shithub: diff

Download patch

ref: 33d0b194f3b0df7b830e643c5a9847b6f9a1f74c
author: Ori Bernstein <[email protected]>
date: Sat Feb 11 18:01:57 EST 2023 

diff: refactor to use struct


--- /dev/null
+++ b/diff.c
@@ -1,0 +1,261 @@
+#include <u.h>
+#include <libc.h>
+#include <bio.h>
+#include "diff.h"
+
+#define	DIRECTORY(s)		((s)->qid.type&QTDIR)
+#define	REGULAR_FILE(s)		((s)->type == 'M' && !DIRECTORY(s))
+
+Biobuf	stdout;
+char	mode;			/* '\0', 'e', 'f', 'h' */
+char	bflag;			/* ignore multiple and trailing blanks */
+char	rflag;			/* recurse down directory trees */
+char	mflag;			/* pseudo flag: doing multiple files, one dir */
+int	anychange;
+
+static char *tmp[] = {"/tmp/diff1XXXXXXXXXXX", "/tmp/diff2XXXXXXXXXXX"};
+static int whichtmp;
+
+static void
+rmtmpfiles(void)
+{
+	while (whichtmp > 0) {
+		whichtmp--;
+		remove(tmp[whichtmp]);
+	}
+}
+
+void	
+done(int status)
+{
+	rmtmpfiles();
+	switch(status)
+	{
+	case 0:
+		exits("");
+	case 1:
+		exits("some");
+	default:
+		exits("error");
+	}
+	/*NOTREACHED*/
+}
+
+void
+panic(int status, char *fmt, ...)
+{
+	va_list arg;
+
+	Bflush(&stdout);
+
+	fprint(2, "%s: ", argv0);
+	va_start(arg, fmt);
+	vfprint(2, fmt, arg);
+	va_end(arg);
+	if (status)
+		done(status);
+		/*NOTREACHED*/
+}
+
+static int
+catch(void *a, char *msg)
+{
+	USED(a);
+	panic(2, msg);
+	return 1;
+}
+
+int
+mkpathname(char *pathname, char *path, char *name)
+{
+	if (strlen(path) + strlen(name) > MAXPATHLEN) {
+		panic(0, "pathname %s/%s too long\n", path, name);
+		return 1;
+	}
+	sprint(pathname, "%s/%s", path, name);
+	return 0;
+}
+	
+static char *
+mktmpfile(int input, Dir **sb)
+{
+	int fd, i;
+	char *p;
+	char buf[8192];
+
+	atnotify(catch, 1);
+	p = mktemp(tmp[whichtmp++]);
+	fd = create(p, OWRITE, 0600);
+	if (fd < 0) {
+		panic(mflag ? 0: 2, "cannot create %s: %r\n", p);
+		return 0;
+	}
+	while ((i = read(input, buf, sizeof(buf))) > 0) {
+		if ((i = write(fd, buf, i)) < 0)
+			break;
+	}
+	*sb = dirfstat(fd);
+	close(fd);
+	if (i < 0) {
+		panic(mflag ? 0: 2, "cannot read/write %s: %r\n", p);
+		return 0;
+	}
+	return p;
+}
+
+static char *
+statfile(char *file, Dir **sb)
+{
+	Dir *dir;
+	int input;
+
+	dir = dirstat(file);
+	if(dir == nil) {
+		if (strcmp(file, "-") || (dir = dirfstat(0)) == nil) {
+			panic(mflag ? 0: 2, "cannot stat %s: %r\n", file);
+			return 0;
+		}
+		free(dir);
+		return mktmpfile(0, sb);
+	} else if (!REGULAR_FILE(dir) && !DIRECTORY(dir)) {
+		free(dir);
+		if ((input = open(file, OREAD)) == -1) {
+			panic(mflag ? 0: 2, "cannot open %s: %r\n", file);
+			return 0;
+		}
+		file = mktmpfile(input, sb);
+		close(input);
+	} else
+		*sb = dir;
+	return file;
+}
+
+void
+diff(char *f, char *t, int level)
+{
+	char *fp, *tp, *p, fb[MAXPATHLEN+1], tb[MAXPATHLEN+1];
+	Dir *fsb, *tsb;
+
+	if ((fp = statfile(f, &fsb)) == 0)
+		goto Return;
+	if ((tp = statfile(t, &tsb)) == 0){
+		free(fsb);
+		goto Return;
+	}
+	if (DIRECTORY(fsb) && DIRECTORY(tsb)) {
+		if (rflag || level == 0)
+			diffdir(fp, tp, level);
+		else
+			Bprint(&stdout, "Common subdirectories: %s and %s\n", fp, tp);
+	}
+	else if (REGULAR_FILE(fsb) && REGULAR_FILE(tsb))
+		diffreg(fp, f, tp, t);
+	else {
+		if (REGULAR_FILE(fsb)) {
+			if ((p = utfrrune(f, '/')) == 0)
