[PATCH 0/6] Merge David's SVN exporter

STALE3732d

15 messages, 4 authors, 2016-06-15 · open the first message on its own page

[PATCH 0/6] Merge David's SVN exporter

From: Ramkumar Ramachandra <hidden>
Date: 2016-06-15 22:48:54

Hi,

Sorry about the mess- The authorship is corrected in this series.

-- Ram

David Barr (5):
  Add memory pool library
  Add library for string-specific memory pool
  Add stream helper library
  Add infrastructure to write revisions in fast-export format
  Add SVN dump parser

Jason Evans (1):
  Add cpp macro implementation of treaps

 vcs-svn/fast_export.c |   69 ++++++++++
 vcs-svn/fast_export.h |   14 ++
 vcs-svn/line_buffer.c |  129 ++++++++++++++++++
 vcs-svn/line_buffer.h |   14 ++
 vcs-svn/obj_pool.h    |   98 ++++++++++++++
 vcs-svn/repo_tree.c   |  353 +++++++++++++++++++++++++++++++++++++++++++++++++
 vcs-svn/repo_tree.h   |   27 ++++
 vcs-svn/string_pool.c |  110 +++++++++++++++
 vcs-svn/string_pool.h |   14 ++
 vcs-svn/svndump.c     |  294 ++++++++++++++++++++++++++++++++++++++++
 vcs-svn/svndump.h     |    7 +
 vcs-svn/trp.h         |  118 ++++++++++++++++
 vcs-svn/trp.txt       |   62 +++++++++
 13 files changed, 1309 insertions(+), 0 deletions(-)
 create mode 100644 vcs-svn/fast_export.c
 create mode 100644 vcs-svn/fast_export.h
 create mode 100644 vcs-svn/line_buffer.c
 create mode 100644 vcs-svn/line_buffer.h
 create mode 100644 vcs-svn/obj_pool.h
 create mode 100644 vcs-svn/repo_tree.c
 create mode 100644 vcs-svn/repo_tree.h
 create mode 100644 vcs-svn/string_pool.c
 create mode 100644 vcs-svn/string_pool.h
 create mode 100644 vcs-svn/svndump.c
 create mode 100644 vcs-svn/svndump.h
 create mode 100644 vcs-svn/trp.h
 create mode 100644 vcs-svn/trp.txt

[PATCH 1/6] Add memory pool library

From: Ramkumar Ramachandra <hidden>
Date: 2016-06-15 22:48:54

From: David Barr <redacted>

Add a memory pool library implemented using cpp macros. The library
provides macros that can be used to create a type-specific memory pool
API.

Signed-off-by: David Barr <redacted>
Signed-off-by: Ramkumar Ramachandra <redacted>
---
 vcs-svn/obj_pool.h |   98 ++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 files changed, 98 insertions(+), 0 deletions(-)
 create mode 100644 vcs-svn/obj_pool.h
diff --git a/vcs-svn/obj_pool.h b/vcs-svn/obj_pool.h
new file mode 100644
index 0000000..84c8321
--- /dev/null
+++ b/vcs-svn/obj_pool.h
@@ -0,0 +1,98 @@
+#ifndef OBJ_POOL_H_
+#define OBJ_POOL_H_
+
+#include "git-compat-util.h"
+
+/*
+ * The obj_pool_gen() macro generates a type-specific memory pool
+ * implementation.
+ *
+ * Arguments:
+ *
+ *   pre              : Prefix for generated functions (ex: string_).
+ *   obj_t            : Type for treap data structure (ex: char).
+ *   intial_capacity  : The initial size of the memory pool (ex: 4096).
+ *
+ */
+#define obj_pool_gen(pre, obj_t, initial_capacity) \
+static struct { \
+	uint32_t size; \
+	uint32_t capacity; \
+	obj_t *base; \
+        FILE *file; \
+} pre##_pool = { 0, 0, NULL, NULL}; \
+static void pre##_init(void) \
+{ \
+	struct stat st; \
+	size_t ps = sysconf (_SC_PAGESIZE); \
+	/* Touch binary file before opening read/write */ \
+	pre##_pool.file = fopen(#pre ".bin", "a"); \
+	fclose(pre##_pool.file); \
+	/* Open, check size, compute capacity */ \
+	pre##_pool.file = fopen(#pre ".bin", "r+"); \
+	fstat(fileno(pre##_pool.file), &st); \
+	pre##_pool.size = st.st_size / sizeof(obj_t); \
+	pre##_pool.capacity = ((st.st_size + ps - 1) & ~(ps - 1)) / sizeof(obj_t); \
+	if (pre##_pool.capacity < initial_capacity) \
+		pre##_pool.capacity = initial_capacity; \
+	/* Truncate to calculated capacity and map to VM */ \
+	ftruncate(fileno(pre##_pool.file), pre##_pool.capacity * sizeof(obj_t)); \
+	pre##_pool.base = mmap(0, pre##_pool.capacity * sizeof(obj_t), \
+				PROT_READ | PROT_WRITE, MAP_SHARED, \
+				fileno(pre##_pool.file), 0); \
+} \
+static uint32_t pre##_alloc(uint32_t count) \
+{ \
+	uint32_t offset; \
+	if (pre##_pool.size + count > pre##_pool.capacity) { \
+		if (NULL == pre##_pool.base) \
+			pre##_init(); \
+		fsync(fileno(pre##_pool.file)); \
+		munmap(pre##_pool.base, \
+			pre##_pool.capacity * sizeof(obj_t)); \
+		pre##_pool.base = NULL; \
+		while (pre##_pool.size + count > pre##_pool.capacity) \
+			if (pre##_pool.capacity) \
+				pre##_pool.capacity *= 2; \
+			else \
+				pre##_pool.capacity = initial_capacity; \
+		ftruncate(fileno(pre##_pool.file), \
+				pre##_pool.capacity * sizeof(obj_t)); \
+		pre##_pool.base = \
+			mmap(0, pre##_pool.capacity * sizeof(obj_t), \
+				PROT_READ | PROT_WRITE, MAP_SHARED, \
+				fileno(pre##_pool.file), 0); \
+	} \
+	offset = pre##_pool.size; \
+	pre##_pool.size += count; \
+	return offset; \
+} \
+static void pre##_free(uint32_t count) \
+{ \
+	pre##_pool.size -= count; \
+} \
+static uint32_t pre##_offset(obj_t *obj) \
+{ \
+	return obj == NULL ? ~0 : obj - pre##_pool.base; \
+} \
+static obj_t *pre##_pointer(uint32_t offset) \
+{ \
+	return offset >= pre##_pool.size ? NULL : &pre##_pool.base[offset]; \
+} \
+static void pre##_reset(void) \
+{ \
+	if (pre##_pool.base) { \
+		fsync(fileno(pre##_pool.file)); \
+		munmap(pre##_pool.base, \
+			pre##_pool.capacity * sizeof(obj_t)); \
+		ftruncate(fileno(pre##_pool.file), \
+				pre##_pool.size * sizeof(obj_t)); \
+		fclose(pre##_pool.file); \
+	} \
+	pre##_pool.base = NULL; \
+	pre##_pool.size = 0; \
+	pre##_pool.capacity = 0; \
+	pre##_pool.file = NULL; \
+}
+
+#endif
-- 
1.7.1

[PATCH 2/6] Add cpp macro implementation of treaps

From: Ramkumar Ramachandra <hidden>
Date: 2016-06-15 22:48:54

From: Jason Evans <redacted>

The implementation exposes an API to generate type-specific treap
implmentation and various functions to operate on it. It uses
obj_pool.h to store memory nodes in a treap.

Treaps provide a memory-efficient binary search tree structure.
Insertion/deletion/search are about as about as fast in the average
case as red-black trees and the chances of worst-case behavior are
vanishingly small, thanks to (pseudo-)randomness.  That is a small
price to pay, given that treaps are much simpler to implement.

[db: Altered to reference nodes by offset from a common base pointer]
[db: Bob Jenkins' hashing implementation dropped for Knuth's]
[db: Methods unnecessary for search and insert dropped]

