mirror of
https://github.com/WinampDesktop/winamp.git
synced 2024-12-18 16:55:54 +01:00
974 lines
19 KiB
C++
974 lines
19 KiB
C++
#include "../nde.h"
|
|
#include "../NDEString.h"
|
|
#include "Query.h"
|
|
|
|
//---------------------------------------------------------------------------
|
|
|
|
bool Scanner::Query(const char *query)
|
|
{
|
|
if (!query) return false;
|
|
ndestring_release(last_query);
|
|
last_query = ndestring_wcsdup(query);
|
|
RemoveFilters();
|
|
in_query_parser = 1;
|
|
bool r = Query_Parse(query);
|
|
|
|
if (r == false)
|
|
{
|
|
if (!disable_date_resolution) RemoveFilters();
|
|
last_query_failed = true;
|
|
}
|
|
in_query_parser = 0;
|
|
Query_CleanUp();
|
|
return r & CheckFilters();
|
|
}
|
|
|
|
const char *Scanner::GetLastQuery()
|
|
{
|
|
return last_query;
|
|
}
|
|
|
|
|
|
typedef struct
|
|
{
|
|
const char *token;
|
|
int tid;
|
|
} tokenstruct;
|
|
|
|
tokenstruct Tokens[] = // Feel free to add more...
|
|
{
|
|
{"AND", TOKEN_AND },
|
|
{"OR", TOKEN_OR },
|
|
{"HAS", TOKEN_CONTAINS },
|
|
{"NOTHAS",TOKEN_NOTCONTAINS},
|
|
{"BEGINS", TOKEN_BEGINS },
|
|
{"ENDS", TOKEN_ENDS },
|
|
{"ISEMPTY", TOKEN_ISEMPTY},
|
|
{"ISNOTEMPTY",TOKEN_ISNOTEMPTY},
|
|
{"LIKE", TOKEN_LIKE},
|
|
{"BEGINSLIKE", TOKEN_BEGINSLIKE},
|
|
};
|
|
|
|
|
|
typedef struct
|
|
{
|
|
int Op;
|
|
int Level;
|
|
} OpLevel;
|
|
|
|
static int Query_ParseLength(const char *str)
|
|
{
|
|
int i = atoi(str);
|
|
|
|
const char *p;
|
|
if ((p=strstr(str,":")))
|
|
{
|
|
i*=60;
|
|
i+=atoi(++p);
|
|
if ((p=strstr(p,":")))
|
|
{
|
|
i*=60;
|
|
i+=atoi(++p);
|
|
}
|
|
}
|
|
return i;
|
|
}
|
|
|
|
/*
|
|
our state machine
|
|
|
|
&, |
|
|
----------<-------------------------<----------------------<-----------
|
|
| |
|
|
v ID (Col) =, >, <... ID / Data / [f] ) |
|
|
---->(0) ----->-----> (1) ------>-----> (2) ------>------> (3) ------>-----> (4) <--
|
|
| |^ \isempty------------->------------/ |^ | |
|
|
| !( || ||---- | ) |
|
|
--<-- ---------<---------------------------<-------------<-| | -->--
|
|
&, | v [f] |
|
|
-->--
|
|
|
|
*/
|
|
|
|
//---------------------------------------------------------------------------
|
|
bool Scanner::Query_Parse(const char *query)
|
|
{
|
|
const char *p = query; // pointer on next token to read
|
|
int size;
|
|
int state = 0;
|
|
int pcount = 0;
|
|
VListEntry<OpLevel> *entry;
|
|
|
|
if (pstack.GetNElements() > 0)
|
|
Query_CleanUp();
|
|
|
|
while (1)
|
|
{
|
|
p = Query_EatSpace(p);