+				p = f;
+			else
+				p++;
+			if (mkpathname(tb, tp, p) == 0)
+				diffreg(fp, f, tb, t);
+		} else {
+			if ((p = utfrrune(t, '/')) == 0)
+				p = t;
+			else
+				p++;
+			if (mkpathname(fb, fp, p) == 0)
+				diffreg(fb, f, tp, t);
+		}
+	}
+	free(fsb);
+	free(tsb);
+Return:
+	rmtmpfiles();
+}
+
+void
+usage(void)
+{
+	fprint(2, "usage: %s [-abcefmnrw] file1 ... file2\n", argv0);
+	exits("usage");
+}
+
+void
+main(int argc, char *argv[])
+{
+	int i;
+	Dir *fsb, *tsb;
+
+	Binit(&stdout, 1, OWRITE);
+	ARGBEGIN{
+	case 'e':
+	case 'f':
+	case 'n':
+	case 'c':
+	case 'a':
+	case 'u':
+		mode = ARGC();
+		break;
+	case 'w':
+		bflag = 2;
+		break;
+
+	case 'b':
+		bflag = 1;
+		break;
+
+	case 'r':
+		rflag = 1;
+		break;
+
+	case 'm':
+		mflag = 1;	
+		break;
+
+	case 'h':
+	default:
+		usage();
+	}ARGEND;
+	if (argc < 2)
+		usage();
+	if ((tsb = dirstat(argv[argc-1])) == nil)
+		panic(2, "can't stat %s\n", argv[argc-1]);
+	if (argc > 2) {
+		if (!DIRECTORY(tsb))
+			panic(2, "not directory: %s", argv[argc-1]);
+		mflag = 1;
+	} else {
+		if ((fsb = dirstat(argv[0])) == nil)
+			panic(2, "can't stat %s\n", argv[0]);
+		if (DIRECTORY(fsb) && DIRECTORY(tsb))
+			mflag = 1;
+		free(fsb);
+	}
+	free(tsb);
+	for (i = 0; i < argc-1; i++)
+		diff(argv[i], argv[argc-1], 0);
+	done(anychange);
+	/*NOTREACHED*/
+}
+
+static char noroom[] = "out of memory - try diff -h\n";
+
+void *
+emalloc(unsigned n)
+{
+	register void *p;
+
+	if ((p = malloc(n)) == 0)
+		panic(2, noroom);
+	return p;
+}
+
+void *
+erealloc(void *p, unsigned n)
+{
+	void *rp;
+
+	if ((rp = realloc(p, n)) == 0)
+		panic(2, noroom);
+	return rp;
+}
--- /dev/null
+++ b/diff.h
@@ -1,0 +1,69 @@
+typedef struct Line	Line;
+typedef struct Cand	Cand;
+typedef struct Diff	Diff;
+typedef struct Change	Change;
+
+struct Line {
+	int	serial;
+	int	value;
+};
+
+struct Cand {
+	int x;
+	int y;
+	int pred;
+};
+
+struct Change
+{
+	int a;
+	int b;
+	int c;
+	int d;
+};
+
+struct Diff {
+	Cand cand;
+	Line *file[2], line;
+	int len[2];
+	int binary;
+	Line *sfile[2];	/*shortened by pruning common prefix and suffix*/
+	int slen[2];
+	int pref, suff;	/*length of prefix and suffix*/
+	int *class;	/*will be overlaid on file[0]*/
+	int *member;	/*will be overlaid on file[1]*/
+	int *klist;	/*will be overlaid on file[0] after class*/
+	Cand *clist;	/* merely a free storage pot for candidates */
+	int clen;
+	int *J;		/*will be overlaid on class*/
+	long *ixold;	/*will be overlaid on klist*/
+	long *ixnew;	/*will be overlaid on file[1]*/
+	char *file1;
+	char *file2;
+	Biobuf *input[2];
+	int firstchange;
+	Change *changes;
+	int nchanges;
+};
+
+extern char mode;
+extern char bflag;
+extern char rflag;
+extern char mflag;
+extern int anychange;
+extern Biobuf	stdout;
+
+#define MAXPATHLEN	1024
+
+int mkpathname(char *, char *, char *);
+void *emalloc(unsigned);
+void *erealloc(void *, unsigned);
+void diff(char *, char *, int);
+void diffdir(char *, char *, int);
+void diffreg(char *, char *, char *, char *);
+Biobuf *prepare(Diff*, int, char *, char *);
+void panic(int, char *, ...);
+void check(Diff *, Biobuf *, Biobuf *);
+void change(Diff *, int, int, int, int);
+void flushchanges(Diff *d);
+
--- /dev/null
+++ b/diffdir.c