Signed-off-by: David Barr <redacted>
Signed-off-by: Ramkumar Ramachandra <redacted>
---
 vcs-svn/trp.h   |  118 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
 vcs-svn/trp.txt |   62 +++++++++++++++++++++++++++++
 2 files changed, 180 insertions(+), 0 deletions(-)
 create mode 100644 vcs-svn/trp.h
 create mode 100644 vcs-svn/trp.txt
diff --git a/vcs-svn/trp.h b/vcs-svn/trp.h
new file mode 100644
index 0000000..c90f5c3
--- /dev/null
+++ b/vcs-svn/trp.h
@@ -0,0 +1,118 @@
+/*
+ * cpp macro implementation of treaps.
+ *
+ * Usage:
+ *   #include <stdint.h>
+ *   #include <trp.h>
+ *   trp_gen(...)
+ */
+
+#ifndef TRP_H_
+#define TRP_H_
+
+/* Node structure. */
+struct trp_node {
+	uint32_t trpn_left;
+	uint32_t trpn_right;
+};
+
+/* Root structure. */
+struct trp_root {
+	uint32_t trp_root;
+};
+
+/* Pointer/Offset conversion */
+#define trpn_pointer(a_base, a_offset) (a_base##_pointer(a_offset))
+#define trpn_offset(a_base, a_pointer) (a_base##_offset(a_pointer))
+
+/* Left accessors. */
+#define trp_left_get(a_base, a_field, a_node) \
+	trpn_pointer(a_base, (a_node)->a_field.trpn_left)
+#define trp_left_set(a_base, a_field, a_node, a_left) \
+	(a_node)->a_field.trpn_left = trpn_offset(a_base, a_left)
+
+/* Right accessors. */
+#define trp_right_get(a_base, a_field, a_node) \
+	trpn_pointer(a_base, (a_node)->a_field.trpn_right)
+#define trp_right_set(a_base, a_field, a_node, a_right) \
+	(a_node)->a_field.trpn_right = trpn_offset(a_base, a_right)
+
+/* Priority accessors. */
+#define KNUTH_GOLDEN_RATIO_32BIT 2654435761u
+#define trp_prio_get(a_node) \
+	(KNUTH_GOLDEN_RATIO_32BIT*(uint32_t)(uintptr_t)(a_node))
+
+/* Node initializer. */
+#define trp_node_new(a_base, a_field, a_node) \
+	trp_left_set(a_base, a_field, (a_node), NULL); \
+	trp_right_set(a_base, a_field, (a_node), NULL)
+
+/* Internal utility macros. */
+#define trpn_rotate_left(a_base, a_field, a_node, r_node) \
+	do { (r_node) = trp_right_get(a_base, a_field, (a_node)); \
+	trp_right_set(a_base, a_field, (a_node), \
+		trp_left_get(a_base, a_field, (r_node))); \
+	trp_left_set(a_base, a_field, (r_node), (a_node)); } while(0)
+
+#define trpn_rotate_right(a_base, a_field, a_node, r_node) \
+	do { (r_node) = trp_left_get(a_base, a_field, (a_node)); \
+	trp_left_set(a_base, a_field, (a_node), \
+		trp_right_get(a_base, a_field, (r_node))); \
+	trp_right_set(a_base, a_field, (r_node), (a_node)); } while(0)
+
+#define trp_gen(a_attr, a_pre, a_type, a_field, a_base, a_cmp) \
+a_attr a_type *a_pre##psearch(struct trp_root *treap, a_type *key) \
+{ \
+	a_type *ret; \
+	a_type *tnode = trpn_pointer(a_base, treap->trp_root); \
+	ret = NULL; \
+	while (tnode != NULL) { \
+		int cmp = (a_cmp)(key, tnode); \
+		if (cmp < 0) \
+			tnode = trp_left_get(a_base, a_field, tnode); \
+		else if (cmp > 0) { \
+			ret = tnode; \
+			tnode = trp_right_get(a_base, a_field, tnode); \
+		} else { \
+			ret = tnode; \
+			break; \
+		} \
+	} \
+	return (ret); \
+} \
+a_attr a_type *a_pre##insert_recurse(a_type *cur_node, a_type *ins_node) \
+{ \
+	if (cur_node == NULL) \
+		return (ins_node); \
+	else { \
+		a_type *ret; \
+		int cmp = a_cmp(ins_node, cur_node); \
+		if (cmp < 0) { \
+			a_type *left = a_pre##insert_recurse( \
+				trp_left_get(a_base, a_field, cur_node), ins_node); \
+			trp_left_set(a_base, a_field, cur_node, left); \
+			if (trp_prio_get(left) < trp_prio_get(cur_node)) \
+				trpn_rotate_right(a_base, a_field, cur_node, ret); \
+			else \
+				ret = cur_node; \
+		} else { \
+			a_type *right = a_pre##insert_recurse( \
+				trp_right_get(a_base, a_field, cur_node), ins_node); \
+			trp_right_set(a_base, a_field, cur_node, right); \
+			if (trp_prio_get(right) < trp_prio_get(cur_node)) \
+				trpn_rotate_left(a_base, a_field, cur_node, ret); \
+			else \
+				ret = cur_node; \
+		} \
+		return (ret); \
+	} \
+} \
+a_attr void a_pre##insert(struct trp_root *treap, a_type *node) \
+{ \
+	trp_node_new(a_base, a_field, node); \
+	treap->trp_root = trpn_offset(a_base, a_pre##insert_recurse( \
+					      trpn_pointer(a_base, treap->trp_root), \
+					      node)); \
+}
+
+#endif
diff --git a/vcs-svn/trp.txt b/vcs-svn/trp.txt
new file mode 100644
index 0000000..7cf9b40
--- /dev/null
+++ b/vcs-svn/trp.txt
@@ -0,0 +1,62 @@
+TODO: Update this documentation to match the changes to trp.h
+
+The trp_gen() macro generates a type-specific treap implementation,
+based on the above cpp macros.
+
+Arguments:
+
+  a_attr     : Function attribute for generated functions (ex: static).
+  a_pre      : Prefix for generated functions (ex: treap_).
+  a_t_type   : Type for treap data structure (ex: treap_t).
+  a_type     : Type for treap node data structure (ex: treap_node_t).
+  a_field    : Name of treap node linkage (ex: treap_link).
+  a_base     : Expression for the base pointer from which nodes are offset.
+  a_cmp      : Node comparison function name, with the following prototype:
+                 int (a_cmp *)(a_type *a_node, a_type *a_other);
+                                       ^^^^^^
+                                    or a_key
+               Interpretation of comparision function return values:
+                 -1 : a_node <  a_other
+                  0 : a_node == a_other
+                  1 : a_node >  a_other
+               In all cases, the a_node or a_key macro argument is the first
+               argument to the comparison function, which makes it possible
+               to write comparison functions that treat the first argument
+               specially.
+
+Assuming the following setup:
+
+  typedef struct ex_node_s ex_node_t;
+  struct ex_node_s {
+      trp_node(ex_node_t) ex_link;
+  };
+  typedef trp(ex_node_t) ex_t;
+  static ex_node_t ex_base[MAX_NODES];
+  trp_gen(static, ex_, ex_t, ex_node_t, ex_link, ex_base, ex_cmp)
+
+The following API is generated:
+
+  static void
+  ex_new(ex_t *treap);
+      Description: Initialize a treap structure.
+      Args:
+        treap: Pointer to an uninitialized treap object.
+
+  static ex_node_t *
+  ex_psearch(ex_t *treap, ex_node_t *key);
+      Description: Search for node that matches key.  If no match is found,
+                   return what would be key's successor/predecessor, were
+                   key in treap.
+      Args:
+        treap: Pointer to a initialized treap object.
+        key  : Search key.
+      Ret: Node in treap that matches key, or if no match, hypothetical
+           node's successor/predecessor (NULL if no successor/predecessor).
+
+  static void
+  ex_insert(ex_t *treap, ex_node_t *node);
+      Description: Insert node into treap.
+      Args:
+        treap: Pointer to a initialized treap object.
+        node : Node to be inserted into treap.
+
-- 
1.7.1

[PATCH 3/6] Add library for string-specific memory pool

From: Ramkumar Ramachandra <hidden>
Date: 2016-06-15 22:48:54

From: David Barr <redacted>

This library uses the macros in the obj_pool.h and trp.h to create a
memory pool for strings and expose an API for handling them.

