Commits


start caching file handles to packfiles


avoid a round-trip via tempfile when reading packed commits


rename a function for clarity


avoid a round-trip of data through a temp file when reading trees


process small deltas in memory; unfortunately it is not faster...


remove a useless and leaky allocation


In dump_delta_chain(), don't open the packfile if it's not needed.


fix open file handle leak in dump_delta_chain()


propagate error from delta cache insertion; fixes mem leak


add a delta cache; saves us from reading + decompressing some deltas


process delta chains in memory if max size is < 32 MB


use stricter CPPFLAGS from mpi


rename library-private headers from *_priv.h to *_lib.h


rename got_zb_priv.h to got_zbuf_priv.h


unwrap short line


return error from search_packidx() if dup_packidx() fails


add a pack index cache; speeds tree listing up quite a lot


We don't use a delta combiner, we use delta application. A "delta combiner" (as used in SVN) combines deltas and applies the combined delta to the base. We don't do this yet; we apply deltas one by one.


rename private header filenames so they are in got_ namespace


move struct got_zstream_buf to private header zb.h


read delta data from the right offset


we must store delta's type-and-size length separately to skip it


don't read plain objects to memory during unpack


don't assume objects IDs are sorted by their values in packidx


make got_inflate_read() properly indicate if it needs to be called again