@@ -1,0 +1,113 @@
+#include <u.h>
+#include <libc.h>
+#include <bio.h>
+#include "diff.h"
+
+static int
+itemcmp(void *v1, void *v2)
+{
+	char **d1 = v1, **d2 = v2;
+
+	return strcmp(*d1, *d2);
+}
+
+static char **
+scandir(char *name)
+{
+	char **cp;
+	Dir *db;
+	int nitems;
+	int fd, n;
+
+	if ((fd = open(name, OREAD)) < 0) {
+		fprint(2, "%s: can't open %s: %r\n", argv0, name);
+		/* fake an empty directory */
+		cp = emalloc(sizeof(char*));
+		cp[0] = 0;
+		return cp;
+	}
+	cp = 0;
+	nitems = 0;
+	if((n = dirreadall(fd, &db)) > 0){
+		while (n--) {
+			cp = erealloc(cp, (nitems+1)*sizeof(char*));
+			cp[nitems] = emalloc(strlen((db+n)->name)+1);
+			strcpy(cp[nitems], (db+n)->name);
+			nitems++;
+		}
+		free(db);
+	}
+	cp = erealloc(cp, (nitems+1)*sizeof(char*));
+	cp[nitems] = 0;
+	close(fd);
+	qsort((char *)cp, nitems, sizeof(char*), itemcmp);
+	return cp;
+}
+
+static int
+isdotordotdot(char *p)
+{
+	if (*p == '.') {
+		if (!p[1])
+			return 1;
+		if (p[1] == '.' && !p[2])
+			return 1;
+	}
+	return 0;
+}
+
+void
+diffdir(char *f, char *t, int level)
+{
+	char  **df, **dt, **dirf, **dirt;
+	char *from, *to;
+	int res;
+	char fb[MAXPATHLEN+1], tb[MAXPATHLEN+1];
+
+	df = scandir(f);
+	dt = scandir(t);
+	dirf = df;
+	dirt = dt;
+	while (*df || *dt) {
+		from = *df;
+		to = *dt;
+		if (from && isdotordotdot(from)) {
+			df++;
+			continue;
+		}
+		if (to && isdotordotdot(to)) {
+			dt++;
+			continue;
+		}
+		if (!from)
+			res = 1;
+		else if (!to)
+			res = -1;
+		else
+			res = strcmp(from, to);
+		if (res < 0) {
+			if (mode == 0 || mode == 'n')
+				Bprint(&stdout, "Only in %s: %s\n", f, from);
+			df++;
+			continue;
+		}
+		if (res > 0) {
+			if (mode == 0 || mode == 'n')
+				Bprint(&stdout, "Only in %s: %s\n", t, to);
+			dt++;
+			continue;
+		}
+		if (mkpathname(fb, f, from))
+			continue;
+		if (mkpathname(tb, t, to))
+			continue;
+		diff(fb, tb, level+1);
+		df++; dt++;
+	}
+	for (df = dirf; *df; df++)
+		free(*df);
+	for (dt = dirt; *dt; dt++)
+		free(*dt);
+	free(dirf);
+	free(dirt);
+}
--- /dev/null
+++ b/diffio.c
@@ -1,0 +1,373 @@
+#include <u.h>
+#include <libc.h>
+#include <bio.h>
+#include <ctype.h>
+#include "diff.h"
+
+#define MAXLINELEN	4096
+#define MIN(x, y)	((x) < (y) ? (x): (y))
+
+static int
+readline(Biobuf *bp, char *buf)
+{
+	int c;
+	char *p, *e;
+
+	p = buf;
+	e = p + MAXLINELEN-1;
+	do {
+		c = Bgetc(bp);
+		if (c < 0) {
+			if (p == buf)
+				return -1;
+			break;
+		}
+		if (c == '\n')
+			break;
+		*p++ = c;
+	} while (p < e);
+	*p = 0;
+	if (c != '\n' && c >= 0) {
+		do c = Bgetc(bp);
+		while (c >= 0 && c != '\n');
+	}
+	return p - buf;
+}
+
+#define HALFLONG 16
+#define low(x)	(x&((1L<<HALFLONG)-1))
+#define high(x)	(x>>HALFLONG)
+
+/*
+ * hashing has the effect of
+ * arranging line in 7-bit bytes and then
+ * summing 1-s complement in 16-bit hunks 
+ */
+static int
+readhash(Biobuf *bp, char *buf)
+{
+	long sum;
+	unsigned shift;
+	char *p;
+	int len, space;
+
+	sum = 1;
+	shift = 0;
+	if ((len = readline(bp, buf)) == -1)
+		return 0;
+	p = buf;
+	switch(bflag)	/* various types of white space handling */
+	{
+	case 0:
+		while (len--) {
+			sum += (long)*p++ << (shift &= (HALFLONG-1));
+			shift += 7;
+		}
+		break;
+	case 1:
+		/*