Signed-off-by: David Barr <redacted>
Signed-off-by: Ramkumar Ramachandra <redacted>
---
 vcs-svn/string_pool.c |  110 +++++++++++++++++++++++++++++++++++++++++++++++++
 vcs-svn/string_pool.h |   14 ++++++
 2 files changed, 124 insertions(+), 0 deletions(-)
 create mode 100644 vcs-svn/string_pool.c
 create mode 100644 vcs-svn/string_pool.h
diff --git a/vcs-svn/string_pool.c b/vcs-svn/string_pool.c
new file mode 100644
index 0000000..cfcf127
--- /dev/null
+++ b/vcs-svn/string_pool.c
@@ -0,0 +1,110 @@
+#include "git-compat-util.h"
+
+#include "trp.h"
+#include "obj_pool.h"
+#include "string_pool.h"
+
+typedef struct node_s node_t;
+static struct trp_root tree = { ~0 };
+
+struct node_s {
+	uint32_t offset;
+	struct trp_node children;
+};
+
+/* Create two memory pools: one for node_t, and another for strings */
+obj_pool_gen(node, node_t, 4096);
+obj_pool_gen(string, char, 4096);
+
+static char *node_value(node_t *node)
+{
+	return node ? string_pointer(node->offset) : NULL;
+}
+
+static int node_value_cmp(node_t *a, node_t *b)
+{
+	return strcmp(node_value(a), node_value(b));
+}
+
+static int node_indentity_cmp(node_t *a, node_t *b)
+{
+	int r = node_value_cmp(a, b);
+	return r ? r : (((uintptr_t) a) > ((uintptr_t) b))
+		- (((uintptr_t) a) < ((uintptr_t) b));
+}
+
+/* Build a Treap from the node_s structure (a trp_node w/ offset) */
+trp_gen(static, tree_, node_t, children, node, node_indentity_cmp);
+
+char *pool_fetch(uint32_t entry)
+{
+	return node_value(node_pointer(entry));
+}
+
+uint32_t pool_intern(char *key)
+{
+	/* Canonicalize key */
+	node_t *match = NULL;
+	uint32_t key_len;
+	if (key == NULL)
+		return ~0;
+	key_len = strlen(key) + 1;
+	node_t *node = node_pointer(node_alloc(1));
+	node->offset = string_alloc(key_len);
+	strcpy(node_value(node), key);
+	match = tree_psearch(&tree, node);
+	if (!match || node_value_cmp(node, match)) {
+		tree_insert(&tree, node);
+	} else {
+		node_free(1);
+		string_free(key_len);
+		node = match;
+	}
+	return node_offset(node);
+}
+
+uint32_t pool_tok_r(char *str, const char *delim, char **saveptr)
+{
+	char *token = strtok_r(str, delim, saveptr);
+	return token ? pool_intern(token) : ~0;
+}
+
+void pool_print_seq(uint32_t len, uint32_t *seq, char delim, FILE *stream)
+{
+	uint32_t i;
+	for (i = 0; i < len && ~seq[i]; i++) {
+		fputs(pool_fetch(seq[i]), stream);
+		if (i < len - 1 && ~seq[i + 1])
+			fputc(delim, stream);
+	}
+}
+
+uint32_t pool_tok_seq(uint32_t max, uint32_t *seq, char *delim, char *str)
+{
+	char *context = NULL;
+	uint32_t length = 0, token = str ? pool_tok_r(str, delim, &context) : ~0;
+	while (length < max) {
+		seq[length++] = token;
+		if (token == ~0)
+			break;
+		token = pool_tok_r(NULL, delim, &context);
+	}
+	seq[length ? length - 1 : 0] = ~0;
+	return length;
+}
+
+void pool_init(void)
+{
+	uint32_t node;
+	node_init();
+	string_init();
+	for (node = 0; node < node_pool.size; node++) {
+		tree_insert(&tree, node_pointer(node));
+	}
+}
+
+void pool_reset(void)
+{
+	node_reset();
+	string_reset();
+}
diff --git a/vcs-svn/string_pool.h b/vcs-svn/string_pool.h
new file mode 100644
index 0000000..e2cc447
--- /dev/null
+++ b/vcs-svn/string_pool.h
@@ -0,0 +1,14 @@
+#ifndef STRING_POOL_H_
+#define	STRING_POOL_H_
+
+#include "git-compat-util.h"
+
+uint32_t pool_intern(char *key);
+char *pool_fetch(uint32_t entry);
+uint32_t pool_tok_r(char *str, const char *delim, char **saveptr);
+void pool_print_seq(uint32_t len, uint32_t *seq, char delim, FILE *stream);
+uint32_t pool_tok_seq(uint32_t max, uint32_t *seq, char *delim, char *str);
+void pool_init(void);
+void pool_reset(void);
+
+#endif
-- 
1.7.1

[PATCH 5/6] Add infrastructure to write revisions in fast-export format

From: Ramkumar Ramachandra <hidden>
Date: 2016-06-15 22:48:54

From: David Barr <redacted>

repo_tree maintains the exporter's state and provides a facility to
to call fast_export, which then writes objects to stdout suitable for
consumption by git-fast-import.

