UEFITool/common/bstrlib/bstrwrap.cpp

1721 lines
39 KiB
C++
Raw Permalink Normal View History

2017-01-14 08:24:56 +08:00
/*
* This source file is part of the bstring string library. This code was
* written by Paul Hsieh in 2002-2015, and is covered by the BSD open source
* license and the GPL. Refer to the accompanying documentation for details
* on usage and license.
*/
/*
* bstrwrap.c
*
* This file is the C++ wrapper for the bstring functions.
*/
#if defined (_MSC_VER)
# define _CRT_SECURE_NO_WARNINGS
#endif
#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <stdarg.h>
#include <limits.h>
#include "bstrwrap.h"
#if defined(MEMORY_DEBUG) || defined(BSTRLIB_MEMORY_DEBUG)
#include "memdbg.h"
#endif
#ifndef bstr__alloc
#define bstr__alloc(x) malloc (x)
#endif
#ifndef bstr__free
#define bstr__free(p) free (p)
#endif
#ifndef bstr__realloc
#define bstr__realloc(p,x) realloc ((p), (x))
#endif
#ifndef bstr__memcpy
#define bstr__memcpy(d,s,l) memcpy ((d), (s), (l))
#endif
#ifndef bstr__memmove
#define bstr__memmove(d,s,l) memmove ((d), (s), (l))
#endif
#ifndef bstr__memset
#define bstr__memset(d,c,l) memset ((d), (c), (l))
#endif
#ifndef bstr__memcmp
#define bstr__memcmp(d,c,l) memcmp ((d), (c), (l))
#endif
#ifndef bstr__memchr
#define bstr__memchr(s,c,l) memchr ((s), (c), (l))
#endif
#if defined(BSTRLIB_CAN_USE_IOSTREAM)
#include <iostream>
#endif
namespace Bstrlib {
// Constructors.
CBString::CBString () {
slen = 0;
mlen = 8;
data = (unsigned char *) bstr__alloc (mlen);
if (!data) {
mlen = 0;
bstringThrow ("Failure in default constructor");
} else {
data[0] = '\0';
}
}
CBString::CBString (const void * blk, int len) {
data = NULL;
if (len >= 0) {
mlen = len + 1;
slen = len;
data = (unsigned char *) bstr__alloc (mlen);
}
if (!data) {
mlen = slen = 0;
bstringThrow ("Failure in block constructor");
} else {
if (slen > 0) bstr__memcpy (data, blk, slen);
data[slen] = '\0';
}
}
CBString::CBString (char c, int len) {
data = NULL;
if (len >= 0) {
mlen = len + 1;
slen = len;
data = (unsigned char *) bstr__alloc (mlen);
}
if (!data) {
mlen = slen = 0;
bstringThrow ("Failure in repeat(char) constructor");
} else {
if (slen > 0) bstr__memset (data, c, slen);
data[slen] = '\0';
}
}
CBString::CBString (char c) {
mlen = 2;
slen = 1;
if (NULL == (data = (unsigned char *) bstr__alloc (mlen))) {
mlen = slen = 0;
bstringThrow ("Failure in (char) constructor");
} else {
data[0] = (unsigned char) c;
data[1] = '\0';
}
}
CBString::CBString (unsigned char c) {
mlen = 2;
slen = 1;
if (NULL == (data = (unsigned char *) bstr__alloc (mlen))) {
mlen = slen = 0;
bstringThrow ("Failure in (char) constructor");
} else {
data[0] = c;
data[1] = '\0';
}
}
CBString::CBString (const char *s) {
if (s) {
size_t sslen = strlen (s);
if (sslen >= INT_MAX) bstringThrow ("Failure in (char *) constructor, string too large")
slen = (int) sslen;
mlen = slen + 1;
if (NULL != (data = (unsigned char *) bstr__alloc (mlen))) {
bstr__memcpy (data, s, mlen);
return;
}
}
data = NULL;
bstringThrow ("Failure in (char *) constructor");
}
CBString::CBString (int len, const char *s) {
if (s) {
size_t sslen = strlen (s);
if (sslen >= INT_MAX) bstringThrow ("Failure in (char *) constructor, string too large")
slen = (int) sslen;
mlen = slen + 1;
if (mlen < len) mlen = len;
if (NULL != (data = (unsigned char *) bstr__alloc (mlen))) {
bstr__memcpy (data, s, slen + 1);
return;
}
}
data = NULL;
bstringThrow ("Failure in (int len, char *) constructor");
}
CBString::CBString (const CBString& b) {
slen = b.slen;
mlen = slen + 1;
data = NULL;
if (mlen > 0) data = (unsigned char *) bstr__alloc (mlen);
if (!data) {
bstringThrow ("Failure in (CBString) constructor");
} else {
bstr__memcpy (data, b.data, slen);
data[slen] = '\0';
}
}
CBString::CBString (const tagbstring& x) {
slen = x.slen;
mlen = slen + 1;
data = NULL;
if (slen >= 0 && x.data != NULL) data = (unsigned char *) bstr__alloc (mlen);
if (!data) {
bstringThrow ("Failure in (tagbstring) constructor");
} else {
bstr__memcpy (data, x.data, slen);
data[slen] = '\0';
}
}
// Destructor.