+		 * coalesce multiple white-space
+		 */
+		for (space = 0; len--; p++) {
+			if (isspace(*p)) {
+				space++;
+				continue;
+			}
+			if (space) {
+				shift += 7;
+				space = 0;
+			}
+			sum += (long)*p << (shift &= (HALFLONG-1));
+			shift += 7;
+		}
+		break;
+	default:
+		/*
+		 * strip all white-space
+		 */
+		while (len--) {
+			if (isspace(*p)) {
+				p++;
+				continue;
+			}
+			sum += (long)*p++ << (shift &= (HALFLONG-1));
+			shift += 7;
+		}
+		break;
+	}
+	sum = low(sum) + high(sum);
+	return ((short)low(sum) + (short)high(sum));
+}
+
+Biobuf *
+prepare(Diff *d, int i, char *arg, char *orig)
+{
+	Line *p;
+	int j, h;
+	Biobuf *bp;
+	char *cp, buf[MAXLINELEN];
+	int nbytes;
+	Rune r;
+
+	bp = Bopen(arg, OREAD);
+	if (!bp) {
+		panic(mflag ? 0: 2, "cannot open %s: %r\n", arg);
+		return 0;
+	}
+	if (d->binary)
+		return bp;
+	nbytes = Bread(bp, buf, MIN(1024, MAXLINELEN));
+	if (nbytes > 0) {
+		cp = buf;
+		while (cp < buf+nbytes-UTFmax) {
+			/*
+			 * heuristic for a binary file in the
+			 * brave new UNICODE world
+			 */
+			cp += chartorune(&r, cp);
+			if (r == 0 || (r > 0x7f && r <= 0xa0)) {
+				d->binary++;
+				return bp;
+			}
+		}
+		Bseek(bp, 0, 0);
+	}
+	p = emalloc(3*sizeof(Line));
+	for (j = 0; h = readhash(bp, buf); p[j].value = h)
+		p = erealloc(p, (++j+3)*sizeof(Line));
+	d->len[i] = j;
+	d->file[i] = p;
+	d->input[i] = bp;
+	if (i == 0) {
+		d->file1 = orig;
+		d->firstchange = 0;
+	} else
+		d->file2 = orig;
+	return bp;
+}
+
+static int
+squishspace(char *buf)
+{
+	char *p, *q;
+	int space;
+
+	for (space = 0, q = p = buf; *q; q++) {
+		if (isspace(*q)) {
+			space++;
+			continue;
+		}
+		if (space && bflag == 1) {
+			*p++ = ' ';
+			space = 0;
+		}
+		*p++ = *q;
+	}
+	*p = 0;
+	return p - buf;
+}
+
+/*
+ * need to fix up for unexpected EOF's
+ */
+void
+check(Diff *d, Biobuf *bf, Biobuf *bt)
+{
+	int f, t, flen, tlen;
+	char fbuf[MAXLINELEN], tbuf[MAXLINELEN];
+
+	d->ixold[0] = 0;
+	d->ixnew[0] = 0;
+	for (f = t = 1; f < d->len[0]; f++) {
+		flen = readline(bf, fbuf);
+		d->ixold[f] = d->ixold[f-1] + flen + 1;		/* ftell(bf) */
+		if (d->J[f] == 0)
+			continue;
+		do {
+			tlen = readline(bt, tbuf);
+			d->ixnew[t] = d->ixnew[t-1] + tlen + 1;	/* ftell(bt) */
+		} while (t++ < d->J[f]);
+		if (bflag) {
+			flen = squishspace(fbuf);
+			tlen = squishspace(tbuf);
+		}
+		if (flen != tlen || strcmp(fbuf, tbuf))
+			d->J[f] = 0;
+	}
+	while (t < d->len[1]) {
+		tlen = readline(bt, tbuf);
+		d->ixnew[t] = d->ixnew[t-1] + tlen + 1;	/* fseek(bt) */
+		t++;
+	}
+}
+
+static void
+range(int a, int b, char *separator)
+{
+	Bprint(&stdout, "%d", a > b ? b: a);
+	if (a < b)
+		Bprint(&stdout, "%s%d", separator, b);
+}
+
+static void
+fetch(Diff *d, long *f, int a, int b, Biobuf *bp, char *s)
+{
+	char buf[MAXLINELEN];
+	int maxb;
+
+	if(a <= 1)
+		a = 1;
+	if(bp == d->input[0])
+		maxb = d->len[0];
+	else
+		maxb = d->len[1];
+	if(b > maxb)
+		b = maxb;
+	if(a > maxb)
+		return;
+	Bseek(bp, f[a-1], 0);
+	while (a++ <= b) {
+		readline(bp, buf);
+		Bprint(&stdout, "%s%s\n", s, buf);
+	}
+}
+
+void
+change(Diff *df, int a, int b, int c, int d)
+{
+	char verb;
+	char buf[4];
+	Change *ch;
+
+	if (a > b && c > d)
+		return;
+	anychange = 1;