Signed-off-by: David Barr <redacted>
Signed-off-by: Ramkumar Ramachandra <redacted>
---
 vcs-svn/fast_export.c |   69 ++++++++++
 vcs-svn/fast_export.h |   14 ++
 vcs-svn/repo_tree.c   |  353 +++++++++++++++++++++++++++++++++++++++++++++++++
 vcs-svn/repo_tree.h   |   27 ++++
 4 files changed, 463 insertions(+), 0 deletions(-)
 create mode 100644 vcs-svn/fast_export.c
 create mode 100644 vcs-svn/fast_export.h
 create mode 100644 vcs-svn/repo_tree.c
 create mode 100644 vcs-svn/repo_tree.h
diff --git a/vcs-svn/fast_export.c b/vcs-svn/fast_export.c
new file mode 100644
index 0000000..e5eb409
--- /dev/null
+++ b/vcs-svn/fast_export.c
@@ -0,0 +1,69 @@
+#include "git-compat-util.h"
+
+#include "fast_export.h"
+#include "line_buffer.h"
+#include "repo_tree.h"
+#include "string_pool.h"
+
+#define MAX_GITSVN_LINE_LEN 4096
+
+static uint32_t first_commit_done;
+
+void fast_export_delete(uint32_t depth, uint32_t *path)
+{
+	putchar('D');
+	putchar(' ');
+	pool_print_seq(depth, path, '/', stdout);
+	putchar('\n');
+}
+
+void fast_export_modify(uint32_t depth, uint32_t *path, uint32_t mode,
+                        uint32_t mark)
+{
+	printf("M %06o :%d ", mode, mark);
+	pool_print_seq(depth, path, '/', stdout);
+	putchar('\n');
+}
+
+static char gitsvnline[MAX_GITSVN_LINE_LEN];
+void fast_export_commit(uint32_t revision, uint32_t author, char *log,
+                        uint32_t uuid, uint32_t url,
+                        unsigned long timestamp)
+{
+	if (!log)
+		log = "";
+	if (~uuid && ~url) {
+		snprintf(gitsvnline, MAX_GITSVN_LINE_LEN, "\n\ngit-svn-id: %s@%d %s\n",
+				 pool_fetch(url), revision, pool_fetch(uuid));
+	} else {
+		*gitsvnline = '\0';
+	}
+	printf("commit refs/heads/master\n");
+	printf("committer %s <%s@%s> %ld +0000\n",
+		   ~author ? pool_fetch(author) : "nobody",
+		   ~author ? pool_fetch(author) : "nobody",
+		   ~uuid ? pool_fetch(uuid) : "local", timestamp);
+	printf("data %zd\n%s%s\n",
+		   strlen(log) + strlen(gitsvnline), log, gitsvnline);
+	if (!first_commit_done) {
+		if (revision > 1)
+			printf("from refs/heads/master^0\n");
+		first_commit_done = 1;
+	}
+	repo_diff(revision - 1, revision);
+	fputc('\n', stdout);
+
+	printf("progress Imported commit %d.\n\n", revision);
+}
+
+void fast_export_blob(uint32_t mode, uint32_t mark, uint32_t len)
+{
+	if (mode == REPO_MODE_LNK) {
+		/* svn symlink blobs start with "link " */
+		buffer_skip_bytes(5);
+		len -= 5;
+	}
+	printf("blob\nmark :%d\ndata %d\n", mark, len);
+	buffer_copy_bytes(len);
+	fputc('\n', stdout);
+}
diff --git a/vcs-svn/fast_export.h b/vcs-svn/fast_export.h
new file mode 100644
index 0000000..e9ea3ed
--- /dev/null
+++ b/vcs-svn/fast_export.h
@@ -0,0 +1,14 @@
+#ifndef FAST_EXPORT_H_
+#define FAST_EXPORT_H_
+
+#include <stdint.h>
+#include <time.h>
+
+void fast_export_delete(uint32_t depth, uint32_t *path);
+void fast_export_modify(uint32_t depth, uint32_t *path, uint32_t mode,
+                        uint32_t mark);
+void fast_export_commit(uint32_t revision, uint32_t author, char *log,
+                        uint32_t uuid, uint32_t url, unsigned long timestamp);
+void fast_export_blob(uint32_t mode, uint32_t mark, uint32_t len);
+
+#endif
diff --git a/vcs-svn/repo_tree.c b/vcs-svn/repo_tree.c
new file mode 100644
index 0000000..5dfffb8
--- /dev/null
+++ b/vcs-svn/repo_tree.c
@@ -0,0 +1,353 @@
+#include "git-compat-util.h"
+
+#include "string_pool.h"
+#include "repo_tree.h"
+#include "obj_pool.h"
+#include "fast_export.h"
+
+struct repo_dirent {
+	uint32_t name_offset;
+	uint32_t mode;
+	uint32_t content_offset;
+};
+
+struct repo_dir {
+	uint32_t size;
+	uint32_t first_offset;
+};
+
+struct repo_commit {
+	uint32_t mark;
+	uint32_t root_dir_offset;
+};
+
+/* Generate memory pools for commit, dir and dirent */
+obj_pool_gen(commit, struct repo_commit, 4096);
+obj_pool_gen(dir, struct repo_dir, 4096);
+obj_pool_gen(dirent, struct repo_dirent, 4096);
+
+static uint32_t num_dirs_saved;
+static uint32_t num_dirents_saved;
+static uint32_t active_commit;
+static uint32_t _mark;
+
+uint32_t next_blob_mark(void)
+{
+	return _mark++;
+}
+
+static struct repo_dir *repo_commit_root_dir(struct repo_commit *commit)
+{
+	return dir_pointer(commit->root_dir_offset);
+}
+
+static struct repo_dirent *repo_first_dirent(struct repo_dir *dir)
+{
+	return dirent_pointer(dir->first_offset);
+}
+
+static int repo_dirent_name_cmp(const void *a, const void *b)
+{
+	const struct repo_dirent *dirent1 = a, *dirent2 = b;
+	uint32_t a_offset = dirent1->name_offset;
+	uint32_t b_offset = dirent2->name_offset;
+	return (a_offset > b_offset) - (a_offset < b_offset);
+}
+
+static struct repo_dirent *repo_dirent_by_name(struct repo_dir *dir,
+                                          uint32_t name_offset)
+{
+	struct repo_dirent key;
+	if (dir == NULL || dir->size == 0)
+		return NULL;
+	key.name_offset = name_offset;
+	return bsearch(&key, repo_first_dirent(dir), dir->size,
+				   sizeof(struct repo_dirent), repo_dirent_name_cmp);
+}
+
+static int repo_dirent_is_dir(struct repo_dirent *dirent)
+{
+	return dirent != NULL && dirent->mode == REPO_MODE_DIR;
+}
+
+static struct repo_dir *repo_dir_from_dirent(struct repo_dirent *dirent)
+{
+	if (!repo_dirent_is_dir(dirent))
+		return NULL;
+	return dir_pointer(dirent->content_offset);
+}
+
+static uint32_t dir_with_dirents_alloc(uint32_t size)
+{
+	uint32_t offset = dir_alloc(1);
+	dir_pointer(offset)->size = size;
+	dir_pointer(offset)->first_offset = dirent_alloc(size);
+	return offset;
+}
+
+static struct repo_dir *repo_clone_dir(struct repo_dir *orig_dir, uint32_t padding)
+{
+	uint32_t orig_o, new_o, dirent_o;
+	orig_o = dir_offset(orig_dir);
+	if (orig_o < num_dirs_saved) {
+		new_o = dir_with_dirents_alloc(orig_dir->size + padding);
+		orig_dir = dir_pointer(orig_o);
+		dirent_o = dir_pointer(new_o)->first_offset;
+	} else {
+		if (padding == 0)
+			return orig_dir;
+		new_o = orig_o;
+		dirent_o = dirent_alloc(orig_dir->size + padding);
+	}
+	memcpy(dirent_pointer(dirent_o), repo_first_dirent(orig_dir),
+		   orig_dir->size * sizeof(struct repo_dirent));
+	dir_pointer(new_o)->size = orig_dir->size + padding;
+	dir_pointer(new_o)->first_offset = dirent_o;
+	return dir_pointer(new_o);
+}
+
+static struct repo_dirent *repo_read_dirent(uint32_t revision, uint32_t *path)
+{
+	uint32_t name = 0;
+	struct repo_dir *dir = NULL;
+	struct repo_dirent *dirent = NULL;
+	dir = repo_commit_root_dir(commit_pointer(revision));
+	while (~(name = *path++)) {
+		dirent = repo_dirent_by_name(dir, name);
+		if (dirent == NULL) {
+			return NULL;
+		} else if (repo_dirent_is_dir(dirent)) {
+			dir = repo_dir_from_dirent(dirent);
+		} else {
+			break;
+		}
+	}
+	return dirent;
+}
+
+static void
+repo_write_dirent(uint32_t *path, uint32_t mode, uint32_t content_offset,
+                  uint32_t del)
+{
+	uint32_t name, revision, dirent_o = ~0, dir_o = ~0, parent_dir_o = ~0;
+	struct repo_dir *dir;
+	struct repo_dirent *dirent = NULL;
+	revision = active_commit;
+	dir = repo_commit_root_dir(commit_pointer(revision));
+	dir = repo_clone_dir(dir, 0);
+	commit_pointer(revision)->root_dir_offset = dir_offset(dir);
+	while (~(name = *path++)) {
+		parent_dir_o = dir_offset(dir);
+		dirent = repo_dirent_by_name(dir, name);
+		if (dirent == NULL) {
+			dir = repo_clone_dir(dir, 1);
+			dirent = &repo_first_dirent(dir)[dir->size - 1];
+			dirent->name_offset = name;
+			dirent->mode = REPO_MODE_DIR;
+			qsort(repo_first_dirent(dir), dir->size,
+				  sizeof(struct repo_dirent), repo_dirent_name_cmp);
+			dirent = repo_dirent_by_name(dir, name);
+			dir_o = dir_with_dirents_alloc(0);
+			dirent->content_offset = dir_o;
+			dir = dir_pointer(dir_o);
+		} else if ((dir = repo_dir_from_dirent(dirent))) {
+			dirent_o = dirent_offset(dirent);
+			dir = repo_clone_dir(dir, 0);
+			if (dirent_o != ~0)
+				dirent_pointer(dirent_o)->content_offset = dir_offset(dir);
+		} else {
+			dirent->mode = REPO_MODE_DIR;
+			dirent_o = dirent_offset(dirent);
+			dir_o = dir_with_dirents_alloc(0);
+			dirent = dirent_pointer(dirent_o);
+			dir = dir_pointer(dir_o);
+			dirent->content_offset = dir_o;
+		}
+	}
+	if (dirent) {
+		dirent->mode = mode;
+		dirent->content_offset = content_offset;
+		if (del && ~parent_dir_o) {
+			dirent->name_offset = ~0;
+			dir = dir_pointer(parent_dir_o);
+			qsort(repo_first_dirent(dir), dir->size,
+				  sizeof(struct repo_dirent), repo_dirent_name_cmp);
+			dir->size--;
+		}
+	}
+}
+
+uint32_t repo_copy(uint32_t revision, uint32_t *src, uint32_t *dst)
+{
+	uint32_t mode = 0, content_offset = 0;
+	struct repo_dirent *src_dirent;
+	src_dirent = repo_read_dirent(revision, src);
+	if (src_dirent != NULL) {
+		mode = src_dirent->mode;
+		content_offset = src_dirent->content_offset;
+		repo_write_dirent(dst, mode, content_offset, 0);
+	}
+	return mode;
+}
+
+void repo_add(uint32_t *path, uint32_t mode, uint32_t blob_mark)
+{
+	repo_write_dirent(path, mode, blob_mark, 0);
+}
+
+uint32_t repo_replace(uint32_t *path, uint32_t blob_mark)
+{
+	uint32_t mode = 0;
+	struct repo_dirent *src_dirent;
+	src_dirent = repo_read_dirent(active_commit, path);
+	if (src_dirent != NULL) {
+		mode = src_dirent->mode;
+		repo_write_dirent(path, mode, blob_mark, 0);
+	}
+	return mode;
+}
+
+void repo_modify(uint32_t *path, uint32_t mode, uint32_t blob_mark)
+{
+	struct repo_dirent *src_dirent;
+	src_dirent = repo_read_dirent(active_commit, path);
+	if (src_dirent != NULL && blob_mark == 0) {
+		blob_mark = src_dirent->content_offset;
+	}
+	repo_write_dirent(path, mode, blob_mark, 0);
+}
+
+void repo_delete(uint32_t *path)
+{
+	repo_write_dirent(path, 0, 0, 1);
+}
+
+static void repo_git_add_r(uint32_t depth, uint32_t *path, struct repo_dir *dir);
+
+static void repo_git_add(uint32_t depth, uint32_t *path, struct repo_dirent *dirent)
+{
+	if (repo_dirent_is_dir(dirent)) {
+		repo_git_add_r(depth, path, repo_dir_from_dirent(dirent));
+	} else {
+		fast_export_modify(depth, path, dirent->mode, dirent->content_offset);
+	}
+}
+
+static void repo_git_add_r(uint32_t depth, uint32_t *path, struct repo_dir *dir)
+{
+	uint32_t o;
+	struct repo_dirent *de;
+	de = repo_first_dirent(dir);
+	for (o = 0; o < dir->size; o++) {
+		path[depth] = de[o].name_offset;
+		repo_git_add(depth + 1, path, &de[o]);
+	}
+}
+
+static void repo_diff_r(uint32_t depth, uint32_t *path, struct repo_dir *dir1,
+			struct repo_dir *dir2)
+{
+	struct repo_dirent *de1, *de2, *max_de1, *max_de2;
+	de1 = repo_first_dirent(dir1);
+	de2 = repo_first_dirent(dir2);
+	max_de1 = &de1[dir1->size];
+	max_de2 = &de2[dir2->size];
+
+	while (de1 < max_de1 && de2 < max_de2) {
+		if (de1->name_offset < de2->name_offset) {
+			path[depth] = (de1++)->name_offset;
+			fast_export_delete(depth + 1, path);
+		} else if (de1->name_offset > de2->name_offset) {
+			path[depth] = de2->name_offset;
+			repo_git_add(depth + 1, path, de2++);
+		} else {
+			path[depth] = de1->name_offset;
+			if (de1->mode != de2->mode ||
+				de1->content_offset != de2->content_offset) {
+				if (repo_dirent_is_dir(de1) && repo_dirent_is_dir(de2)) {
+					repo_diff_r(depth + 1, path,
+								repo_dir_from_dirent(de1),
+								repo_dir_from_dirent(de2));
+				} else {
+					if (repo_dirent_is_dir(de1) != repo_dirent_is_dir(de2)) {
+						fast_export_delete(depth + 1, path);
+					}
+					repo_git_add(depth + 1, path, de2);
+				}
+			}
+			de1++;
+			de2++;
+		}
+	}
+	while (de1 < max_de1) {
+		path[depth] = (de1++)->name_offset;
+		fast_export_delete(depth + 1, path);
+	}
+	while (de2 < max_de2) {
+		path[depth] = de2->name_offset;
+		repo_git_add(depth + 1, path, de2++);
+	}
+}
+
+static uint32_t path_stack[REPO_MAX_PATH_DEPTH];
+
+void repo_diff(uint32_t r1, uint32_t r2)
+{
+	repo_diff_r(0,
+	            path_stack,
+	            repo_commit_root_dir(commit_pointer(r1)),
+	            repo_commit_root_dir(commit_pointer(r2)));
+}
+
+void repo_commit(uint32_t revision, uint32_t author, char *log, uint32_t uuid,
+                 uint32_t url, unsigned long timestamp)
+{
+	fast_export_commit(revision, author, log, uuid, url, timestamp);
+	num_dirs_saved = dir_pool.size;
+	num_dirents_saved = dirent_pool.size;
+	active_commit = commit_alloc(1);
+	commit_pointer(active_commit)->root_dir_offset =
+		commit_pointer(active_commit - 1)->root_dir_offset;
+}
+
+static void mark_init(void)
+{
+	uint32_t i;
+	_mark = 0;
+	for (i = 0; i < dirent_pool.size; i++)
+		if (!repo_dirent_is_dir(dirent_pointer(i)) &&
+			dirent_pointer(i)->content_offset > _mark)
+			_mark = dirent_pointer(i)->content_offset;
+	_mark++;
+}
+
+void repo_init() {
+	pool_init();
+	commit_init();
+	dir_init();
+	dirent_init();
+	mark_init();
+	num_dirs_saved = dir_pool.size;
+	num_dirents_saved = dirent_pool.size;
+	active_commit = commit_pool.size - 1;
+	if (active_commit == -1) {
+		commit_alloc(2);
+		/* Create empty tree for commit 0. */
+		commit_pointer(0)->root_dir_offset =
+			dir_with_dirents_alloc(0);
+		/* Preallocate commit 1, ready for changes. */
+		commit_pointer(1)->root_dir_offset =
+			commit_pointer(0)->root_dir_offset;
+		active_commit = 1;
+		num_dirs_saved = dir_pool.size;
+		num_dirents_saved = dirent_pool.size;
+	}
+}
+
+void repo_reset(void)
+{
+	pool_reset();
+	commit_reset();
+	dir_reset();
+	dirent_reset();
+}
diff --git a/vcs-svn/repo_tree.h b/vcs-svn/repo_tree.h
new file mode 100644
index 0000000..c49b65a
--- /dev/null
+++ b/vcs-svn/repo_tree.h
@@ -0,0 +1,27 @@
+#ifndef REPO_TREE_H_
+#define REPO_TREE_H_
+
+#include <stdint.h>
+#include <time.h>
+
+#define REPO_MODE_DIR 0040000
+#define REPO_MODE_BLB 0100644
+#define REPO_MODE_EXE 0100755
+#define REPO_MODE_LNK 0120000
+
+#define REPO_MAX_PATH_LEN 4096
+#define REPO_MAX_PATH_DEPTH 1000
+
+uint32_t next_blob_mark(void);
+uint32_t repo_copy(uint32_t revision, uint32_t *src, uint32_t *dst);
+void repo_add(uint32_t *path, uint32_t mode, uint32_t blob_mark);
+uint32_t repo_replace(uint32_t *path, uint32_t blob_mark);
+void repo_modify(uint32_t *path, uint32_t mode, uint32_t blob_mark);
+void repo_delete(uint32_t *path);
+void repo_commit(uint32_t revision, uint32_t author, char *log, uint32_t uuid,
+                 uint32_t url, long unsigned timestamp);
+void repo_diff(uint32_t r1, uint32_t r2);
+void repo_init(void);
+void repo_reset(void);
+
+#endif
-- 
1.7.1