CBString::~CBString () {
if (data != NULL) {
bstr__free (data);
data = NULL;
}
mlen = 0;
slen = -__LINE__;
}
// = operator.
const CBString& CBString::operator = (char c) {
if (mlen <= 0) bstringThrow ("Write protection error");
if (2 >= mlen) alloc (2);
if (!data) {
mlen = slen = 0;
bstringThrow ("Failure in =(char) operator");
} else {
slen = 1;
data[0] = (unsigned char) c;
data[1] = '\0';
}
return *this;
}
const CBString& CBString::operator = (unsigned char c) {
if (mlen <= 0) bstringThrow ("Write protection error");
if (2 >= mlen) alloc (2);
if (!data) {
mlen = slen = 0;
bstringThrow ("Failure in =(char) operator");
} else {
slen = 1;
data[0] = c;
data[1] = '\0';
}
return *this;
}
const CBString& CBString::operator = (const char *s) {
size_t tmpSlen;
if (mlen <= 0) bstringThrow ("Write protection error");
if (NULL == s) s = "";
if ((tmpSlen = strlen (s)) >= (size_t) mlen) {
if (tmpSlen >= INT_MAX-1) bstringThrow ("Failure in =(const char *) operator, string too large");
alloc ((int) tmpSlen);
}
if (data) {
slen = (int) tmpSlen;
bstr__memcpy (data, s, tmpSlen + 1);
} else {
mlen = slen = 0;
bstringThrow ("Failure in =(const char *) operator");
}
return *this;
}
const CBString& CBString::operator = (const CBString& b) {
if (mlen <= 0) bstringThrow ("Write protection error");
if (b.slen >= mlen) alloc (b.slen);
slen = b.slen;
if (!data) {
mlen = slen = 0;
bstringThrow ("Failure in =(CBString) operator");
} else {
bstr__memcpy (data, b.data, slen);
data[slen] = '\0';
}
return *this;
}
const CBString& CBString::operator = (const tagbstring& x) {
if (mlen <= 0) bstringThrow ("Write protection error");
if (x.slen < 0) bstringThrow ("Failure in =(tagbstring) operator, badly formed tagbstring");
if (x.slen >= mlen) alloc (x.slen);
slen = x.slen;
if (!data) {
mlen = slen = 0;
bstringThrow ("Failure in =(tagbstring) operator");
} else {
bstr__memcpy (data, x.data, slen);
data[slen] = '\0';
}
return *this;
}
const CBString& CBString::operator += (const CBString& b) {
if (BSTR_ERR == bconcat (this, (bstring) &b)) {
bstringThrow ("Failure in concatenate");
}
return *this;
}
const CBString& CBString::operator += (const char *s) {
char * d;
int i, l;
if (mlen <= 0) bstringThrow ("Write protection error");
/* Optimistically concatenate directly */
l = mlen - slen;
d = (char *) &data[slen];
for (i=0; i < l; i++) {
if ((*d++ = *s++) == '\0') {
slen += i;
return *this;
}
}
slen += i;
if (BSTR_ERR == bcatcstr (this, s)) {
bstringThrow ("Failure in concatenate");
}
return *this;
}
const CBString& CBString::operator += (char c) {
if (BSTR_ERR == bconchar (this, c)) {
bstringThrow ("Failure in concatenate");
}
return *this;
}
const CBString& CBString::operator += (unsigned char c) {
if (BSTR_ERR == bconchar (this, (char) c)) {
bstringThrow ("Failure in concatenate");
}
return *this;
}
const CBString& CBString::operator += (const tagbstring& x) {
if (mlen <= 0) bstringThrow ("Write protection error");
if (x.slen < 0) bstringThrow ("Failure in +=(tagbstring) operator, badly formed tagbstring");
alloc (x.slen + slen + 1);
if (!data) {
mlen = slen = 0;
bstringThrow ("Failure in +=(tagbstring) operator");
} else {
bstr__memcpy (data + slen, x.data, x.slen);
slen += x.slen;
data[slen] = '\0';
}
return *this;
}
const CBString CBString::operator + (char c) const {
CBString retval (*this);
retval += c;
return retval;
}
const CBString CBString::operator + (unsigned char c) const {
CBString retval (*this);
retval += c;
return retval;
}
const CBString CBString::operator + (const CBString& b) const {
CBString retval (*this);
retval += b;
return retval;
}
const CBString CBString::operator + (const char *s) const {
if (s == NULL) bstringThrow ("Failure in + (char *) operator, NULL");
CBString retval (*this);
retval += s;
return retval;
}
const CBString CBString::operator + (const unsigned char *s) const {
if (s == NULL) bstringThrow ("Failure in + (unsigned char *) operator, NULL");
CBString retval (*this);
retval += (const char *) s;
return retval;
}
const CBString CBString::operator + (const tagbstring& x) const {
if (x.slen < 0) bstringThrow ("Failure in + (tagbstring) operator, badly formed tagbstring");
CBString retval (*this);
retval += x;
return retval;
}
bool CBString::operator == (const CBString& b) const {
int retval;
if (BSTR_ERR == (retval = biseq ((bstring)this, (bstring)&b))) {
bstringThrow ("Failure in compare (==)");
}
return retval > 0;
}
bool CBString::operator == (const char * s) const {
int retval;
if (NULL == s) {
bstringThrow ("Failure in compare (== NULL)");
}
if (BSTR_ERR == (retval = biseqcstr ((bstring) this, s))) {
bstringThrow ("Failure in compare (==)");
}
return retval > 0;
}
bool CBString::operator == (const unsigned char * s) const {
int retval;
if (NULL == s) {
bstringThrow ("Failure in compare (== NULL)");
}
if (BSTR_ERR == (retval = biseqcstr ((bstring) this, (const char *) s))) {
bstringThrow ("Failure in compare (==)");
}
return retval > 0;
}
bool CBString::operator != (const CBString& b) const {
return ! ((*this) == b);
}
bool CBString::operator != (const char * s) const {
return ! ((*this) == s);
}
bool CBString::operator != (const unsigned char * s) const {
return ! ((*this) == s);
}
bool CBString::operator < (const CBString& b) const {
int retval;
if (SHRT_MIN == (retval = bstrcmp ((bstring) this, (bstring)&b))) {
bstringThrow ("Failure in compare (<)");
}
return retval < 0;
}
bool CBString::operator < (const char * s) const {
if (s == NULL) {
bstringThrow ("Failure in compare (<)");
}
return strcmp ((const char *)this->data, s) < 0;
}
bool CBString::operator < (const unsigned char * s) const {
if (s == NULL) {
bstringThrow ("Failure in compare (<)");
}
return strcmp ((const char *)this->data, (const char *)s) < 0;
}
bool CBString::operator <= (const CBString& b) const {
