Skip to content

Commit

Permalink
Speedup PreProcessXmlString()
Browse files Browse the repository at this point in the history 
There are more than necessary comparions made on every char. By
1. do a quick check before proceeding to more complicated paths
2. use pointers instead of indices
we see ~50% fewer time spent in the function.
  • Loading branch information
@bbshelper
bbshelper committed Sep 25, 2023
1 parent 84d2ff0 commit e5281de
Showing 1 changed file with 125 additions and 123 deletions.
248 changes: 125 additions & 123 deletions crengine/src/lvxml.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -5428,143 +5428,145 @@ int codeconvert(int code)
}
}

/// in-place XML string decoding, don't expand tabs, returns new length (may be less than initial len)
int PreProcessXmlString(lChar32 * str, int len, lUInt32 flags, const lChar32 * enc_table)
{
int state = 0;
static inline int collapsable(int ch) {
return ch == ' ' || ch == '\n' || ch == '\r' || ch == '\t';
}

static void decode_character(const lChar32 *end, const lChar32 *enc_table,
const lChar32 *&src, lChar32 *&dst) {
int state = 1, ch;
lChar32 nch = 0;
lChar32 lch = 0;
lChar32 nsp = 0;
bool cdata = (flags & TXTFLG_CDATA) != 0;
bool pre = (flags & TXTFLG_PRE) != 0;
bool pre_para_splitting = (flags & TXTFLG_PRE_PARA_SPLITTING)!=0;
if ( pre_para_splitting )
pre = false;
bool attribute = (flags & TXTFLG_PROCESS_ATTRIBUTE) != 0;
//CRLog::trace("before: '%s' %s, len=%d", LCSTR(str), pre ? "pre ":" ", len);
int j = 0;
for (int i=0; i<len; ++i ) {
if (j >= len)
break;
lChar32 ch = str[i];
if (pre) {
if (ch == '\r') {
if ((i==0 || lch!='\n') && (i==len-1 || str[i+1]!='\n')) {
str[j++] = '\n';
lch = '\n';
for (; src < end; ++src) {
ch = *src;
if (state == 2 && ch == 'x')
state = 22;
else if (state == 22 && hexDigit(ch) >= 0)
nch = (lChar32) ((nch << 4) | hexDigit(ch));
else if (state == 2 && ch >= '0' && ch <= '9')
nch = (lChar32) (nch * 10 + (ch - '0'));
else if (ch == '#' && state == 1)
state = 2;
else if (state == 1 && ((ch >= 'a' && ch <= 'z') || (ch >= 'A' && ch <= 'Z'))) {
int k;
lChar32 entname[32];
for (k = 0; k < 32; k++) {
entname[k] = src[k];
if (!entname[k] || entname[k] == ';' || entname[k] == ' ')
break;
}
if (32 == k)
k--;
entname[k] = 0;
lChar32 code = 0;
lChar32 code2 = 0;
if (src[k] == ';' || src[k] == ' ') {
// Nb of iterations for some classic named entities:
// nbsp: 5 - amp: 7 - lt: 8 - quot: 9
// apos gt shy eacute 10
// Let's have some early straight comparisons for the ones we
// have a chance to find in huge quantities in some documents.
if (!lStr_cmp(entname, U"nbsp"))
code = 160;
else if (!lStr_cmp(entname, U"shy"))
code = 173;
else {
// Binary search (usually takes 5 to 12 iterations)
int left = 0;
int right = sizeof(def_entity_table) / sizeof((def_entity_table)[0]) - 1; // ignore last NULL
int middle;
int iters = 0;
while (left < right) {
iters++;
middle = (left + right) / 2;
int res = lStr_cmp(entname, def_entity_table[middle].name);
if (res == 0) {
code = def_entity_table[middle].code;
code2 = def_entity_table[middle].code2;
break;
} else if (res < 0) {
right = middle;
} else {
left = middle + 1;
}
}
}
continue;
} else if (ch == '\n') {
str[j++] = '\n';
lch = ch;
continue;
}
} else if ( !attribute ) {
if (ch=='\r' || ch=='\n' || ch=='\t')
ch = ' ';
}
if (ch == '&' && !cdata) {
state = 1;
nch = 0;
} else if (state == 0) {
if (ch == ' ') {
if ( pre || attribute || !nsp )
str[j++] = ch;
nsp++;
if (code) {
src += k - 1;
state = 0;
if (enc_table && code < 256 && code >= 128)
code = enc_table[code - 128];
*dst++ = code;
if (code2) {
if (enc_table && code2 < 256 && code2 >= 128)
code2 = enc_table[code2 - 128];
*dst++ = code2;
}
} else {
str[j++] = ch;
nsp = 0;
// include & and rest of entity into output string
if (dst < end - 1) {
*dst++ = '&';
*dst++ = *src;
}
break;
}
} else if (ch == ';') {
if (nch)
*dst++ = codeconvert(nch);
break;
} else {
if (state == 2 && ch=='x')
state = 22;
else if (state == 22 && hexDigit(ch)>=0)
nch = (lChar32)((nch << 4) | hexDigit(ch));
else if (state == 2 && ch>='0' && ch<='9')
nch = (lChar32)(nch * 10 + (ch - '0'));
else if (ch=='#' && state==1)
state = 2;
else if (state==1 && ((ch>='a' && ch<='z') || (ch>='A' && ch<='Z')) ) {
int k;
lChar32 entname[32];
for ( k = 0; k < 32; k++ ) {
entname[k] = str[k + i];
if (!entname[k] || entname[k]==';' || entname[k]==' ')
break;
break;
}
}
}