+	if (mflag && df->firstchange == 0) {
+		if(mode) {
+			buf[0] = '-';
+			buf[1] = mode;
+			buf[2] = ' ';
+			buf[3] = '\0';
+		} else {
+			buf[0] = '\0';
+		}
+		Bprint(&stdout, "diff %s%s %s\n", buf, df->file1, df->file2);
+		df->firstchange = 1;
+	}
+	verb = a > b ? 'a': c > d ? 'd': 'c';
+	switch(mode) {
+	case 'e':
+		range(a, b, ",");
+		Bputc(&stdout, verb);
+		break;
+	case 0:
+		range(a, b, ",");
+		Bputc(&stdout, verb);
+		range(c, d, ",");
+		break;
+	case 'n':
+		Bprint(&stdout, "%s:", df->file1);
+		range(a, b, ",");
+		Bprint(&stdout, " %c ", verb);
+		Bprint(&stdout, "%s:", df->file2);
+		range(c, d, ",");
+		break;
+	case 'f':
+		Bputc(&stdout, verb);
+		range(a, b, " ");
+		break;
+	case 'c':
+	case 'a':
+	case 'u':
+		if(df->nchanges%1024 == 0)
+			df->changes = erealloc(df->changes, (df->nchanges+1024)*sizeof(df->changes[0]));
+		ch = &df->changes[df->nchanges++];
+		ch->a = a;
+		ch->b = b;
+		ch->c = c;
+		ch->d = d;
+		return;
+	}
+	Bputc(&stdout, '\n');
+	if (mode == 0 || mode == 'n') {
+		fetch(df, df->ixold, a, b, df->input[0], "< ");
+		if (a <= b && c <= d)
+			Bprint(&stdout, "---\n");
+	}
+	fetch(df, df->ixnew, c, d, df->input[1], mode == 0 || mode == 'n' ? "> ": "");
+	if (mode != 0 && mode != 'n' && c <= d)
+		Bprint(&stdout, ".\n");
+}
+
+enum
+{
+	Lines = 3,	/* number of lines of context shown */
+};
+
+int
+changeset(Diff *d, int i)
+{
+	while(i < d->nchanges && d->changes[i].b + 1 + 2*Lines > d->changes[i+1].a)
+		i++;
+	if(i < d->nchanges)
+		return i+1;
+	return d->nchanges;
+}
+
+void
+flushchanges(Diff *df)
+{
+	vlong a, b, c, d, at, hdr;
+	vlong i, j;
+
+	if(df->nchanges == 0)
+		return;
+
+	hdr = 0;
+	for(i=0; i < df->nchanges; ){
+		j = changeset(df, i);
+		a = df->changes[i].a - Lines;
+		b = df->changes[j-1].b + Lines;
+		c = df->changes[i].c - Lines;
+		d = df->changes[j-1].d + Lines;
+		if(a < 1)
+			a = 1;
+		if(c < 1)
+			c = 1;
+		if(b > df->len[0])
+			b = df->len[0];
+		if(d > df->len[1])
+			d = df->len[1];
+		if(mode == 'a'){
+			a = 1;
+			b = df->len[0];
+			c = 1;
+			d = df->len[1];
+			j = df->nchanges;
+		}
+		if(mode == 'u'){
+			if(!hdr){
+				Bprint(&stdout, "--- %s\n", df->file1);
+				Bprint(&stdout, "+++ %s\n", df->file2);
+				hdr = 1;
+			}
+			Bprint(&stdout, "@@ -%lld,%lld +%lld,%lld @@\n", a, b-a+1, c, d-c+1);
+		}else{
+			Bprint(&stdout, "%s:", df->file1);
+			range(a, b, ",");
+			Bprint(&stdout, " - ");
+			Bprint(&stdout, "%s:", df->file2);
+			range(c, d, ",");
+			Bputc(&stdout, '\n');
+		}
+		at = a;
+		for(; i<j; i++){
+			fetch(df, df->ixold, at, df->changes[i].a-1, df->input[0], mode == 'u' ? " " : "  ");
+			fetch(df, df->ixold, df->changes[i].a, df->changes[i].b, df->input[0], mode == 'u' ? "-" : "- ");
+			fetch(df, df->ixnew, df->changes[i].c, df->changes[i].d, df->input[1], mode == 'u' ? "+" : "+ ");
+			at = df->changes[i].b+1;
+		}
+		fetch(df, df->ixold, at, b, df->input[0], mode == 'u' ? " " : "  ");
+	}
+	df->nchanges = 0;
+}
--- /dev/null
+++ b/diffreg.c
@@ -1,0 +1,399 @@
+#include <u.h>
+#include <libc.h>
+#include <bio.h>
+#include "diff.h"
+
+/*	diff - differential file comparison
+*
+*	Uses an algorithm due to Harold Stone, which finds
+*	a pair of longest identical subsequences in the two
+*	files.
