mercurial/mpatch.c
author mpm@selenic.com
Sun, 21 Aug 2005 21:59:55 -0700
changeset 990 5007e0bdeed2
parent 597 e530637ea060
child 1722 681c5c211b92
permissions -rw-r--r--
Fix long-standing excessive file merges Since switching to the multihead approach, we've been creating excessive file-level merges where files are marked as merged with their ancestors. This explicitly checks at commit time whether the two parent versions are linearly related, and if so, reduces the file check-in to a non-merge. Then the file is compared against the remaining parent, and, if equal, skips check-in of that file (as it's not changed). Since we're not checking in all files that were different between versions, we no longer need to mark so many files for merge. This removes most of the 'm' state marking as well. Finally, it is possible to do a tree-level merge with no file-level changes. This will happen if one user changes file A and another changes file B. Thus, if we have have two parents, we allow commit to proceed even if there are no file-level changes.

/*
 mpatch.c - efficient binary patching for Mercurial

 This implements a patch algorithm that's O(m + nlog n) where m is the
 size of the output and n is the number of patches.

 Given a list of binary patches, it unpacks each into a hunk list,
 then combines the hunk lists with a treewise recursion to form a
 single hunk list. This hunk list is then applied to the original
 text.

 The text (or binary) fragments are copied directly from their source
 Python objects into a preallocated output string to avoid the
 allocation of intermediate Python objects. Working memory is about 2x
 the total number of hunks.

 Copyright 2005 Matt Mackall <mpm@selenic.com>

 This software may be used and distributed according to the terms
 of the GNU General Public License, incorporated herein by reference.
*/

#include <Python.h>
#include <stdlib.h>
#include <string.h>
#ifdef _WIN32
#ifdef _MSC_VER
#define inline __inline
typedef unsigned long uint32_t;
#else
#include <stdint.h>
#endif
static uint32_t ntohl(uint32_t x)
{
	return ((x & 0x000000ffUL) << 24) |
		((x & 0x0000ff00UL) <<  8) |
		((x & 0x00ff0000UL) >>  8) |
		((x & 0xff000000UL) >> 24);
}
#else
#include <sys/types.h>
#include <arpa/inet.h>
#endif

static char mpatch_doc[] = "Efficient binary patching.";

struct frag {
	int start, end, len;
	char *data;
};

struct flist {
	struct frag *base, *head, *tail;
};

static struct flist *lalloc(int size)
{
	struct flist *a = NULL;

	a = malloc(sizeof(struct flist));
	if (a) {
		a->base = malloc(sizeof(struct frag) * size);
		if (!a->base) {
			free(a);
			a = NULL;
		} else
			a->head = a->tail = a->base;
	}
	return a;
}

static void lfree(struct flist *a)
{
	if (a) {
		free(a->base);
		free(a);
	}
}

static int lsize(struct flist *a)
{
	return a->tail - a->head;
}

/* move hunks in source that are less cut to dest, compensating
   for changes in offset. the last hunk may be split if necessary.
*/
static int gather(struct flist *dest, struct flist *src, int cut, int offset)
{
	struct frag *d = dest->tail, *s = src->head;
	int postend, c, l;

	while (s != src->tail) {
		if (s->start + offset >= cut)
			break; /* we've gone far enough */

		postend = offset + s->start + s->len;
		if (postend <= cut) {
			/* save this hunk */
			offset += s->start + s->len - s->end;
			*d++ = *s++;
		}
		else {
			/* break up this hunk */
			c = cut - offset;
			if (s->end < c)
				c = s->end;
			l = cut - offset - s->start;
			if (s->len < l)
				l = s->len;

			offset += s->start + l - c;

			d->start = s->start;
			d->end = c;
			d->len = l;
			d->data = s->data;
			d++;
			s->start = c;
			s->len = s->len - l;
			s->data = s->data + l;

			break;
		}
	}

	dest->tail = d;
	src->head = s;
	return offset;
}

/* like gather, but with no output list */
static int discard(struct flist *src, int cut, int offset)
{
	struct frag *s = src->head;
	int postend, c, l;

	while (s != src->tail) {
		if (s->start + offset >= cut)
			break;

		postend = offset + s->start + s->len;
		if (postend <= cut) {
			offset += s->start + s->len - s->end;
			s++;
		}
		else {
			c = cut - offset;
			if (s->end < c)
				c = s->end;
			l = cut - offset - s->start;
			if (s->len < l)
				l = s->len;