int retval;
if (SHRT_MIN == (retval = bstrcmp ((bstring) this, (bstring)&b))) {
bstringThrow ("Failure in compare (<=)");
}
return retval <= 0;
}
bool CBString::operator <= (const char * s) const {
if (s == NULL) {
bstringThrow ("Failure in compare (<=)");
}
return strcmp ((const char *)this->data, s) <= 0;
}
bool CBString::operator <= (const unsigned char * s) const {
if (s == NULL) {
bstringThrow ("Failure in compare (<=)");
}
return strcmp ((const char *)this->data, (const char *)s) <= 0;
}
bool CBString::operator > (const CBString& b) const {
return ! ((*this) <= b);
}
bool CBString::operator > (const char * s) const {
return ! ((*this) <= s);
}
bool CBString::operator > (const unsigned char * s) const {
return ! ((*this) <= s);
}
bool CBString::operator >= (const CBString& b) const {
return ! ((*this) < b);
}
bool CBString::operator >= (const char * s) const {
return ! ((*this) < s);
}
bool CBString::operator >= (const unsigned char * s) const {
return ! ((*this) < s);
}
CBString::operator double () const {
double d = 0;
if (1 != sscanf ((const char *)this->data, "%lf", &d)) {
bstringThrow ("Unable to convert to a double");
}
return d;
}
CBString::operator float () const {
float d = 0;
if (1 != sscanf ((const char *)this->data, "%f", &d)) {
bstringThrow ("Unable to convert to a float");
}
return d;
}
CBString::operator int () const {
int d = 0;
if (1 != sscanf ((const char *)this->data, "%d", &d)) {
bstringThrow ("Unable to convert to an int");
}
return d;
}
CBString::operator unsigned int () const {
unsigned int d = 0;
if (1 != sscanf ((const char *)this->data, "%u", &d)) {
bstringThrow ("Unable to convert to an unsigned int");
}
return d;
}
#ifdef __TURBOC__
# ifndef BSTRLIB_NOVSNP
# define BSTRLIB_NOVSNP
# endif
#endif
/* Give WATCOM C/C++, MSVC some latitude for their non-support of vsnprintf */
#if defined(__WATCOMC__) || defined(_MSC_VER)
#define exvsnprintf(r,b,n,f,a) {r = _vsnprintf (b,n,f,a);}
#else
#ifdef BSTRLIB_NOVSNP
/* This is just a hack. If you are using a system without a vsnprintf, it is
not recommended that bformat be used at all. */
#define exvsnprintf(r,b,n,f,a) {vsprintf (b,f,a); r = -1;}
#define START_VSNBUFF (256)
#else
#if defined (__GNUC__) && !defined (__PPC__)
/* Something is making gcc complain about this prototype not being here, so
I've just gone ahead and put it in. */
extern "C" {
extern int vsnprintf (char *buf, size_t count, const char *format, va_list arg);
}
#endif
#define exvsnprintf(r,b,n,f,a) {r = vsnprintf (b,n,f,a);}
#endif
#endif
#ifndef START_VSNBUFF
#define START_VSNBUFF (16)
#endif
/*
* Yeah I'd like to just call a vformat function or something, but because of
* the ANSI specified brokeness of the va_* macros, it is actually not
* possible to do this correctly.
*/
void CBString::format (const char * fmt, ...) {
bstring b;
va_list arglist;
int r, n;
if (mlen <= 0) bstringThrow ("Write protection error");
if (fmt == NULL) {
*this = "<NULL>";
bstringThrow ("CBString::format (NULL, ...) is erroneous.");
} else {
if ((b = bfromcstr ("")) == NULL) {
#ifdef BSTRLIB_THROWS_EXCEPTIONS
bstringThrow ("CBString::format out of memory.");
#else
*this = "<NULL>";
#endif
} else {
if ((n = (int) (2 * (strlen) (fmt))) < START_VSNBUFF) n = START_VSNBUFF;
for (;;) {
if (BSTR_OK != balloc (b, n + 2)) {
#ifdef BSTRLIB_THROWS_EXCEPTIONS
bstringThrow ("CBString::format out of memory.");
#else
b = bformat ("<NULL>");
break;
#endif
}
va_start (arglist, fmt);
exvsnprintf (r, (char *) b->data, n + 1, fmt, arglist);
va_end (arglist);
b->data[n] = '\0';
b->slen = (int) (strlen) ((char *) b->data);
if (b->slen < n) break;
if (r > n) n = r; else n += n;
}
*this = *b;
bdestroy (b);
}
}
}
void CBString::formata (const char * fmt, ...) {
bstring b;
va_list arglist;
int r, n;
if (mlen <= 0) bstringThrow ("Write protection error");
if (fmt == NULL) {
*this += "<NULL>";
bstringThrow ("CBString::formata (NULL, ...) is erroneous.");
} else {
if ((b = bfromcstr ("")) == NULL) {
#ifdef BSTRLIB_THROWS_EXCEPTIONS
bstringThrow ("CBString::format out of memory.");
#else
*this += "<NULL>";
#endif
} else {
if ((n = (int) (2 * (strlen) (fmt))) < START_VSNBUFF) n = START_VSNBUFF;
for (;;) {
if (BSTR_OK != balloc (b, n + 2)) {
#ifdef BSTRLIB_THROWS_EXCEPTIONS
bstringThrow ("CBString::format out of memory.");
#else
b = bformat ("<NULL>");
break;
#endif
}
va_start (arglist, fmt);
exvsnprintf (r, (char *) b->data, n + 1, fmt, arglist);
va_end (arglist);
b->data[n] = '\0';
b->slen = (int) (strlen) ((char *) b->data);
if (b->slen < n) break;
if (r > n) n = r; else n += n;
}
*this += *b;
bdestroy (b);
}
}
}
int CBString::caselessEqual (const CBString& b) const {
int ret;
if (BSTR_ERR == (ret = biseqcaseless ((bstring) this, (bstring) &b))) {
bstringThrow ("CBString::caselessEqual Unable to compare");
}
return ret;
}
int CBString::caselessCmp (const CBString& b) const {
int ret;
if (SHRT_MIN == (ret = bstricmp ((bstring) this, (bstring) &b))) {
bstringThrow ("CBString::caselessCmp Unable to compare");
}
return ret;
}
int CBString::find (const CBString& b, int pos) const {
return binstr ((bstring) this, pos, (bstring) &b);
}
/*
int CBString::find (const char * b, int pos) const;
Uses and unrolling and sliding paired indexes character matching. Since
the unrolling is the primary real world impact the true purpose of this
algorithm choice is maximize the effectiveness of the unrolling. The
idea is to scan until at least one match of the current indexed character
from each string, and then shift indexes of both down by and repeat until
the last character form b matches. When the last character from b
matches if the were no mismatches in previous strlen(b) characters then
we know we have a full match, otherwise shift both indexes back strlen(b)
characters and continue.