[PATCH 4/6] Add stream helper library

From: Ramkumar Ramachandra <hidden>
Date: 2016-06-15 22:48:54

From: David Barr <redacted>

This library provides facilities to read streams into buffers. It
maintains a couple of static buffers and provides an API to use them.

Signed-off-by: David Barr <redacted>
Signed-off-by: Ramkumar Ramachandra <redacted>
---
 vcs-svn/line_buffer.c |  129 +++++++++++++++++++++++++++++++++++++++++++++++++
 vcs-svn/line_buffer.h |   14 +++++
 2 files changed, 143 insertions(+), 0 deletions(-)
 create mode 100644 vcs-svn/line_buffer.c
 create mode 100644 vcs-svn/line_buffer.h
diff --git a/vcs-svn/line_buffer.c b/vcs-svn/line_buffer.c
new file mode 100644
index 0000000..740676d
--- /dev/null
+++ b/vcs-svn/line_buffer.c
@@ -0,0 +1,129 @@
+#include "git-compat-util.h"
+
+#include "line_buffer.h"
+#include "obj_pool.h"
+
+#define LINE_BUFFER_LEN 10000
+#define COPY_BUFFER_LEN 4096
+
+/* Create memory pool for char sequence of known length */
+obj_pool_gen(blob, char, 4096);
+
+static char line_buffer[LINE_BUFFER_LEN];
+static char byte_buffer[COPY_BUFFER_LEN];
+static uint32_t line_buffer_len = 0;
+static uint32_t line_len = 0;
+static FILE *infile;
+
+int buffer_init(char *filename)
+{
+	infile = fopen(filename, "r");
+	if (!infile)
+		return 1;
+	return 0;
+}
+
+int buffer_deinit()
+{
+	fclose(infile);
+	return 0;
+}
+
+char *buffer_read_line(void)
+{
+	char *end;
+	uint32_t n_read;
+
+	if (line_len) {
+		memmove(line_buffer, &line_buffer[line_len],
+			line_buffer_len - line_len);
+		line_buffer_len -= line_len;
+		line_len = 0;
+	}
+
+	end = memchr(line_buffer, '\n', line_buffer_len);
+	while (line_buffer_len < LINE_BUFFER_LEN - 1 &&
+	       !feof(infile) && ferror(infile) && NULL == end) {
+		n_read = fread(&line_buffer[line_buffer_len], 1,
+			       LINE_BUFFER_LEN - 1 - line_buffer_len,
+			       infile);
+		end = memchr(&line_buffer[line_buffer_len], '\n', n_read);
+		line_buffer_len += n_read;
+	}
+
+	if (ferror(infile))
+		return NULL;
+
+	if (end != NULL) {
+		line_len = end - line_buffer;
+		line_buffer[line_len++] = '\0';
+	} else {
+		line_len = line_buffer_len;
+		line_buffer[line_buffer_len] = '\0';
+	}
+
+	if (line_len == 0)
+		return NULL;
+
+	return line_buffer;
+}
+
+char *buffer_read_string(uint32_t len)
+{
+	char *s;
+	blob_free(blob_pool.size);
+	s = blob_pointer(blob_alloc(len + 1));
+	uint32_t offset = 0;
+	if (line_buffer_len > line_len) {
+		offset = line_buffer_len - line_len;
+		if (offset > len)
+			offset = len;
+		memcpy(s, &line_buffer[line_len], offset);
+		line_len += offset;
+	}
+	if (offset < len)
+		offset += fread(&s[offset], 1, len - offset, infile);
+	s[offset] = '\0';
+	return s;
+}
+
+void buffer_copy_bytes(uint32_t len)
+{
+	uint32_t in;
+	if (line_buffer_len > line_len) {
+		in = line_buffer_len - line_len;
+		if (in > len)
+			in = len;
+		fwrite(&line_buffer[line_len], 1, in, stdout);
+		len -= in;
+		line_len += in;
+	}
+	while (len > 0 && !feof(infile)) {
+		in = len < COPY_BUFFER_LEN ? len : COPY_BUFFER_LEN;
+		in = fread(byte_buffer, 1, in, infile);
+		len -= in;
+		fwrite(byte_buffer, 1, in, stdout);
+	}
+}
+
+void buffer_skip_bytes(uint32_t len)
+{
+	uint32_t in;
+	if (line_buffer_len > line_len) {
+		in = line_buffer_len - line_len;
+		if (in > len)
+			in = len;
+		line_len += in;
+		len -= in;
+	}
+	while (len > 0 && !feof(infile) && !ferror(infile)) {
+		in = len < COPY_BUFFER_LEN ? len : COPY_BUFFER_LEN;
+		in = fread(byte_buffer, 1, in, infile);
+		len -= in;
+	}
+}
+
+void buffer_reset(void)
+{
+	blob_reset();
+}
diff --git a/vcs-svn/line_buffer.h b/vcs-svn/line_buffer.h
new file mode 100644
index 0000000..a6c42d7
--- /dev/null
+++ b/vcs-svn/line_buffer.h
@@ -0,0 +1,14 @@
+#ifndef LINE_BUFFER_H_
+#define LINE_BUFFER_H_
+
+#include <stdint.h>
+
+int buffer_init(char *filename);
+int buffer_deinit(void);
+char *buffer_read_line(void);
+char *buffer_read_string(uint32_t len);
+void buffer_copy_bytes(uint32_t len);
+void buffer_skip_bytes(uint32_t len);
+void buffer_reset(void);
+
+#endif
-- 
1.7.1

[PATCH 6/6] Add SVN dump parser

From: Ramkumar Ramachandra <hidden>
Date: 2016-06-15 22:48:54

From: David Barr <redacted>

svndump parses data that is in SVN dumpfile format produced by
`svnadmin dump` with the help of line_buffer, and uses repo_tree and
fast_export to emit a git fast-import stream.

