Fixed _perf module, which built, but didn't link dynamically. All the time, I think it never successfully imported, but its hard to believe this slipped by.

Added performance test for pack-writing, which isn't really showing what I want as it currently read data from a densly compressed pack which takes most of the time in the nearly pure python implementation. Compared to c++, all the measured performance is just below anything I'd want to use. But we shouldn't forget this is just a test implementation, writing packs is quite simple actually, if you leave out the delta compression part and the delta logic

3 files changed, 55 insertions, 13 deletions

diff --git a/gitdb/_delta_apply.c b/gitdb/_delta_apply.c
index 96ab30a..f03e7ea 100644
--- a/gitdb/_delta_apply.c
+++ b/gitdb/_delta_apply.c
@@ -1,4 +1,4 @@
-#include "_delta_apply.h"
+#include <_delta_apply.h>
 #include <stdint.h>
 #include <assert.h>
 #include <stdio.h>
@@ -463,7 +463,7 @@ void DIV_reset(DeltaInfoVector* vec)
 
 // Append one chunk to the end of the list, and return a pointer to it
 // It will not have been initialized !
-static inline
+inline
 DeltaInfo* DIV_append(DeltaInfoVector* vec)
 {
 	if (vec->size + 1 > vec->reserved_size){
@@ -703,7 +703,7 @@ typedef struct {
 } DeltaChunkList;
 
 
-static 
+ 
 int DCL_init(DeltaChunkList*self, PyObject *args, PyObject *kwds)
 {
 	if(args && PySequence_Size(args) > 0){
@@ -715,20 +715,20 @@ int DCL_init(DeltaChunkList*self, PyObject *args, PyObject *kwds)
 	return 0;
 }
 
-static
+
 void DCL_dealloc(DeltaChunkList* self)
 {
 	TSI_destroy(&(self->istream));
 }
 
-static
+
 PyObject* DCL_py_rbound(DeltaChunkList* self)
 {
 	return PyLong_FromUnsignedLongLong(self->istream.target_size);
 }
 
 // Write using a write function, taking remaining bytes from a base buffer
-static
+
 PyObject* DCL_apply(DeltaChunkList* self, PyObject* args)
 {
 	PyObject* pybuf = 0;
@@ -769,13 +769,13 @@ PyObject* DCL_apply(DeltaChunkList* self, PyObject* args)
 	Py_RETURN_NONE;
 }
 
-static PyMethodDef DCL_methods[] = {
+PyMethodDef DCL_methods[] = {
     {"apply", (PyCFunction)DCL_apply, METH_VARARGS, "Apply the given iterable of delta streams" },
     {"rbound", (PyCFunction)DCL_py_rbound, METH_NOARGS, NULL},
     {NULL}  /* Sentinel */
 };
 
-static PyTypeObject DeltaChunkListType = {
+PyTypeObject DeltaChunkListType = {
 	PyObject_HEAD_INIT(NULL)
 	0,						   /*ob_size*/
 	"DeltaChunkList",			/*tp_name*/
@@ -897,7 +897,7 @@ uint compute_chunk_count(const uchar* data, const uchar* dend, bool read_header)
 	return num_chunks;
 }
 
-static PyObject* connect_deltas(PyObject *self, PyObject *dstreams)
+PyObject* connect_deltas(PyObject *self, PyObject *dstreams)
 {
 	// obtain iterator
 	PyObject* stream_iter = 0;
@@ -1088,7 +1088,6 @@ _error:
 
 // Write using a write function, taking remaining bytes from a base buffer
 // replaces the corresponding method in python
-static
 PyObject* apply_delta(PyObject* self, PyObject* args)
 {
 	PyObject* pybbuf = 0;
diff --git a/gitdb/_delta_apply.h b/gitdb/_delta_apply.h
index 3e7e5f9..1fcd538 100644
--- a/gitdb/_delta_apply.h
+++ b/gitdb/_delta_apply.h
@@ -1,6 +1,6 @@
 #include <Python.h>
 
-static PyObject* connect_deltas(PyObject *self, PyObject *dstreams);
-static PyObject* apply_delta(PyObject* self, PyObject* args);
+extern PyObject* connect_deltas(PyObject *self, PyObject *dstreams);
+extern PyObject* apply_delta(PyObject* self, PyObject* args);
 
-static PyTypeObject DeltaChunkListType;
+extern PyTypeObject DeltaChunkListType;
diff --git a/gitdb/test/performance/test_pack_streaming.py b/gitdb/test/performance/test_pack_streaming.py
index 22a62a3..795ed1e 100644
--- a/gitdb/test/performance/test_pack_streaming.py
+++ b/gitdb/test/performance/test_pack_streaming.py
@@ -8,14 +8,57 @@ from lib import (
 	)
 
 from gitdb.db.pack import PackedDB
+from gitdb.stream import NullStream
+from gitdb.pack import PackEntity
 
 import os
 import sys
 from time import time
+from nose import SkipTest
+
+class CountedNullStream(NullStream):
+	__slots__ = '_bw'
+	def __init__(self):
+		self._bw = 0
+		
+	def bytes_written(self):
+		return self._bw
+		
+	def write(self, d):
+		self._bw += NullStream.write(self, d)
+	
 
 class TestPackStreamingPerformance(TestBigRepoR):
 	
+	def test_pack_writing(self):
+		# see how fast we can write a pack from object streams.
+		# This will not be fast, as we take time for decompressing the streams as well
+		ostream = CountedNullStream()
+		pdb = PackedDB(os.path.join(self.gitrepopath, "objects/pack"))
+		
+		ni = 5000
+		count = 0
+		total_size = 0
+		st = time()
+		objs = list()
+		for sha in pdb.sha_iter():
+			count += 1
+			objs.append(pdb.stream(sha))
+			if count == ni:
+				break
+		#END gather objects for pack-writing
+		elapsed = time() - st
+		print >> sys.stderr, "PDB Streaming: Got %i streams by sha in in %f s ( %f streams/s )" % (ni, elapsed, ni / elapsed)
+		
+		st = time()
+		PackEntity.write_pack(objs, ostream.write)
+		elapsed = time() - st
+		total_kb = ostream.bytes_written() / 1000
+		print >> sys.stderr, "PDB Streaming: Wrote pack of size %i kb in %f s (%f kb/s)" % (total_kb, elapsed, total_kb/elapsed)
+		
+	
 	def test_stream_reading(self):
+		raise SkipTest()
 		pdb = PackedDB(os.path.join(self.gitrepopath, "objects/pack"))
 		
 		# streaming only, meant for --with-profile runs