In general, if there is any character in b that is not at all in this
CBString, then this algorithm is O(slen). The algorithm does not easily
degenerate into O(slen * strlen(b)) performance except in very uncommon
situations. Thus from a real world perspective, the overhead of
precomputing suffix shifts in the Boyer-Moore algorithm is avoided, while
delivering an unrolled matching inner loop most of the time.
*/
int CBString::find (const char * b, int pos) const {
int ii, j;
unsigned char c0;
int i, l;
unsigned char cx;
unsigned char * pdata;
2017-01-14 08:24:56 +08:00
if (NULL == b) {
#ifdef BSTRLIB_THROWS_EXCEPTIONS
bstringThrow ("CBString::find NULL.");
#else
return BSTR_ERR;
#endif
}
if ((unsigned int) pos > (unsigned int) slen) return BSTR_ERR;
if ('\0' == b[0]) return pos;
if (pos == slen) return BSTR_ERR;
if ('\0' == b[1]) return find (b[0], pos);
cx = c0 = (unsigned char) b[0];
l = slen - 1;
pdata = data;
for (ii = -1, i = pos, j = 0; i < l;) {
/* Unrolled current character test */
if (cx != pdata[i]) {
if (cx != pdata[1+i]) {
i += 2;
continue;
}
i++;
}
/* Take note if this is the start of a potential match */
if (0 == j) ii = i;
/* Shift the test character down by one */
j++;
i++;
/* If this isn't past the last character continue */
if ('\0' != (cx = b[j])) continue;
N0:;
/* If no characters mismatched, then we matched */
if (i == ii+j) return ii;
/* Shift back to the beginning */
i -= j;
j = 0;
cx = c0;
}
/* Deal with last case if unrolling caused a misalignment */
if (i == l && cx == pdata[i] && '\0' == b[j+1]) goto N0;
return BSTR_ERR;
}
int CBString::caselessfind (const CBString& b, int pos) const {
return binstrcaseless ((bstring) this, pos, (bstring) &b);
}
int CBString::caselessfind (const char * b, int pos) const {
struct tagbstring t;
if (NULL == b) {
#ifdef BSTRLIB_THROWS_EXCEPTIONS
bstringThrow ("CBString::caselessfind NULL.");
#else
return BSTR_ERR;
#endif
}
if ((unsigned int) pos > (unsigned int) slen) return BSTR_ERR;
if ('\0' == b[0]) return pos;
if (pos == slen) return BSTR_ERR;
btfromcstr (t, b);
return binstrcaseless ((bstring) this, pos, (bstring) &t);
}
int CBString::find (char c, int pos) const {
if (pos < 0) return BSTR_ERR;
for (;pos < slen; pos++) {
if (data[pos] == (unsigned char) c) return pos;
}
return BSTR_ERR;
}
int CBString::reversefind (const CBString& b, int pos) const {
return binstrr ((bstring) this, pos, (bstring) &b);
}
int CBString::reversefind (const char * b, int pos) const {
struct tagbstring t;
if (NULL == b) {
#ifdef BSTRLIB_THROWS_EXCEPTIONS
bstringThrow ("CBString::reversefind NULL.");
#else
return BSTR_ERR;
#endif
}
cstr2tbstr (t, b);
return binstrr ((bstring) this, pos, &t);
}
int CBString::caselessreversefind (const CBString& b, int pos) const {
return binstrrcaseless ((bstring) this, pos, (bstring) &b);
}
int CBString::caselessreversefind (const char * b, int pos) const {
struct tagbstring t;
if (NULL == b) {
#ifdef BSTRLIB_THROWS_EXCEPTIONS
bstringThrow ("CBString::caselessreversefind NULL.");
#else
return BSTR_ERR;
#endif
}
if ((unsigned int) pos > (unsigned int) slen) return BSTR_ERR;
if ('\0' == b[0]) return pos;
if (pos == slen) return BSTR_ERR;
btfromcstr (t, b);
return binstrrcaseless ((bstring) this, pos, (bstring) &t);
}
int CBString::reversefind (char c, int pos) const {
if (pos > slen) return BSTR_ERR;
if (pos == slen) pos--;
for (;pos >= 0; pos--) {
if (data[pos] == (unsigned char) c) return pos;
}
return BSTR_ERR;
}
int CBString::findchr (const CBString& b, int pos) const {
return binchr ((bstring) this, pos, (bstring) &b);
}
int CBString::findchr (const char * s, int pos) const {
struct tagbstring t;
if (NULL == s) {
#ifdef BSTRLIB_THROWS_EXCEPTIONS
bstringThrow ("CBString::findchr NULL.");
#else
return BSTR_ERR;
#endif
}
cstr2tbstr (t, s);
return binchr ((bstring) this, pos, (bstring) &t);
}
int CBString::nfindchr (const CBString& b, int pos) const {
return bninchr ((bstring) this, pos, (bstring) &b);
}
int CBString::nfindchr (const char * s, int pos) const {
struct tagbstring t;
if (NULL == s) {
#ifdef BSTRLIB_THROWS_EXCEPTIONS
bstringThrow ("CBString::nfindchr NULL.");
#else
return BSTR_ERR;
#endif
}
cstr2tbstr (t, s);
return bninchr ((bstring) this, pos, &t);
}
int CBString::reversefindchr (const CBString& b, int pos) const {