			offset += s->start + l - c;
			s->start = c;
			s->len = s->len - l;
			s->data = s->data + l;

			break;
		}
	}

	src->head = s;
	return offset;
}

/* combine hunk lists a and b, while adjusting b for offset changes in a/
   this deletes a and b and returns the resultant list. */
static struct flist *combine(struct flist *a, struct flist *b)
{
	struct flist *c = NULL;
	struct frag *bh, *ct;
	int offset = 0, post;

	if (a && b)
		c = lalloc((lsize(a) + lsize(b)) * 2);

	if (c) {

		for (bh = b->head; bh != b->tail; bh++) {
			/* save old hunks */
			offset = gather(c, a, bh->start, offset);

			/* discard replaced hunks */
			post = discard(a, bh->end, offset);

			/* insert new hunk */
			ct = c->tail;
			ct->start = bh->start - offset;
			ct->end = bh->end - post;
			ct->len = bh->len;
			ct->data = bh->data;
			c->tail++;
			offset = post;
		}

		/* hold on to tail from a */
		memcpy(c->tail, a->head, sizeof(struct frag) * lsize(a));
		c->tail += lsize(a);
	}

	lfree(a);
	lfree(b);
	return c;
}

/* decode a binary patch into a hunk list */
static struct flist *decode(char *bin, int len)
{
	struct flist *l;
	struct frag *lt;
	char *end = bin + len;
	char decode[12]; /* for dealing with alignment issues */

	/* assume worst case size, we won't have many of these lists */
	l = lalloc(len / 12);
	lt = l->tail;

	while (bin < end) {
		memcpy(decode, bin, 12);
		lt->start = ntohl(*(uint32_t *)decode);
		lt->end = ntohl(*(uint32_t *)(decode + 4));
		lt->len = ntohl(*(uint32_t *)(decode + 8));
		lt->data = bin + 12;
		bin += 12 + lt->len;
		lt++;
	}

	l->tail = lt;
	return l;
}

/* calculate the size of resultant text */
static int calcsize(int len, struct flist *l)
{
	int outlen = 0, last = 0;
	struct frag *f = l->head;

	while (f != l->tail) {
		outlen += f->start - last;
		last = f->end;
		outlen += f->len;
		f++;
	}

	outlen += len - last;
	return outlen;
}

static void apply(char *buf, char *orig, int len, struct flist *l)
{
	struct frag *f = l->head;
	int last = 0;
	char *p = buf;

	while (f != l->tail) {
		memcpy(p, orig + last, f->start - last);
		p += f->start - last;
		memcpy(p, f->data, f->len);
		last = f->end;
		p += f->len;
		f++;
	}
	memcpy(p, orig + last, len - last);
}

/* recursively generate a patch of all bins between start and end */
static struct flist *fold(PyObject *bins, int start, int end)
{
	int len;

	if (start + 1 == end) {
		/* trivial case, output a decoded list */
		PyObject *tmp = PyList_GetItem(bins, start);
		if (!tmp)
			return NULL;
		return decode(PyString_AsString(tmp), PyString_Size(tmp));
	}

	/* divide and conquer, memory management is elsewhere */
	len = (end - start) / 2;
	return combine(fold(bins, start, start + len),
		       fold(bins, start + len, end));
}

static PyObject *
patches(PyObject *self, PyObject *args)
{
	PyObject *text, *bins, *result;
	struct flist *patch;
	char *in, *out;
	int len, outlen;

	if (!PyArg_ParseTuple(args, "SO:mpatch", &text, &bins))
		return NULL;

	len = PyList_Size(bins);
	if (!len) {
		/* nothing to do */
		Py_INCREF(text);
		return text;
	}

	patch = fold(bins, 0, len);
	if (!patch)
		return PyErr_NoMemory();

	outlen = calcsize(PyString_Size(text), patch);
	result = PyString_FromStringAndSize(NULL, outlen);
	if (result) {
		in = PyString_AsString(text);
		out = PyString_AsString(result);
		apply(out, in, PyString_Size(text), patch);
	}

	lfree(patch);
	return result;
}

static PyMethodDef methods[] = {
	{"patches", patches, METH_VARARGS, "apply a series of patches\n"},
	{NULL, NULL}
};

PyMODINIT_FUNC
initmpatch(void)
{
	Py_InitModule3("mpatch", methods, mpatch_doc);
}