consumed, even though we may not see literally "@(EmbeddedResource)"
// in the tag anywhere. Adding it to this list allows us (down below in this
// method) to check that every item in this list has a value for each
// unqualified metadata reference.
consumedItemReferenceNames = Utilities.CreateTableIfNecessary(consumedItemReferenceNames);
consumedItemReferenceNames[consumedMetadataReference.itemName] = String.Empty;
}
}
// Loop through all the metadata references and find the ones that are unqualified.
foreach (MetadataReference consumedMetadataReference in consumedMetadataReferences.Values)
{
if (consumedMetadataReference.itemName == null)
{
// Rule #2. Unqualified metadata reference.
// For metadata references that are unqualified, every single consumed item
// must contain a value for that metadata. If any item doesn't, it's an error
// to use unqualified metadata.
if (consumedItemReferenceNames != null)
{
foreach (string consumedItemName in consumedItemReferenceNames.Keys)
{
// Loop through all the items in the item list.
BuildItemGroup items = lookup.GetItems(consumedItemName);
if (items != null)
{
// Loop through all the items in the BuildItemGroup.
foreach (BuildItem item in items)
{
ProjectErrorUtilities.VerifyThrowInvalidProject(
item.HasMetadata(consumedMetadataReference.metadataName),
parentNode, "ItemDoesNotContainValueForUnqualifiedMetadata",
item.Include, consumedItemName, consumedMetadataReference.metadataName);
}
}
// This item list passes the test of having every single item containing
// a value for this metadata. Therefore, add this item list to the batching list.
// Also, to save doing lookup.GetItems again, put the items in the table as the value.
itemListsToBeBatched[consumedItemName] = items;
}
}
}
}
return itemListsToBeBatched;
}