return binchrr ((bstring) this, pos, (bstring) &b);
}
int CBString::reversefindchr (const char * s, int pos) const {
struct tagbstring t;
if (NULL == s) {
#ifdef BSTRLIB_THROWS_EXCEPTIONS
bstringThrow ("CBString::reversefindchr NULL.");
#else
return BSTR_ERR;
#endif
}
cstr2tbstr (t, s);
return binchrr ((bstring) this, pos, &t);
}
int CBString::nreversefindchr (const CBString& b, int pos) const {
return bninchrr ((bstring) this, pos, (bstring) &b);
}
int CBString::nreversefindchr (const char * s, int pos) const {
struct tagbstring t;
if (NULL == s) {
#ifdef BSTRLIB_THROWS_EXCEPTIONS
bstringThrow ("CBString::nreversefindchr NULL.");
#else
return BSTR_ERR;
#endif
}
cstr2tbstr (t, s);
return bninchrr ((bstring) this, pos, &t);
}
CBString CBString::midstr (int left, int len) const {
struct tagbstring t;
if (left < 0) {
len += left;
left = 0;
}
if (len > slen - left) len = slen - left;
if (len <= 0) return CBString ("");
blk2tbstr (t, data + left, len);
return CBString (t);
}
void CBString::alloc (int len) {
if (BSTR_ERR == balloc ((bstring)this, len)) {
bstringThrow ("Failure in alloc");
}
}
void CBString::fill (int len, unsigned char cfill) {
slen = 0;
if (BSTR_ERR == bsetstr (this, len, NULL, cfill)) {
bstringThrow ("Failure in fill");
}
}
void CBString::setsubstr (int pos, const CBString& b, unsigned char cfill) {
if (BSTR_ERR == bsetstr (this, pos, (bstring) &b, cfill)) {
bstringThrow ("Failure in setsubstr");
}
}
void CBString::setsubstr (int pos, const char * s, unsigned char cfill) {
struct tagbstring t;
if (NULL == s) {
#ifdef BSTRLIB_THROWS_EXCEPTIONS
bstringThrow ("setsubstr NULL.");
#else
return;
#endif
}
cstr2tbstr (t, s);
if (BSTR_ERR == bsetstr (this, pos, &t, cfill)) {
bstringThrow ("Failure in setsubstr");
}
}
void CBString::insert (int pos, const CBString& b, unsigned char cfill) {
if (BSTR_ERR == binsert (this, pos, (bstring) &b, cfill)) {
bstringThrow ("Failure in insert");
}
}
void CBString::insert (int pos, const char * s, unsigned char cfill) {
struct tagbstring t;
if (NULL == s) {
#ifdef BSTRLIB_THROWS_EXCEPTIONS
bstringThrow ("insert NULL.");
#else
return;
#endif
}
cstr2tbstr (t, s);
if (BSTR_ERR == binsert (this, pos, &t, cfill)) {
bstringThrow ("Failure in insert");
}
}
void CBString::insertchrs (int pos, int len, unsigned char cfill) {
if (BSTR_ERR == binsertch (this, pos, len, cfill)) {
bstringThrow ("Failure in insertchrs");
}
}
void CBString::replace (int pos, int len, const CBString& b, unsigned char cfill) {
if (BSTR_ERR == breplace (this, pos, len, (bstring) &b, cfill)) {
bstringThrow ("Failure in replace");
}
}
void CBString::replace (int pos, int len, const char * s, unsigned char cfill) {
struct tagbstring t;
size_t q;
if (mlen <= 0) bstringThrow ("Write protection error");
if (NULL == s || (pos|len) < 0) {
bstringThrow ("Failure in replace");
} else {
if (pos + len >= slen) {
cstr2tbstr (t, s);
if (BSTR_ERR == bsetstr (this, pos, &t, cfill)) {
bstringThrow ("Failure in replace");
} else if (pos + t.slen < slen) {
slen = pos + t.slen;
data[slen] = '\0';
}
} else {
/* Aliasing case */
if ((unsigned int) (data - (unsigned char *) s) < (unsigned int) slen) {
replace (pos, len, CBString(s), cfill);
return;
}
if ((q = strlen (s)) > (size_t) len || len < 0) {
if (slen + q - len >= INT_MAX) bstringThrow ("Failure in replace, result too long.");
alloc ((int) (slen + q - len));
if (NULL == data) return;
}
if ((int) q != len) bstr__memmove (data + pos + q, data + pos + len, slen - (pos + len));
bstr__memcpy (data + pos, s, q);
slen += ((int) q) - len;
data[slen] = '\0';
}
}
}
void CBString::findreplace (const CBString& sfind, const CBString& repl, int pos) {
if (BSTR_ERR == bfindreplace (this, (bstring) &sfind, (bstring) &repl, pos)) {
bstringThrow ("Failure in findreplace");
}
}
void CBString::findreplace (const CBString& sfind, const char * repl, int pos) {
struct tagbstring t;
if (NULL == repl) {
#ifdef BSTRLIB_THROWS_EXCEPTIONS
bstringThrow ("findreplace NULL.");
#else
return;
#endif
}
cstr2tbstr (t, repl);
if (BSTR_ERR == bfindreplace (this, (bstring) &sfind, (bstring) &t, pos)) {
bstringThrow ("Failure in findreplace");
}
}
void CBString::findreplace (const char * sfind, const CBString& repl, int pos) {
struct tagbstring t;