+*
+*	The major goal is to generate the match vector J.
+*	J[i] is the index of the line in file1 corresponding
+*	to line i file0. J[i] = 0 if there is no
+*	such line in file1.
+*
+*	Lines are hashed so as to work in core. All potential
+*	matches are located by sorting the lines of each file
+*	on the hash (called value). In particular, this
+*	collects the equivalence classes in file1 together.
+*	Subroutine equiv replaces the value of each line in
+*	file0 by the index of the first element of its 
+*	matching equivalence in (the reordered) file1.
+*	To save space equiv squeezes file1 into a single
+*	array member in which the equivalence classes
+*	are simply concatenated, except that their first
+*	members are flagged by changing sign.
+*
+*	Next the indices that point into member are unsorted into
+*	array class according to the original order of file0.
+*
+*	The cleverness lies in routine stone. This marches
+*	through the lines of file0, developing a vector klist
+*	of "k-candidates". At step i a k-candidate is a matched
+*	pair of lines x,y (x in file0 y in file1) such that
+*	there is a common subsequence of lenght k
+*	between the first i lines of file0 and the first y 
+*	lines of file1, but there is no such subsequence for
+*	any smaller y. x is the earliest possible mate to y
+*	that occurs in such a subsequence.
+*
+*	Whenever any of the members of the equivalence class of
+*	lines in file1 matable to a line in file0 has serial number 
+*	less than the y of some k-candidate, that k-candidate 
+*	with the smallest such y is replaced. The new 
+*	k-candidate is chained (via pred) to the current
+*	k-1 candidate so that the actual subsequence can
+*	be recovered. When a member has serial number greater
+*	that the y of all k-candidates, the klist is extended.
+*	At the end, the longest subsequence is pulled out
+*	and placed in the array J by unravel.
+*
+*	With J in hand, the matches there recorded are
+*	check'ed against reality to assure that no spurious
+*	matches have crept in due to hashing. If they have,
+*	they are broken, and "jackpot " is recorded--a harmless
+*	matter except that a true match for a spuriously
+*	mated line may now be unnecessarily reported as a change.
+*
+*	Much of the complexity of the program comes simply
+*	from trying to minimize core utilization and
+*	maximize the range of doable problems by dynamically
+*	allocating what is needed and reusing what is not.
+*	The core requirements for problems larger than somewhat
+*	are (in words) 2*length(file0) + length(file1) +
+*	3*(number of k-candidates installed),  typically about
+*	6n words for files of length n. 
