org.eclipse.jgit.storage.pack
Interface ObjectReuseAsIs

All Known Implementing Classes:

DfsReader

public interface ObjectReuseAsIs

Extension of ObjectReader that supports reusing objects in packs.

ObjectReader implementations may also optionally implement this interface to support PackWriter with a means of copying an object that is already in pack encoding format directly into the output stream, without incurring decompression and recompression overheads.

Method Summary

void	copyObjectAsIs(PackOutputStream out, ObjectToPack otp, boolean validate) Output a previously selected representation.
void	copyPackAsIs(PackOutputStream out, CachedPack pack, boolean validate) Append an entire pack's contents onto the output stream.
Collection<CachedPack>	getCachedPacks() Obtain the available cached packs.
ObjectToPack	newObjectToPack(RevObject obj) Allocate a new PackWriter state structure for an object.
void	selectObjectRepresentation(PackWriter packer, ProgressMonitor monitor, Iterable<ObjectToPack> objects) Select the best object representation for a packer.
void	writeObjects(PackOutputStream out, List<ObjectToPack> list) Write objects to the pack stream in roughly the order given.

Method Detail

newObjectToPack

ObjectToPack newObjectToPack(RevObject obj)

Allocate a new PackWriter state structure for an object.

PackWriter allocates these objects to keep track of the per-object state, and how to load the objects efficiently into the generated stream. Implementers may subclass this type with additional object state, such as to remember what file and offset contains the object's pack encoded data.

Parameters:

obj - identity of the object that will be packed. The object's parsed status is undefined here. Implementers must not rely on the object being parsed.

Returns:

a new instance for this object.

selectObjectRepresentation

void selectObjectRepresentation(PackWriter packer,
                                ProgressMonitor monitor,
                                Iterable<ObjectToPack> objects)
                                throws IOException,
                                       MissingObjectException

Select the best object representation for a packer.

Implementations should iterate through all available representations of an object, and pass them in turn to the PackWriter though PackWriter.select(ObjectToPack, StoredObjectRepresentation) so the writer can select the most suitable representation to reuse into the output stream.

If the implementation returns CachedPack from getCachedPacks(), it must consider the representation of any object that is stored in any of the offered CachedPacks. PackWriter relies on this behavior to prune duplicate objects out of the pack stream when it selects a CachedPack and the object was also reached through the thin-pack enumeration.

The implementation may choose to consider multiple objects at once on concurrent threads, but must evaluate all representations of an object within the same thread.

Parameters:

packer - the packer that will write the object in the near future.

monitor - progress monitor, implementation should update the monitor once for each item in the iteration when selection is done.

objects - the objects that are being packed.

Throws:

MissingObjectException - there is no representation available for the object, as it is no longer in the repository. Packing will abort.

IOException - the repository cannot be accessed. Packing will abort.

writeObjects

void writeObjects(PackOutputStream out,
                  List<ObjectToPack> list)
                  throws IOException

Write objects to the pack stream in roughly the order given. PackWriter invokes this method to write out one or more objects, in approximately the order specified by the iteration over the list. A simple implementation of this method would just iterate the list and output each object:

 for (ObjectToPack obj : list)
   out.writeObject(obj)
 

However more sophisticated implementors may try to perform some (small) reordering to access objects that are stored close to each other at roughly the same time. Implementations may choose to write objects out of order, but this may increase pack file size due to using a larger header format to reach a delta base that is later in the stream. It may also reduce data locality for the reader, slowing down data access. Invoking PackOutputStream.writeObject(ObjectToPack) will cause copyObjectAsIs(PackOutputStream, ObjectToPack, boolean) to be invoked recursively on this if the current object is scheduled for reuse.

Parameters:

out - the stream to write each object to.

list - the list of objects to write. Objects should be written in approximately this order. Implementors may resort the list elements in-place during writing if desired.

Throws:

IOException - the stream cannot be written to, or one or more required objects cannot be accessed from the object database.

copyObjectAsIs

void copyObjectAsIs(PackOutputStream out,
                    ObjectToPack otp,
                    boolean validate)
                    throws IOException,
                           StoredObjectRepresentationNotAvailableException

Output a previously selected representation.

PackWriter invokes this method only if a representation previously given to it by selectObjectRepresentation was chosen for reuse into the output stream. The otp argument is an instance created by this reader's own newObjectToPack, and the representation data saved within it also originated from this reader.

Implementors must write the object header before copying the raw data to the output stream. The typical implementation is like:

 MyToPack mtp = (MyToPack) otp;
 byte[] raw;
 if (validate)
         raw = validate(mtp); // throw SORNAE here, if at all
 else
         raw = readFast(mtp);
 out.writeHeader(mtp, mtp.inflatedSize);
 out.write(raw);
 

Parameters:

out - stream the object should be written to.

otp - the object's saved representation information.

validate - if true the representation must be validated and not be corrupt before being reused. If false, validation may be skipped as it will be performed elsewhere in the processing pipeline.

Throws:

StoredObjectRepresentationNotAvailableException - the previously selected representation is no longer available. If thrown before out.writeHeader the pack writer will try to find another representation, and write that one instead. If throw after out.writeHeader, packing will abort.

IOException - the stream's write method threw an exception. Packing will abort.

getCachedPacks

Collection<CachedPack> getCachedPacks()
                                      throws IOException

Obtain the available cached packs.

A cached pack has known starting points and may be sent entirely as-is, with almost no effort on the sender's part.

Returns:

the available cached packs.

Throws:

IOException - the cached packs cannot be listed from the repository. Callers may choose to ignore this and continue as-if there were no cached packs.

copyPackAsIs

void copyPackAsIs(PackOutputStream out,
                  CachedPack pack,
                  boolean validate)
                  throws IOException

Append an entire pack's contents onto the output stream.

The entire pack, excluding its header and trailing footer is sent.

Parameters:

out - stream to append the pack onto.

pack - the cached pack to send.

validate - if true the representation must be validated and not be corrupt before being reused. If false, validation may be skipped as it will be performed elsewhere in the processing pipeline.

Throws:

IOException - the pack cannot be read, or stream did not accept a write.