Signed-off-by: David Barr <redacted>
Signed-off-by: Ramkumar Ramachandra <redacted>
---
 vcs-svn/svndump.c |  294 +++++++++++++++++++++++++++++++++++++++++++++++++++++
 vcs-svn/svndump.h |    7 ++
 2 files changed, 301 insertions(+), 0 deletions(-)
 create mode 100644 vcs-svn/svndump.c
 create mode 100644 vcs-svn/svndump.h
diff --git a/vcs-svn/svndump.c b/vcs-svn/svndump.c
new file mode 100644
index 0000000..9ee1246
--- /dev/null
+++ b/vcs-svn/svndump.c
@@ -0,0 +1,294 @@
+/*
+ * Parse and rearrange a svnadmin dump.
+ * Create the dump with:
+ * svnadmin dump --incremental -r<startrev>:<endrev> <repository> >outfile
+ */
+
+#include "cache.h"
+#include "git-compat-util.h"
+
+#include "repo_tree.h"
+#include "fast_export.h"
+#include "line_buffer.h"
+#include "obj_pool.h"
+#include "string_pool.h"
+
+#define NODEACT_REPLACE 4
+#define NODEACT_DELETE 3
+#define NODEACT_ADD 2
+#define NODEACT_CHANGE 1
+#define NODEACT_UNKNOWN 0
+
+#define DUMP_CTX 0
+#define REV_CTX  1
+#define NODE_CTX 2
+
+#define LENGTH_UNKNOWN (~0)
+#define DATE_RFC2822_LEN 31
+
+/* Create memory pool for log messages */
+obj_pool_gen(log, char, 4096);
+
+static char* log_copy(uint32_t length, char *log)
+{
+	char *buffer;
+	log_free(log_pool.size);
+	buffer = log_pointer(log_alloc(length));
+	strncpy(buffer, log, length);
+	return buffer;
+}
+
+static struct {
+	uint32_t action, propLength, textLength, srcRev, srcMode, mark, type;
+	uint32_t src[REPO_MAX_PATH_DEPTH], dst[REPO_MAX_PATH_DEPTH];
+} node_ctx;
+
+static struct {
+	uint32_t revision, author;
+	unsigned long timestamp;
+	char *log;
+} rev_ctx;
+
+static struct {
+	uint32_t uuid, url;
+} dump_ctx;
+
+static struct {
+	uint32_t svn_log, svn_author, svn_date, svn_executable, svn_special, uuid,
+		revision_number, node_path, node_kind, node_action,
+		node_copyfrom_path, node_copyfrom_rev, text_content_length,
+		prop_content_length, content_length;
+} keys;
+
+static void reset_node_ctx(char *fname)
+{
+	node_ctx.type = 0;
+	node_ctx.action = NODEACT_UNKNOWN;
+	node_ctx.propLength = LENGTH_UNKNOWN;
+	node_ctx.textLength = LENGTH_UNKNOWN;
+	node_ctx.src[0] = ~0;
+	node_ctx.srcRev = 0;
+	node_ctx.srcMode = 0;
+	pool_tok_seq(REPO_MAX_PATH_DEPTH, node_ctx.dst, "/", fname);
+	node_ctx.mark = 0;
+}
+
+static void reset_rev_ctx(uint32_t revision)
+{
+	rev_ctx.revision = revision;
+	rev_ctx.timestamp = "";
+	rev_ctx.log = NULL;
+	rev_ctx.author = ~0;
+}
+
+static void reset_dump_ctx(uint32_t url)
+{
+	dump_ctx.url = url;
+	dump_ctx.uuid = ~0;
+}
+
+static void init_keys(void)
+{
+	keys.svn_log = pool_intern("svn:log");
+	keys.svn_author = pool_intern("svn:author");
+	keys.svn_date = pool_intern("svn:date");
+	keys.svn_executable = pool_intern("svn:executable");
+	keys.svn_special = pool_intern("svn:special");
+	keys.uuid = pool_intern("UUID");
+	keys.revision_number = pool_intern("Revision-number");
+	keys.node_path = pool_intern("Node-path");
+	keys.node_kind = pool_intern("Node-kind");
+	keys.node_action = pool_intern("Node-action");
+	keys.node_copyfrom_path = pool_intern("Node-copyfrom-path");
+	keys.node_copyfrom_rev = pool_intern("Node-copyfrom-rev");
+	keys.text_content_length = pool_intern("Text-content-length");
+	keys.prop_content_length = pool_intern("Prop-content-length");
+	keys.content_length = pool_intern("Content-length");
+}
+
+static void read_props(void)
+{
+	uint32_t len;
+	uint32_t key = ~0;
+	char buffer[27];
+	char *val = NULL;
+	char *t;
+	while ((t = buffer_read_line()) && strcmp(t, "PROPS-END")) {
+		if (!strncmp(t, "K ", 2)) {
+			len = atoi(&t[2]);
+			key = pool_intern(buffer_read_string(len));
+			buffer_read_line();
+		} else if (!strncmp(t, "V ", 2)) {
+			len = atoi(&t[2]);
+			val = buffer_read_string(len);
+			if (key == keys.svn_log) {
+				/* Value length excludes terminating nul. */
+				rev_ctx.log = log_copy(len + 1, val);
+			} else if (key == keys.svn_author) {
+				rev_ctx.author = pool_intern(val);
+			} else if (key == keys.svn_date) {
+				if (parse_date(val, buffer, sizeof(buffer)) > 0)
+					rev_ctx.timestamp = strtoul(buffer, NULL, 0);
+				else
+					fprintf(stderr, "Invalid timestamp: %s", val);
+			} else if (key == keys.svn_executable) {
+				node_ctx.type = REPO_MODE_EXE;
+			} else if (key == keys.svn_special) {
+				node_ctx.type = REPO_MODE_LNK;
+			}
+			key = ~0;
+			buffer_read_line();
+		}
+	}
+}
+
+static void handle_node(void)
+{
+	if (node_ctx.propLength != LENGTH_UNKNOWN && node_ctx.propLength) {
+		read_props();
+	}
+
+	if (node_ctx.srcRev) {
+		node_ctx.srcMode = repo_copy(node_ctx.srcRev, node_ctx.src, node_ctx.dst);
+	}
+
+	if (node_ctx.textLength != LENGTH_UNKNOWN &&
+		node_ctx.type != REPO_MODE_DIR) {
+		node_ctx.mark = next_blob_mark();
+	}
+
+	if (node_ctx.action == NODEACT_DELETE) {
+		repo_delete(node_ctx.dst);
+	} else if (node_ctx.action == NODEACT_CHANGE ||
+			   node_ctx.action == NODEACT_REPLACE) {
+		if (node_ctx.action == NODEACT_REPLACE &&
+			node_ctx.type == REPO_MODE_DIR) {
+			repo_replace(node_ctx.dst, node_ctx.mark);
+		} else if (node_ctx.propLength != LENGTH_UNKNOWN ) {
+			repo_modify(node_ctx.dst, node_ctx.type, node_ctx.mark);
+		} else if (node_ctx.textLength != LENGTH_UNKNOWN) {
+			node_ctx.srcMode = repo_replace(node_ctx.dst, node_ctx.mark);
+		}
+	} else if (node_ctx.action == NODEACT_ADD) {
+		if (node_ctx.srcRev &&
+			node_ctx.propLength == LENGTH_UNKNOWN &&
+			node_ctx.textLength != LENGTH_UNKNOWN) {
+			node_ctx.srcMode = repo_replace(node_ctx.dst, node_ctx.mark);
+		} else if ((node_ctx.type == REPO_MODE_DIR && !node_ctx.srcRev) ||
+				   node_ctx.textLength != LENGTH_UNKNOWN){
+			repo_add(node_ctx.dst, node_ctx.type, node_ctx.mark);
+		}
+	}
+
+	if (node_ctx.propLength == LENGTH_UNKNOWN && node_ctx.srcMode) {
+		node_ctx.type = node_ctx.srcMode;
+	}
+
+	if (node_ctx.mark) {
+		fast_export_blob(node_ctx.type, node_ctx.mark, node_ctx.textLength);
+	} else if (node_ctx.textLength != LENGTH_UNKNOWN) {
+		buffer_skip_bytes(node_ctx.textLength);
+	}
+}
+
+static void handle_revision(void)
+{
+	if (rev_ctx.revision)
+		repo_commit(rev_ctx.revision, rev_ctx.author, rev_ctx.log,
+		            dump_ctx.uuid, dump_ctx.url, rev_ctx.timestamp);
+}
+
+void svndump_read(uint32_t url)
+{
+	char *val;
+	char *t;
+	uint32_t active_ctx = DUMP_CTX;
+	uint32_t len;
+	uint32_t key;
+
+	reset_dump_ctx(url);
+	while ((t = buffer_read_line())) {
+		val = strstr(t, ": ");
+		if (!val) continue;
+		*val++ = '\0';
+		*val++ = '\0';
+		key = pool_intern(t);
+
+		if(key == keys.uuid) {
+			dump_ctx.uuid = pool_intern(val);
+		} else if (key == keys.revision_number) {
+			if (active_ctx == NODE_CTX) handle_node();
+			if (active_ctx != DUMP_CTX) handle_revision();
+			active_ctx = REV_CTX;
+			reset_rev_ctx(atoi(val));
+		} else if (key == keys.node_path) {
+			if (active_ctx == NODE_CTX)
+				handle_node();
+			active_ctx = NODE_CTX;
+			reset_node_ctx(val);
+		} else if (key == keys.node_kind) {
+			if (!strcmp(val, "dir")) {
+				node_ctx.type = REPO_MODE_DIR;
+			} else if (!strcmp(val, "file")) {
+				node_ctx.type = REPO_MODE_BLB;
+			} else {
+				fprintf(stderr, "Unknown node-kind: %s\n", val);
+			}
+		} else if (key == keys.node_action) {
+			if (!strcmp(val, "delete")) {
+				node_ctx.action = NODEACT_DELETE;
+			} else if (!strcmp(val, "add")) {
+				node_ctx.action = NODEACT_ADD;
+			} else if (!strcmp(val, "change")) {
+				node_ctx.action = NODEACT_CHANGE;
+			} else if (!strcmp(val, "replace")) {
+				node_ctx.action = NODEACT_REPLACE;
+			} else {
+				fprintf(stderr, "Unknown node-action: %s\n", val);
+				node_ctx.action = NODEACT_UNKNOWN;
+			}
+		} else if (key == keys.node_copyfrom_path) {
+			pool_tok_seq(REPO_MAX_PATH_DEPTH, node_ctx.src, "/", val);
+		} else if (key == keys.node_copyfrom_rev) {
+			node_ctx.srcRev = atoi(val);
+		} else if (key == keys.text_content_length) {
+			node_ctx.textLength = atoi(val);
+		} else if (key == keys.prop_content_length) {
+			node_ctx.propLength = atoi(val);
+		} else if (key == keys.content_length) {
+			len = atoi(val);
+			buffer_read_line();
+			if (active_ctx == REV_CTX) {
+				read_props();
+			} else if (active_ctx == NODE_CTX) {
+				handle_node();
+				active_ctx = REV_CTX;
+			} else {
+				fprintf(stderr, "Unexpected content length header: %d\n", len);
+				buffer_skip_bytes(len);
+			}
+		}
+	}
+	if (active_ctx == NODE_CTX) handle_node();
+	if (active_ctx != DUMP_CTX) handle_revision();
+}
+
+static void svndump_init(void)
+{
+	log_init();
+	repo_init();
+	reset_dump_ctx(~0);
+	reset_rev_ctx(0);
+	reset_node_ctx(NULL);
+	init_keys();
+}
+
+void svndump_reset(void)
+{
+	log_reset();
+	buffer_reset();
+	repo_reset();
+	reset_dump_ctx(~0);
+	reset_rev_ctx(0);
+	reset_node_ctx(NULL);
+}
diff --git a/vcs-svn/svndump.h b/vcs-svn/svndump.h
new file mode 100644
index 0000000..e205f1f
--- /dev/null
+++ b/vcs-svn/svndump.h
@@ -0,0 +1,7 @@
+#ifndef SVNDUMP_H_
+#define SVNDUMP_H_
+
+void svndump_read(char *url);
+void svndump_reset(void);
+
+#endif
-- 
1.7.1