/// in-place XML string decoding, don't expand tabs, returns new length (may be less than initial len)
int PreProcessXmlString(lChar32 * str, int len, lUInt32 flags, const lChar32 * enc_table) {
bool cdata = (flags & TXTFLG_CDATA) != 0;
bool pre = (flags & TXTFLG_PRE) != 0;
bool pre_para_splitting = (flags & TXTFLG_PRE_PARA_SPLITTING) != 0;
if (pre_para_splitting)
pre = false;
bool attribute = (flags & TXTFLG_PROCESS_ATTRIBUTE) != 0;

const lChar32 *end = str + len;
lChar32 *dst = str;
for (const lChar32 *src = str; src < end; ++src) {
lChar32 ch = *src;
if (ch <= '&') [[unlikely]] {
if (ch == '&') {
if (src < end - 1) {
decode_character(end, enc_table, ++src, dst);
continue;
}
if (32 == k)
k--;
entname[k] = 0;
lChar32 code = 0;
lChar32 code2 = 0;
if ( str[i+k]==';' || str[i+k]==' ' ) {
// Nb of iterations for some classic named entities:
// nbsp: 5 - amp: 7 - lt: 8 - quot: 9
// apos gt shy eacute 10
// Let's have some early straight comparisons for the ones we
// have a chance to find in huge quantities in some documents.
if ( !lStr_cmp( entname, U"nbsp" ) )
code = 160;
else if ( !lStr_cmp( entname, U"shy" ) )
code = 173;
else {
// Binary search (usually takes 5 to 12 iterations)
int left = 0;
int right = sizeof(def_entity_table) / sizeof((def_entity_table)[0]) - 1; // ignore last NULL
int middle;
int iters = 0;
while ( left < right ) {
iters++;
middle = (left + right) / 2;
int res = lStr_cmp( entname, def_entity_table[middle].name );
if ( res == 0 ) {
code = def_entity_table[middle].code;
code2 = def_entity_table[middle].code2;
break;
}
else if ( res < 0 ) {
right = middle;
}
else {
left = middle + 1;
}
} else {
if (pre) {
// replace "\r\n" with "\n", otherwise remove it
if (ch == '\r') {
if (src < end - 1 && src[1] == '\n') {
*dst++ = '\n';
++src;
}
continue;
}
}
if ( code ) {
i+=k;
state = 0;
if ( enc_table && code<256 && code>=128 )
code = enc_table[code - 128];
str[j++] = code;
if ( code2 ) {
if ( enc_table && code2<256 && code2>=128 )
code2 = enc_table[code2 - 128];
str[j++] = code2;
}
nsp = 0;
} else {
// include & and rest of entity into output string
if (j < len - 1) {
str[j++] = '&';
str[j++] = str[i];
} else if (!attribute) {
// collapse whitespaces
if (collapsable(ch)) {
const lChar32 *ptr = src;
for (ptr = src + 1; ptr < end; ++ptr) {
if (!collapsable(*ptr))
break;
}
*dst++ = ' ';
src = ptr - 1;
continue;
}
state = 0;
}

} else if (ch == ';') {
if (nch)
str[j++] = codeconvert(nch);
state = 0;
nsp = 0;
} else {
// error: return to normal mode
state = 0;
}
}
lch = ch;
*dst++ = ch;
}
return j;
return dst - str;
}

int CalcTabCount(const lChar32 * str, int nlen) {
Expand Down

0 comments on commit e5281de

Please sign in to comment.