if (NULL == sfind) {
#ifdef BSTRLIB_THROWS_EXCEPTIONS
bstringThrow ("findreplace NULL.");
#else
return;
#endif
}
cstr2tbstr (t, sfind);
if (BSTR_ERR == bfindreplace (this, (bstring) &t, (bstring) &repl, pos)) {
bstringThrow ("Failure in findreplace");
}
}
void CBString::findreplace (const char * sfind, const char * repl, int pos) {
struct tagbstring t, u;
if (NULL == repl || NULL == sfind) {
#ifdef BSTRLIB_THROWS_EXCEPTIONS
bstringThrow ("findreplace NULL.");
#else
return;
#endif
}
cstr2tbstr (t, sfind);
cstr2tbstr (u, repl);
if (BSTR_ERR == bfindreplace (this, (bstring) &t, (bstring) &u, pos)) {
bstringThrow ("Failure in findreplace");
}
}
void CBString::findreplacecaseless (const CBString& sfind, const CBString& repl, int pos) {
if (BSTR_ERR == bfindreplacecaseless (this, (bstring) &sfind, (bstring) &repl, pos)) {
bstringThrow ("Failure in findreplacecaseless");
}
}
void CBString::findreplacecaseless (const CBString& sfind, const char * repl, int pos) {
struct tagbstring t;
if (NULL == repl) {
#ifdef BSTRLIB_THROWS_EXCEPTIONS
bstringThrow ("findreplacecaseless NULL.");
#else
return;
#endif
}
cstr2tbstr (t, repl);
if (BSTR_ERR == bfindreplacecaseless (this, (bstring) &sfind, (bstring) &t, pos)) {
bstringThrow ("Failure in findreplacecaseless");
}
}
void CBString::findreplacecaseless (const char * sfind, const CBString& repl, int pos) {
struct tagbstring t;
if (NULL == sfind) {
#ifdef BSTRLIB_THROWS_EXCEPTIONS
bstringThrow ("findreplacecaseless NULL.");
#else
return;
#endif
}
cstr2tbstr (t, sfind);
if (BSTR_ERR == bfindreplacecaseless (this, (bstring) &t, (bstring) &repl, pos)) {
bstringThrow ("Failure in findreplacecaseless");
}
}
void CBString::findreplacecaseless (const char * sfind, const char * repl, int pos) {
struct tagbstring t, u;
if (NULL == repl || NULL == sfind) {
#ifdef BSTRLIB_THROWS_EXCEPTIONS
bstringThrow ("findreplacecaseless NULL.");
#else
return;
#endif
}
cstr2tbstr (t, sfind);
cstr2tbstr (u, repl);
if (BSTR_ERR == bfindreplacecaseless (this, (bstring) &t, (bstring) &u, pos)) {
bstringThrow ("Failure in findreplacecaseless");
}
}
void CBString::remove (int pos, int len) {
if (BSTR_ERR == bdelete (this, pos, len)) {
bstringThrow ("Failure in remove");
}
}
void CBString::trunc (int len) {
if (len < 0) {
bstringThrow ("Failure in trunc");
}
if (len < slen) {
slen = len;
data[len] = '\0';
}
}
void CBString::ltrim (const CBString& b) {
int l = nfindchr (b, 0);
if (l == BSTR_ERR) l = slen;
remove (0, l);
}
void CBString::rtrim (const CBString& b) {
int l = nreversefindchr (b, slen - 1);
#if BSTR_ERR != -1
if (l == BSTR_ERR) l = -1;
#endif
slen = l + 1;
if (mlen > slen) data[slen] = '\0';
}
void CBString::toupper () {
if (BSTR_ERR == btoupper ((bstring) this)) {
bstringThrow ("Failure in toupper");
}
}
void CBString::tolower () {
if (BSTR_ERR == btolower ((bstring) this)) {
bstringThrow ("Failure in tolower");
}
}
void CBString::repeat (int count) {
count *= slen;
if (count == 0) {
trunc (0);
return;
}
if (count < 0 || BSTR_ERR == bpattern (this, count)) {
bstringThrow ("Failure in repeat");
}
}
int CBString::gets (bNgetc getcPtr, void * parm, char terminator) {
if (mlen <= 0) bstringThrow ("Write protection error");
bstring b = bgets (getcPtr, parm, terminator);
if (b == NULL) {
slen = 0;
return -1;
}
*this = *b;
bdestroy (b);
return 0;
}
int CBString::read (bNread readPtr, void * parm) {
if (mlen <= 0) bstringThrow ("Write protection error");
bstring b = bread (readPtr, parm);
if (b == NULL) {
slen = 0;
return -1;
}
*this = *b;
bdestroy (b);
return 0;
}
const CBString operator + (const char *a, const CBString& b) {
return CBString(a) + b;
}
const CBString operator + (const unsigned char *a, const CBString& b) {
return CBString((const char *)a) + b;
}
const CBString operator + (char c, const CBString& b) {
return CBString(c) + b;
}
const CBString operator + (unsigned char c, const CBString& b) {
return CBString(c) + b;
}
const CBString operator + (const tagbstring& x, const CBString& b) {
return CBString(x) + b;
}
void CBString::writeprotect () {
if (mlen >= 0) mlen = -1;
}
void CBString::writeallow () {
if (mlen == -1) mlen = slen + (slen == 0);
else if (mlen < 0) {
bstringThrow ("Cannot unprotect a constant");
}
}
#if defined(BSTRLIB_CAN_USE_STL)
// Constructors.