Re: [PATCH 1/6] Add memory pool library

From: Jonathan Nieder <hidden>
Date: 2016-06-15 22:48:54

Ramkumar Ramachandra wrote:
Add a memory pool library implemented using cpp macros. The library
provides macros that can be used to create a type-specific memory pool
API.
More details:

  The memory pool library is distinguished from the existing specialized
  allocators in alloc.c by using a contiguous block for all allocations.
  This means that on one hand, long-lived pointers have to be written as
  offsets, since the base address changes as the pool grows, but on the
  other hand, the entire pool can be easily written to the file system.
  This allows the memory pool to persist between runs of an application.
  
  For svn-fe, such a facility is useful because each svn revision can
  copy trees and files from any previous revision.  Therefore the
  relevant information for all revisions has to persist somehow to
  support incremental runs.
  
  The current implementation is backed by the file system using mmap().

Thanks for the explanations, David.
+#define obj_pool_gen(pre, obj_t, initial_capacity) \
+static struct { \
+	uint32_t size; \
+	uint32_t capacity; \
+	obj_t *base; \
+        FILE *file; \
Whitespace damaged.
+} pre##_pool = { 0, 0, NULL, NULL}; \
+static void pre##_init(void) \
+{ \
+	struct stat st; \
+	size_t ps = sysconf (_SC_PAGESIZE); \
Probably an ignorant question, but why?  I do not think the win32 mmap
emulation in git currently supports sysconf().
+	/* Touch binary file before opening read/write */ \
+	pre##_pool.file = fopen(#pre ".bin", "a"); \
+	fclose(pre##_pool.file); \
+	/* Open, check size, compute capacity */ \
+	pre##_pool.file = fopen(#pre ".bin", "r+"); \
Maybe

	..._pool.file = fopen(..., "a+");
	rewind(..._pool.file);

would be simpler.  But given that there is no buffered I/O going on
for this file, why not just

	..._pool.file = open(..., O_RDWR);

?
+	/* Truncate to calculated capacity and map to VM */ \
+	ftruncate(fileno(pre##_pool.file), pre##_pool.capacity * sizeof(obj_t)); \
Reasonable.
+	pre##_pool.base = mmap(0, pre##_pool.capacity * sizeof(obj_t), \
+				PROT_READ | PROT_WRITE, MAP_SHARED, \
+				fileno(pre##_pool.file), 0); \
This is the first use of MAP_SHARED in git.  I wouldn’t be surprised if
the win32 mmap emulation does not support it.
+} \
+static uint32_t pre##_alloc(uint32_t count) \
+{ \
+	uint32_t offset; \
+	if (pre##_pool.size + count > pre##_pool.capacity) { \
+		if (NULL == pre##_pool.base) \
+			pre##_init(); \
+		fsync(fileno(pre##_pool.file)); \
Necessary?  (Maybe yes: we are about to truncate the file, so the
fsync() may be intended to force the data to be committed before
the metadata.)  Portable?  (I suspect the fsync() should go after
the munmap().)

[...]
+static void pre##_reset(void) \
+{ \
+	if (pre##_pool.base) { \
+		fsync(fileno(pre##_pool.file)); \
Same question.
+		munmap(pre##_pool.base, \
+			pre##_pool.capacity * sizeof(obj_t)); \
+		ftruncate(fileno(pre##_pool.file), \
+				pre##_pool.size * sizeof(obj_t)); \
+		fclose(pre##_pool.file); \
Because of these portability concerns, I’d rather use the old version
until the incremental parser is ready.

Hope that helps,
Jonathan

Re: [PATCH 4/6] Add stream helper library

From: Jonathan Nieder <hidden>
Date: 2016-06-15 22:48:54

Ramkumar Ramachandra wrote:
+static uint32_t line_buffer_len = 0;
+static uint32_t line_len = 0;
I still don’t understand why this needs to read ahead.

I can take a stab at cleaning it up...
quoted hunk
--- /dev/null
+++ b/vcs-svn/line_buffer.h
@@ -0,0 +1,14 @@
+#ifndef LINE_BUFFER_H_
+#define LINE_BUFFER_H_
+
+#include <stdint.h>
Should use git-compat-util.h instead.

Jonathan

Re: [PATCH 5/6] Add infrastructure to write revisions in fast-export format

From: Jonathan Nieder <hidden>
Date: 2016-06-15 22:48:54

Hi Ram,

Thanks for the style cleanup.

Ramkumar Ramachandra wrote:
quoted hunk
--- /dev/null
+++ b/vcs-svn/repo_tree.c
[...]
+static struct repo_dir *repo_clone_dir(struct repo_dir *orig_dir, uint32_t padding)
+{
+	uint32_t orig_o, new_o, dirent_o;
+	orig_o = dir_offset(orig_dir);
+	if (orig_o < num_dirs_saved) {
+		new_o = dir_with_dirents_alloc(orig_dir->size + padding);
+		orig_dir = dir_pointer(orig_o);
+		dirent_o = dir_pointer(new_o)->first_offset;
+	} else {
+		if (padding == 0)
+			return orig_dir;
+		new_o = orig_o;
+		dirent_o = dirent_alloc(orig_dir->size + padding);
+	}
+	memcpy(dirent_pointer(dirent_o), repo_first_dirent(orig_dir),
+		   orig_dir->size * sizeof(struct repo_dirent));
+	dir_pointer(new_o)->size = orig_dir->size + padding;
+	dir_pointer(new_o)->first_offset = dirent_o;
+	return dir_pointer(new_o);
+}
This function expands a list of directory entries, copying it if
already committed as part of a previous revision.

The big downside is that as David mentioned it does not scale well with
the size of the directory and number of expansions.  But a patch is in
the pipeline to fix that.  I do not think it should hold the series up.
+
+static struct repo_dirent *repo_read_dirent(uint32_t revision, uint32_t *path)
+{
+	uint32_t name = 0;
+	struct repo_dir *dir = NULL;
+	struct repo_dirent *dirent = NULL;
+	dir = repo_commit_root_dir(commit_pointer(revision));
+	while (~(name = *path++)) {
+		dirent = repo_dirent_by_name(dir, name);
+		if (dirent == NULL) {
+			return NULL;
+		} else if (repo_dirent_is_dir(dirent)) {
+			dir = repo_dir_from_dirent(dirent);
+		} else {
+			break;
+		}
Style: would probably be clearer to write:

	while (~(name = *path++)) {
		dirent = repo_dirent_by_name(dir, name);
		if (!dirent || !repo_dirent_is_dir(dirent))
			break;
		dir = repo_dir_from_dirent(dirent);
	}

i.e., fewer unnecessary braces, and dealing with the exceptional cases
separately from the normal case.

As before, I wonder about the error cases.  Might it make sense to
report the error if someone tries to copy a nonexistent directory
from a previous revision?
+static void
+repo_write_dirent(uint32_t *path, uint32_t mode, uint32_t content_offset,
+                  uint32_t del)
Function is too long for my taste (I realize this is a matter of
taste).  The innermost blocks would make sense as functions in their
own right.

