expo/bsdiff
1
0 Fork 0
mirror of https://github.com/expo/bsdiff.git synced 2026-11-14 17:49:03 -10:00
Code Issues Packages Projects Releases Wiki Activity

Change bsdiff to use divsufsort suffix sort library instead of qsufsort.

Browse Source 
Divsufsort is more efficient and addresses the problem of bsdiff hanging
on certain files.

Here are some measurements of the delta diff generator comparing the two
suffix sort alogrithms.

qsufsort:
real	4m40.670s
user	4m19.170s
sys	0m6.400s

divsufsort:
real	3m29.426s
user	3m7.890s
sys	0m6.630s

BUG=chrome-os-partner:3785
TEST=Unit tests, created + applied delta and ensured the system updated
properly, manually ran bsdiff against problematic files to ensure it
completed in a timely manner

Change-Id: Ic5d4215a2de5a1845d88fad5c466f1adf0eaccee
Reviewed-on: http://gerrit.chromium.org/gerrit/1080
Reviewed-by: Darin Petkov <[email protected]>
Tested-by: Thieu Le <[email protected]>
This commit is contained in:
Thieu Le authored and Alex Deymo committed 2015-09-23 10:01:35 -07:00
1 parent 3ebed9c520
commit c2c68a78ac
1 file changed
+11 -105
Download Patch File Download Diff File Expand all files Collapse all files
bsdiff.c
+11 -105
View File
@@ -38,107 +38,16 @@ __FBSDID("$FreeBSD: src/usr.bin/bsdiff/bsdiff/bsdiff.c,v 1.1 2005/08/06 01:59:05
#include <string.h>
#include <unistd.h>
#if _FILE_OFFSET_BITS == 64
#include "divsufsort64.h"
#define saidx_t saidx64_t
#define divsufsort divsufsort64
#else
#include "divsufsort.h"
#endif
#define MIN(x,y) (((x)<(y)) ? (x) : (y))
static void split(off_t *I,off_t *V,off_t start,off_t len,off_t h)
{
off_t i,j,k,x,tmp,jj,kk;
if(len<16) {
for(k=start;k<start+len;k+=j) {
j=1;x=V[I[k]+h];
for(i=1;k+i<start+len;i++) {
if(V[I[k+i]+h]<x) {
x=V[I[k+i]+h];
j=0;
};
if(V[I[k+i]+h]==x) {
tmp=I[k+j];I[k+j]=I[k+i];I[k+i]=tmp;
j++;
};
};
for(i=0;i<j;i++) V[I[k+i]]=k+j-1;
if(j==1) I[k]=-1;
};
return;
};
x=V[I[start+len/2]+h];
jj=0;kk=0;
for(i=start;i<start+len;i++) {
if(V[I[i]+h]<x) jj++;
if(V[I[i]+h]==x) kk++;
};
jj+=start;kk+=jj;
i=start;j=0;k=0;
while(i<jj) {
if(V[I[i]+h]<x) {
i++;
} else if(V[I[i]+h]==x) {
tmp=I[i];I[i]=I[jj+j];I[jj+j]=tmp;
j++;
} else {
tmp=I[i];I[i]=I[kk+k];I[kk+k]=tmp;
k++;
};
};
while(jj+j<kk) {
if(V[I[jj+j]+h]==x) {
j++;
} else {
tmp=I[jj+j];I[jj+j]=I[kk+k];I[kk+k]=tmp;
k++;
};
};
if(jj>start) split(I,V,start,jj-start,h);
for(i=0;i<kk-jj;i++) V[I[jj+i]]=kk-1;
if(jj==kk-1) I[jj]=-1;
if(start+len>kk) split(I,V,kk,start+len-kk,h);
}
static void qsufsort(off_t *I,off_t *V,u_char *old,off_t oldsize)
{
off_t buckets[256];
off_t i,h,len;
for(i=0;i<256;i++) buckets[i]=0;
for(i=0;i<oldsize;i++) buckets[old[i]]++;
for(i=1;i<256;i++) buckets[i]+=buckets[i-1];
for(i=255;i>0;i--) buckets[i]=buckets[i-1];
buckets[0]=0;
for(i=0;i<oldsize;i++) I[++buckets[old[i]]]=i;
I[0]=oldsize;
for(i=0;i<oldsize;i++) V[i]=buckets[old[i]];
V[oldsize]=0;
for(i=1;i<256;i++) if(buckets[i]==buckets[i-1]+1) I[buckets[i]]=-1;
I[0]=-1;
for(h=1;I[0]!=-(oldsize+1);h+=h) {
len=0;
for(i=0;i<oldsize+1;) {
if(I[i]<0) {
len-=I[i];
i-=I[i];
} else {
if(len) I[i-len]=-len;
len=V[I[i]]+1-i;
split(I,V,i,len,h);
i+=len;
len=0;
};
};
if(len) I[i-len]=-len;
};
for(i=0;i<oldsize+1;i++) I[V[i]]=i;
}
static off_t matchlen(u_char *old,off_t oldsize,u_char *new,off_t newsize)
{
off_t i;
@@ -198,7 +107,7 @@ int main(int argc,char *argv[])
int fd;
u_char *old,*new;
off_t oldsize,newsize;
off_t *I,*V;
saidx_t *I;
off_t scan,pos,len;
off_t lastscan,lastpos,lastoffset;
off_t oldscore,scsc;
@@ -224,12 +133,9 @@ int main(int argc,char *argv[])
(read(fd,old,oldsize)!=oldsize) ||
(close(fd)==-1)) err(1,"%s",argv[1]);
if(((I=malloc((oldsize+1)*sizeof(off_t)))==NULL) ||
((V=malloc((oldsize+1)*sizeof(off_t)))==NULL)) err(1,NULL);
if(((I=malloc((oldsize+1)*sizeof(saidx_t)))==NULL)) err(1,NULL);
qsufsort(I,V,old,oldsize);
free(V);
if(divsufsort(old, I, oldsize)) err(1, "divsufsort");
/* Allocate newsize+1 bytes instead of newsize bytes to ensure
that we never try to malloc(0) and get a NULL pointer */