CBString::CBString (const CBStringList& l) {
int c;
size_t i;
for (c=1, i=0; i < l.size(); i++) {
c += l.at(i).slen;
}
mlen = c;
slen = 0;
data = (unsigned char *) bstr__alloc (c);
if (!data) {
mlen = slen = 0;
bstringThrow ("Failure in (CBStringList) constructor");
} else {
for (i=0; i < l.size(); i++) {
*this += l.at(i);
}
}
}
CBString::CBString (const struct CBStringList& l, const CBString& sep) {
int c, sl = sep.length ();
size_t i;
for (c=1, i=0; i < l.size(); i++) {
c += l.at(i).slen + sl;
}
mlen = c;
slen = 0;
data = (unsigned char *) bstr__alloc (mlen);
if (!data) {
mlen = slen = 0;
bstringThrow ("Failure in (CBStringList) constructor");
} else {
for (i=0; i < l.size(); i++) {
if (i > 0) *this += sep;
*this += l.at(i);
}
}
}
CBString::CBString (const struct CBStringList& l, char sep) {
int c;
size_t i;
for (c=1, i=0; i < l.size(); i++) {
c += l.at(i).slen + 1;
}
mlen = c;
slen = 0;
data = (unsigned char *) bstr__alloc (mlen);
if (!data) {
mlen = slen = 0;
bstringThrow ("Failure in (CBStringList) constructor");
} else {
for (i=0; i < l.size(); i++) {
if (i > 0) *this += sep;
*this += l.at(i);
}
}
}
CBString::CBString (const struct CBStringList& l, unsigned char sep) {
int c;
size_t i;
for (c=1, i=0; i < l.size(); i++) {
c += l.at(i).slen + 1;
}
mlen = c;
slen = 0;
data = (unsigned char *) bstr__alloc (mlen);
if (!data) {
mlen = slen = 0;
bstringThrow ("Failure in (CBStringList) constructor");
} else {
for (i=0; i < l.size(); i++) {
if (i > 0) *this += sep;
*this += l.at(i);
}
}
}
void CBString::join (const struct CBStringList& l) {
int c;
size_t i;
if (mlen <= 0) {
bstringThrow ("Write protection error");
}
for (c=1, i=0; i < l.size(); i++) {
c += l.at(i).slen;
if (c < 0) bstringThrow ("Failure in (CBStringList) constructor, too long");
}
alloc (c);
slen = 0;
if (!data) {
mlen = slen = 0;
bstringThrow ("Failure in (CBStringList) constructor");
} else {
for (i=0; i < l.size(); i++) {
*this += l.at(i);
}
}
}
void CBString::join (const struct CBStringList& l, const CBString& sep) {
int c, sl = sep.length();
size_t i;
if (mlen <= 0) {
bstringThrow ("Write protection error");
}
for (c=1, i=0; i < l.size(); i++) {
c += l.at(i).slen + sl;
if (c < sl) bstringThrow ("Failure in (CBStringList) constructor, too long");
}
alloc (c);
slen = 0;
if (!data) {
mlen = slen = 0;
bstringThrow ("Failure in (CBStringList) constructor");
} else {
for (i=0; i < l.size(); i++) {
if (i > 0) *this += sep;
*this += l.at(i);
}
}
}
void CBString::join (const struct CBStringList& l, char sep) {
int c;
size_t i;
if (mlen <= 0) {
bstringThrow ("Write protection error");
}
for (c=1, i=0; i < l.size(); i++) {
c += l.at(i).slen + 1;
if (c <= 0) bstringThrow ("Failure in (CBStringList) constructor, too long");
}
alloc (c);
slen = 0;
if (!data) {
mlen = slen = 0;
bstringThrow ("Failure in (CBStringList) constructor");
} else {
for (i=0; i < l.size(); i++) {
if (i > 0) *this += sep;
*this += l.at(i);
}
}
}
void CBString::join (const struct CBStringList& l, unsigned char sep) {
int c;
size_t i;
if (mlen <= 0) {
bstringThrow ("Write protection error");
}
for (c=1, i=0; i < l.size(); i++) {
c += l.at(i).slen + 1;
if (c <= 0) bstringThrow ("Failure in (CBStringList) constructor, too long");
}
alloc (c);
slen = 0;
if (!data) {
mlen = slen = 0;
bstringThrow ("Failure in (CBStringList) constructor");
} else {
for (i=0; i < l.size(); i++) {
if (i > 0) *this += sep;
*this += l.at(i);
}
}
}
// Split functions.
void CBStringList::split (const CBString& b, unsigned char splitChar) {
int p, i;
p = 0;
do {
for (i = p; i < b.length (); i++) {
if (b.character (i) == splitChar) break;
}
if (i >= p) this->push_back (CBString (&(b.data[p]), i - p));
p = i + 1;
} while (p <= b.length ());
}
void CBStringList::split (const CBString& b, const CBString& s) {
struct { unsigned long content[(1 << CHAR_BIT) / 32]; } chrs;
unsigned char c;
int p, i;
if (s.length() == 0) bstringThrow ("Null splitstring failure");
if (s.length() == 1) {
this->split (b, s.character (0));
} else {
for (i=0; i < ((1 << CHAR_BIT) / 32); i++) chrs.content[i] = 0x0;
for (i=0; i < s.length(); i++) {
c = s.character (i);
chrs.content[c >> 5] |= ((long)1) << (c & 31);
}
p = 0;
do {
for (i = p; i < b.length (); i++) {
c = b.character (i);
if (chrs.content[c >> 5] & ((long)1) << (c & 31)) break;
}
if (i >= p) this->push_back (CBString (&(b.data[p]), i - p));
p = i + 1;
} while (p <= b.length ());
}
}
void CBStringList::splitstr (const CBString& b, const CBString& s) {
int p, i;
if (s.length() == 1) {
this->split (b, s.character (0));
} else if (s.length() == 0) {
for (i=0; i < b.length (); i++) {
this->push_back (CBString (b.data[i]));
}
} else {
for (p=0; (i = b.find (s, p)) >= 0; p = i + s.length ()) {
this->push_back (b.midstr (p, i - p));
}
if (p <= b.length ()) {
this->push_back (b.midstr (p, b.length () - p));
}
}
}
static int streamSplitCb (void * parm, int ofs, const_bstring entry) {