+*/
+
+static void	
+sort(Line *a, int n)	/*shellsort CACM #201*/
+{
+	int m;
+	Line *ai, *aim, *j, *k;
+	Line w;
+	int i;
+
+	m = 0;
+	for (i = 1; i <= n; i *= 2)
+		m = 2*i - 1;
+	for (m /= 2; m != 0; m /= 2) {
+		k = a+(n-m);
+		for (j = a+1; j <= k; j++) {
+			ai = j;
+			aim = ai+m;
+			do {
+				if (aim->value > ai->value ||
+				   aim->value == ai->value &&
+				   aim->serial > ai->serial)
+					break;
+				w = *ai;
+				*ai = *aim;
+				*aim = w;
+
+				aim = ai;
+				ai -= m;
+			} while (ai > a && aim >= ai);
+		}
+	}
+}
+
+static void
+unsort(Line *f, int l, int *b)
+{
+	int *a;
+	int i;
+
+	a = malloc((l+1)*sizeof(int));
+	for(i=1;i<=l;i++)
+		a[f[i].serial] = f[i].value;
+	for(i=1;i<=l;i++)
+		b[i] = a[i];
+	free(a);
+}
+
+static void
+prune(Diff *d)
+{
+	int i,j;
+
+	for(d->pref = 0; d->pref < d->len[0] && d->pref < d->len[1] &&
+		d->file[0][d->pref+1].value == d->file[1][d->pref+1].value;
+		d->pref++) ;
+	for(d->suff=0; d->suff < d->len[0] - d->pref && d->suff < d->len[1] - d->pref &&
+		d->file[0][d->len[0] - d->suff].value == d->file[1][d->len[1] - d->suff].value;
+		d->suff++) ;
+	for(j=0;j<2;j++) {
+		d->sfile[j] = d->file[j]+d->pref;
+		d->slen[j] = d->len[j]-d->pref-d->suff;
+		for(i=0;i<=d->slen[j];i++)
+			d->sfile[j][i].serial = i;
+	}
+}
+
+static void
+equiv(Line *a, int n, Line *b, int m, int *c)
+{
+	int i, j;
+
+	i = j = 1;
+	while(i<=n && j<=m) {
+		if(a[i].value < b[j].value)
+			a[i++].value = 0;
+		else if(a[i].value == b[j].value)
+			a[i++].value = j;
+		else
+			j++;
+	}
+	while(i <= n)
+		a[i++].value = 0;
+	b[m+1].value = 0;
+	j = 0;
+	while(++j <= m) {
+		c[j] = -b[j].serial;
+		while(b[j+1].value == b[j].value) {
+			j++;
+			c[j] = b[j].serial;
+		}
+	}
+	c[j] = -1;
+}
+
+static int
+newcand(Diff *d, int x, int  y, int pred)
+{
+	Cand *q;
+
+	d->clist = erealloc(d->clist, (d->clen+1)*sizeof(Cand));
+	q = d->clist + d->clen;
+	q->x = x;
+	q->y = y;
+	q->pred = pred;
+	return d->clen++;
+}
+
+static int
+search(Diff *d, int *c, int k, int y)
+{
+	int i, j, l;
+	int t;
+
+	if(d->clist[c[k]].y < y)	/*quick look for typical case*/
+		return k+1;
+	i = 0;
+	j = k+1;
+	while((l=(i+j)/2) > i) {
+		t = d->clist[c[l]].y;
+		if(t > y)
+			j = l;
+		else if(t < y)
+			i = l;
+		else
+			return l;
+	}
+	return l+1;
+}
+
+static int
+stone(Diff *d, int *a, int n, int *b, int *c)
+{
+	int i, k,y;
+	int j, l;
+	int oldc, tc;
+	int oldl;
+
+	k = 0;
+	c[0] = newcand(d, 0, 0, 0);
+	for(i=1; i<=n; i++) {
+		j = a[i];
+		if(j==0)
+			continue;
+		y = -b[j];
+		oldl = 0;
+		oldc = c[0];
+		do {
+			if(y <= d->clist[oldc].y)
+				continue;
+			l = search(d, c, k, y);
+			if(l!=oldl+1)
+				oldc = c[l-1];
+			if(l<=k) {
+				if(d->clist[c[l]].y <= y)
+					continue;
+				tc = c[l];
+				c[l] = newcand(d, i, y, oldc);
+				oldc = tc;
+				oldl = l;
+			} else {
+				c[l] = newcand(d, i,y,oldc);
+				k++;
+				break;
+			}
+		} while((y=b[++j]) > 0);
+	}
+	return k;
+}
+
+static void
+unravel(Diff *d, int p)
+{
+	int i;
+	Cand *q;
+
+	for(i=0; i<=d->len[0]; i++) {
+		if (i <= d->pref)
+			d->J[i] = i;
+		else if (i > d->len[0]-d->suff)
+			d->J[i] = i+d->len[1] - d->len[0];
+		else
+			d->J[i] = 0;
+	}