[...]
+static void repo_diff_r(uint32_t depth, uint32_t *path, struct repo_dir *dir1,
+			struct repo_dir *dir2)
[...]
+	while (de1 < max_de1 && de2 < max_de2) {
+		if (de1->name_offset < de2->name_offset) {
+			path[depth] = (de1++)->name_offset;
+			fast_export_delete(depth + 1, path);
+		} else if (de1->name_offset > de2->name_offset) {
+			path[depth] = de2->name_offset;
+			repo_git_add(depth + 1, path, de2++);
+		} else {
+			path[depth] = de1->name_offset;
+			if (de1->mode != de2->mode ||
+				de1->content_offset != de2->content_offset) {
+				if (repo_dirent_is_dir(de1) && repo_dirent_is_dir(de2)) {
My 80-column terminal is suffering.  Why not use the common

	while (...) {
		if (exceptional case) {
			....
			continue;
		} else if (exceptional case) {
			...
			continue;
		}
		path[depth] = de1->name_offset;
		if (easy case) {
			...
			continue;
		}
		usual case
	}

pattern?
+#define REPO_MAX_PATH_LEN 4096
+#define REPO_MAX_PATH_DEPTH 1000
These limits are not checked; is the caller supposed to check them
itself?  Does svn obey them?

Jonathan

Re: [PATCH 5/6] Add infrastructure to write revisions in fast-export format

From: Ramkumar Ramachandra <hidden>
Date: 2016-06-15 22:48:55

Hi Jonathan,

Jonathan Nieder wrote:
The big downside is that as David mentioned it does not scale well with
the size of the directory and number of expansions.  But a patch is in
the pipeline to fix that.  I do not think it should hold the series up.
Right. The `dirents` branch is now merged.
Style: would probably be clearer to write:

       while (~(name = *path++)) {
               dirent = repo_dirent_by_name(dir, name);
               if (!dirent || !repo_dirent_is_dir(dirent))
                       break;
               dir = repo_dir_from_dirent(dirent);
       }

i.e., fewer unnecessary braces, and dealing with the exceptional cases
separately from the normal case.
This was (probably unintentionally) re-factored by David when merging
in his `dirents` branch.
As before, I wonder about the error cases.  Might it make sense to
report the error if someone tries to copy a nonexistent directory
from a previous revision?
Function is too long for my taste (I realize this is a matter of
taste).  The innermost blocks would make sense as functions in their
own right.
This came up in your previous review- I tried to split it up into more
functions, but due to the number of local variables, it turned out to
be a mess and I had to revert. Would you like to try re-factoring it?
My 80-column terminal is suffering.  Why not use the common
pattern?
Re-factored (again, probably unintentionally) by David during the merge.
These limits are not checked; is the caller supposed to check them
itself?  Does svn obey them?
I asked David too, and as far as we know, these limits are pretty
arbitrary. They're no enforced by SVN or any specific filesystem. We
can discuss about having less arbitrary limits and checking it in
svndump.c (while parsing the dump).

-- Ram

Re: [PATCH 1/6] Add memory pool library

From: Ramkumar Ramachandra <hidden>
Date: 2016-06-15 22:48:55

Hi Jonathan,

Jonathan Nieder wrote:
 The memory pool library is distinguished from the existing specialized
 allocators in alloc.c by using a contiguous block for all allocations.
 This means that on one hand, long-lived pointers have to be written as
 offsets, since the base address changes as the pool grows, but on the
 other hand, the entire pool can be easily written to the file system.
 This allows the memory pool to persist between runs of an application.

 For svn-fe, such a facility is useful because each svn revision can
 copy trees and files from any previous revision.  Therefore the
 relevant information for all revisions has to persist somehow to
 support incremental runs.

 The current implementation is backed by the file system using mmap().

Thanks for the explanations, David.
Thanks. Added to commit message: should be fine the next time I post the series.
Whitespace damaged.
Fixed in b180ad7 (pushed few seconds ago).
Probably an ignorant question, but why?  I do not think the win32 mmap
emulation in git currently supports sysconf().
I'm not sure why the pool capacity should be dependent on the page size.
       ..._pool.file = open(..., O_RDWR);
Fixed in b180ad7 (pushed few seconds ago).
This is the first use of MAP_SHARED in git.  I wouldn’t be surprised if
the win32 mmap emulation does not support it.
You're right; I just checked. compat/mmap.c:7 clearly specifies that
only MAP_PRIVATE is supported.
	if (start != NULL || !(flags & MAP_PRIVATE))
		die("Invalid usage of mmap when built with NO_MMAP");

What can we do about this? Should we attempt to implement MAP_SHARED?
Necessary?  (Maybe yes: we are about to truncate the file, so the
fsync() may be intended to force the data to be committed before
the metadata.)  Portable?  (I suspect the fsync() should go after
the munmap().)
You want to flush changes *after* unmapping the pages of memory from
the file? I can't find any notes on portability relating to this- can
you clarify?
Because of these portability concerns, I’d rather use the old version
until the incremental parser is ready.
Rolling back that many changes can be more painful than actually
fixing these few concerns (especially large number of change after
`dirents` was merged in). It means more delay in getting the exporter
merged.

-- Ram

Re: [PATCH 5/6] Add infrastructure to write revisions in fast-export format

From: David Michael Barr <hidden>
Date: 2016-06-15 22:48:55

Hi Ram,
quoted
Style: would probably be clearer to write:

       while (~(name = *path++)) {
               dirent = repo_dirent_by_name(dir, name);
               if (!dirent || !repo_dirent_is_dir(dirent))
                       break;
               dir = repo_dir_from_dirent(dirent);
       }

i.e., fewer unnecessary braces, and dealing with the exceptional cases
separately from the normal case.
This was (probably unintentionally) re-factored by David when merging
in his `dirents` branch.
Oh no, it was intentional.
quoted
My 80-column terminal is suffering.  Why not use the common
pattern?
Re-factored (again, probably unintentionally) by David during the merge.
As above, totally intentional.

@Jonathan: Thanks for the review, I took it into account when refactoring dirents.
quoted
These limits are not checked; is the caller supposed to check them
itself?  Does svn obey them?
I asked David too, and as far as we know, these limits are pretty
arbitrary. They're no enforced by SVN or any specific filesystem. We
can discuss about having less arbitrary limits and checking it in
svndump.c (while parsing the dump).
I am indeed in favour of less arbitrary limits.
The current magic numbers were chosen as common practical bounds.

Re: [PATCH 1/6] Add memory pool library

From: Erik Faye-Lund <hidden>
Date: 2016-06-15 22:48:55

On Mon, Jun 7, 2010 at 3:28 PM, Ramkumar Ramachandra [off-list ref] wrote:
quoted
Probably an ignorant question, but why?  I do not think the win32 mmap
emulation in git currently supports sysconf().
I'm not sure why the pool capacity should be dependent on the page size.
My guess would be that it's a performance-thing.

Adding a win32-implementation of sysconf for this purpose isn't
tricky, but perhaps it's a little bit overkill. Can't we just use some
large power of two number? Hopefully, we could chose a size that is a
multiple of the page-size on all systems...

Anyway, this is roughly what a Windows-implementation of sysconf would
look like (not tested - only checked that it compiles):
diff --git a/compat/mingw.c b/compat/mingw.c
index f90a114..526f0f9 100644
--- a/compat/mingw.c
+++ b/compat/mingw.c
@@ -1473,3 +1473,13 @@ struct dirent *mingw_readdir(DIR *dir)
 	return (struct dirent*)&dir->dd_dir;
 }
 #endif // !NO_MINGW_REPLACE_READDIR
+
+long mingw_sysconf(int name)
+{
+	SYSTEM_INFO si;
+	if (name != _SC_PAGESIZE)
+		return errno = EINVAL,
+		    error("sysconf emulation only supports _SC_PAGESIZE");
+	GetSystemInfo(&si);
+	return si.dwPageSize;
+}
diff --git a/compat/mingw.h b/compat/mingw.h
index 6c6cbda..db28b0d 100644
--- a/compat/mingw.h
+++ b/compat/mingw.h
@@ -322,3 +322,7 @@ struct dirent *mingw_readdir(DIR *dir);
  * Used by Pthread API implementation for Windows
  */
 extern int err_win_to_posix(DWORD winerr);
+
+#define _SC_PAGESIZE 0
+long mingw_sysconf(int name);
+#define sysconf mingw_sysconf
-- 
Erik "kusma" Faye-Lund

Re: [PATCH 1/6] Add memory pool library

From: Ramkumar Ramachandra <hidden>
Date: 2016-06-15 22:48:55

Hi Erik,

Erik Faye-Lund wrote:
On Mon, Jun 7, 2010 at 3:28 PM, Ramkumar Ramachandra [off-list ref] wrote:
Adding a win32-implementation of sysconf for this purpose isn't
tricky, but perhaps it's a little bit overkill. Can't we just use some
large power of two number? Hopefully, we could chose a size that is a
multiple of the page-size on all systems...
Until an alternative is ready, I'll just use a large power of two and
roll out another revision.
Anyway, this is roughly what a Windows-implementation of sysconf would
look like (not tested - only checked that it compiles):
Thanks for implementing this. Unfortunately, I don't have a Windows
box to test this- why don't you send a patch to the list and sign it
off? While I agree that just implementing sysconf for _SC_PAGESIZE is
an overkill, others can test and extend the code.

-- Ram
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help