CBStringList * r = (CBStringList *) parm;
(void)ofs;
2017-01-14 08:24:56 +08:00
r->push_back (CBString (*entry));
return 0;
}
void CBStringList::split (const CBStream& b, const CBString& s) {
if (0 > bssplitscb (b.m_s, (bstring) &s, streamSplitCb,
(void *) this)) {
bstringThrow ("Split bstream failure");
}
}
void CBStringList::split (const CBStream& b, unsigned char splitChar) {
CBString sc (splitChar);
if (0 > bssplitscb (b.m_s, (bstring) &sc,
streamSplitCb, (void *) this)) {
bstringThrow ("Split bstream failure");
}
}
void CBStringList::splitstr (const CBStream& b, const CBString& s) {
if (0 > bssplitstrcb (b.m_s, (bstring) &s, streamSplitCb,
(void *) this)) {
bstringThrow ("Split bstream failure");
}
}
#endif
#if defined(BSTRLIB_CAN_USE_IOSTREAM)
std::ostream& operator << (std::ostream& sout, CBString b) {
return sout.write ((const char *)b, b.length());
}
#include <ctype.h>
static int istreamGets (void * parm) {
char c = '\n';
((std::istream *)parm)->get(c);
if (isspace (c)) c = '\n';
return c;
}
std::istream& operator >> (std::istream& sin, CBString& b) {
do {
b.gets ((bNgetc) istreamGets, &sin, '\n');
if (b.slen > 0 && b.data[b.slen-1] == '\n') b.slen--;
} while (b.slen == 0 && !sin.eof ());
return sin;
}
struct sgetc {
std::istream * sin;
char terminator;
};
static int istreamGetc (void * parm) {
char c = ((struct sgetc *)parm)->terminator;
((struct sgetc *)parm)->sin->get(c);
return c;
}
std::istream& getline (std::istream& sin, CBString& b, char terminator) {
struct sgetc parm;
parm.sin = &sin;
parm.terminator = terminator;
b.gets ((bNgetc) istreamGetc, &parm, terminator);
if (b.slen > 0 && b.data[b.slen-1] == terminator) b.slen--;
return sin;
}
#endif
CBStream::CBStream (bNread readPtr, void * parm) {
m_s = bsopen (readPtr, parm);
}
CBStream::~CBStream () {
bsclose (m_s);
}
int CBStream::buffLengthSet (int sz) {
if (sz <= 0) {
bstringThrow ("buffLengthSet parameter failure");
}
return bsbufflength (m_s, sz);
}
int CBStream::buffLengthGet () {
return bsbufflength (m_s, 0);
}
CBString CBStream::readLine (char terminator) {
CBString ret("");
if (0 > bsreadln ((bstring) &ret, m_s, terminator) && eof () < 0) {
bstringThrow ("Failed readLine");
}
return ret;
}
CBString CBStream::readLine (const CBString& terminator) {
CBString ret("");
if (0 > bsreadlns ((bstring) &ret, m_s, (bstring) &terminator) && eof () < 0) {
bstringThrow ("Failed readLine");
}
return ret;
}
void CBStream::readLine (CBString& s, char terminator) {
if (0 > bsreadln ((bstring) &s, m_s, terminator) && eof () < 0) {
bstringThrow ("Failed readLine");
}
}
void CBStream::readLine (CBString& s, const CBString& terminator) {
if (0 > bsreadlns ((bstring) &s, m_s, (bstring) &terminator) && eof () < 0) {
bstringThrow ("Failed readLine");
}
}
void CBStream::readLineAppend (CBString& s, char terminator) {
if (0 > bsreadlna ((bstring) &s, m_s, terminator) && eof () < 0) {
bstringThrow ("Failed readLineAppend");
}
}
void CBStream::readLineAppend (CBString& s, const CBString& terminator) {
if (0 > bsreadlnsa ((bstring) &s, m_s, (bstring) &terminator) && eof () < 0) {
bstringThrow ("Failed readLineAppend");
}
}
#define BS_BUFF_SZ (1024)
CBString CBStream::read () {
CBString ret("");
while (!bseof (m_s)) {
if (0 > bsreada ((bstring) &ret, m_s, BS_BUFF_SZ) && eof () < 0) {
bstringThrow ("Failed read");
}
}
return ret;
}
CBString& CBStream::operator >> (CBString& s) {
while (!bseof (m_s)) {
if (0 > bsreada ((bstring) &s, m_s, BS_BUFF_SZ) && eof () < 0) {
bstringThrow ("Failed read");
}
}
return s;
}
CBString CBStream::read (int n) {
CBString ret("");
if (0 > bsread ((bstring) &ret, m_s, n) && eof () < 0) {
bstringThrow ("Failed read");
}
return ret;
}
void CBStream::read (CBString& s) {
s.slen = 0;
while (!bseof (m_s)) {
if (0 > bsreada ((bstring) &s, m_s, BS_BUFF_SZ)) {
bstringThrow ("Failed read");
}
}
}
void CBStream::read (CBString& s, int n) {
if (0 > bsread ((bstring) &s, m_s, n)) {
bstringThrow ("Failed read");
}
}
void CBStream::readAppend (CBString& s) {
while (!bseof (m_s)) {
if (0 > bsreada ((bstring) &s, m_s, BS_BUFF_SZ)) {
bstringThrow ("Failed readAppend");
}
}
}
void CBStream::readAppend (CBString& s, int n) {
if (0 > bsreada ((bstring) &s, m_s, n)) {
bstringThrow ("Failed readAppend");
}
}
void CBStream::unread (const CBString& s) {
if (0 > bsunread (m_s, (bstring) &s)) {
bstringThrow ("Failed unread");
}
}
CBString CBStream::peek () const {
CBString ret ("");
if (0 > bspeek ((bstring) &ret, m_s)) {
bstringThrow ("Failed peek");
}
return ret;
}
void CBStream::peek (CBString& s) const {
s.slen = 0;
if (0 > bspeek ((bstring) &s, m_s)) {
bstringThrow ("Failed peek");
}
}
void CBStream::peekAppend (CBString& s) const {
if (0 > bspeek ((bstring) &s, m_s)) {
bstringThrow ("Failed peekAppend");
}
}
int CBStream::eof () const {
int ret = bseof (m_s);
if (0 > ret) {
bstringThrow ("Failed eof");
}
return ret;
}
} // namespace Bstrlib