+	for(q=d->clist+p; q->y != 0; q= d->clist + q->pred)
+		d->J[q->x+d->pref] = q->y+d->pref;
+}
+
+static void
+output(Diff *d)
+{
+	int m, i0, i1, j0, j1;
+
+	m = d->len[0];
+	d->J[0] = 0;
+	d->J[m+1] = d->len[1]+1;
+	if (mode != 'e') {
+		for (i0 = 1; i0 <= m; i0 = i1+1) {
+			while (i0 <= m && d->J[i0] == d->J[i0-1]+1)
+				i0++;
+			j0 = d->J[i0-1]+1;
+			i1 = i0-1;
+			while (i1 < m && d->J[i1+1] == 0)
+				i1++;
+			j1 = d->J[i1+1]-1;
+			d->J[i1] = j1;
+			change(d, i0, i1, j0, j1);
+		}
+	} else {
+		for (i0 = m; i0 >= 1; i0 = i1-1) {
+			while (i0 >= 1 && d->J[i0] == d->J[i0+1]-1 && d->J[i0])
+				i0--;
+			j0 = d->J[i0+1]-1;
+			i1 = i0+1;
+			while (i1 > 1 && d->J[i1-1] == 0)
+				i1--;
+			j1 = d->J[i1-1]+1;
+			d->J[i1] = j1;
+			change(d, i1 , i0, j1, j0);
+		}
+	}
+	if (m == 0)
+		change(d, 1, 0, 1, d->len[1]);
+	flushchanges(d);
+}
+
+#define BUF 4096
+static int
+cmp(Biobuf* b1, Biobuf* b2)
+{
+	int n;
+	uchar buf1[BUF], buf2[BUF];
+	int f1, f2;
+	vlong nc = 1;
+	uchar *b1s, *b1e, *b2s, *b2e;
+
+	f1 = Bfildes(b1);
+	f2 = Bfildes(b2);
+	seek(f1, 0, 0);
+	seek(f2, 0, 0);
+	b1s = b1e = buf1;
+	b2s = b2e = buf2;
+	for(;;){
+		if(b1s >= b1e){
+			if(b1s >= &buf1[BUF])
+				b1s = buf1;
+			n = read(f1, b1s,  &buf1[BUF] - b1s);
+			b1e = b1s + n;
+		}
+		if(b2s >= b2e){
+			if(b2s >= &buf2[BUF])
+				b2s = buf2;
+			n = read(f2, b2s,  &buf2[BUF] - b2s);
+			b2e = b2s + n;
+		}
+		n = b2e - b2s;
+		if(n > b1e - b1s)
+			n = b1e - b1s;
+		if(n <= 0)
+			break;
+		if(memcmp((void *)b1s, (void *)b2s, n) != 0){
+			return 1;
+		}		
+		nc += n;
+		b1s += n;
+		b2s += n;
+	}
+	if(b1e - b1s == b2e - b2s)
+		return 0;
+	return 1;	
+}
+
+void
+diffreg(char *f, char *fo, char *t, char *to)
+{
+	Biobuf *b0, *b1;
+	Diff d;
+	int k;
+
+	d.binary = 0;
+	memset(&d, 0, sizeof(d));
+	b0 = prepare(&d, 0, f, fo);
+	if (!b0)
+		return;
+	b1 = prepare(&d, 1, t, to);
+	if (!b1) {
+		Bterm(b0);
+		return;
+	}
+	if (d.binary){
+		// could use b0 and b1 but this is simpler.
+		if (cmp(b0, b1))
+			print("binary files %s %s differ\n", f, t);
+		Bterm(b0);
+		Bterm(b1);
+		return;
+	}
+	d.clen = 0;
+	prune(&d);
+	sort(d.sfile[0], d.slen[0]);
+	sort(d.sfile[1], d.slen[1]);
+
+	d.member = (int *)d.file[1];
+	equiv(d.sfile[0], d.slen[0], d.sfile[1], d.slen[1], d.member);
+	d.member = erealloc(d.member, (d.slen[1]+2)*sizeof(int));
+
+	d.class = (int *)d.file[0];
+	unsort(d.sfile[0], d.slen[0], d.class);
+	d.class = erealloc(d.class, (d.slen[0]+2)*sizeof(int));
+
+	d.klist = emalloc((d.slen[0]+2)*sizeof(int));
+	d.clist = emalloc(sizeof(Cand));
+	k = stone(&d, d.class, d.slen[0], d.member, d.klist);
+	free(d.member);
+	free(d.class);
+
+	d.J = emalloc((d.len[0]+2)*sizeof(int));
+	unravel(&d, d.klist[k]);
+	free(d.clist);
+	free(d.klist);
+
+	d.ixold = emalloc((d.len[0]+2)*sizeof(long));
+	d.ixnew = emalloc((d.len[1]+2)*sizeof(long));
+	Bseek(b0, 0, 0); Bseek(b1, 0, 0);
+	check(&d, b0, b1);
+	output(&d);
+	free(d.J);
+	free(d.ixold);
+	free(d.ixnew);
+	Bterm(b0);
+	Bterm(b1);
+}
--- /dev/null
+++ b/mkfile
@@ -1,0 +1,12 @@
+< /$objtype/mkfile
+
+TARG=diff
+OFILES=\
+	diffdir.$O\
+	diffio.$O\
+	diffreg.$O\
+
+HFILES=diff.h
+
+BIN=/$objtype/bin
+</sys/src/cmd/mkmany