|
|
int t = Query_GetNextToken(p, &size, &token);
|
|
if (t == TOKEN_UNKNOWN)
|
|
{
|
|
Query_SyntaxError((int)(p-query));
|
|
return false;
|
|
}
|
|
if (t == TOKEN_EOQ)
|
|
break;
|
|
switch (state)
|
|
{
|
|
case 0:
|
|
switch (t)
|
|
{
|
|
case TOKEN_PAROPEN:
|
|
state = 0;
|
|
// check too many parenthesis open
|
|
if (pcount == 255)
|
|
{
|
|
Query_SyntaxError((int)(p-query)); // should not be _syntax_ error
|
|
return false;
|
|
}
|
|
// up one level
|
|
pcount++;
|
|
break;
|
|
case TOKEN_NOT:
|
|
// push not in this level
|
|
OpLevel o;
|
|
o.Op = FILTER_NOT;
|
|
o.Level = pcount;
|
|
entry = new VListEntry<OpLevel>;
|
|
entry->SetVal(o);
|
|
pstack.AddEntry(entry, true);
|
|
state = 0;
|
|
break;
|
|
case TOKEN_IDENTIFIER:
|
|
state = 1;
|
|
// create filter column
|
|
|
|
if (AddFilterByName(token, NULL, FILTER_NONE) == ADDFILTER_FAILED)
|
|
{
|
|
Query_SyntaxError((int)(p-query));
|
|
return false;
|
|
}
|
|
|
|
break;
|
|
default:
|
|
Query_SyntaxError((int)(p-query));
|
|
return false;
|
|
}
|
|
break;
|
|
case 1:
|
|
switch (t)
|
|
{
|
|
case TOKEN_EQUAL:
|
|
{
|
|
state = 2;
|
|
// set filter op
|
|
Filter *f = GetLastFilter();
|
|
f->SetOp(FILTER_EQUALS);
|
|
break;
|
|
}
|
|
case TOKEN_ABOVE:
|
|
{
|
|
state = 2;
|
|
// set filter op
|
|
Filter *f = GetLastFilter();
|
|
f->SetOp(FILTER_ABOVE);
|
|
break;
|
|
}
|
|
case TOKEN_BELOW:
|
|
{
|
|
state = 2;
|
|
// set filter op
|
|
Filter *f = GetLastFilter();
|
|
f->SetOp(FILTER_BELOW);
|
|
break;
|
|
}
|
|
case TOKEN_CONTAINS:
|
|
{
|
|
state = 2;
|
|
// set filter op
|
|
Filter *f = GetLastFilter();
|
|
f->SetOp(FILTER_CONTAINS);
|
|
break;
|
|
}
|
|
case TOKEN_NOTCONTAINS:
|
|
{
|
|
state = 2;
|
|
// set filter op
|
|
Filter *f = GetLastFilter();
|
|
f->SetOp(FILTER_NOTCONTAINS);
|
|
break;
|
|
}
|
|
case TOKEN_AOREQUAL:
|
|
{
|
|
state = 2;
|
|
// set filter op
|
|
Filter *f = GetLastFilter();
|
|
f->SetOp(FILTER_ABOVEOREQUAL);
|
|
break;
|
|
}
|
|
case TOKEN_BOREQUAL:
|
|
{
|
|
state = 2;
|
|
// set filter op
|
|
Filter *f = GetLastFilter();
|
|
f->SetOp(FILTER_BELOWOREQUAL);
|
|
break;
|
|
}
|
|
case TOKEN_NOTEQUAL:
|
|
{
|
|
state = 2;
|
|
// set filter op
|
|
Filter *f = GetLastFilter();
|
|
f->SetOp(FILTER_NOTEQUALS);
|
|
break;
|
|
}
|
|
case TOKEN_BEGINS:
|
|
{
|
|
state = 2;
|
|
Filter *f = GetLastFilter();
|
|
f->SetOp(FILTER_BEGINS);
|
|
}
|
|
break;
|
|
case TOKEN_ENDS:
|
|
{
|
|
state = 2;
|
|
Filter *f = GetLastFilter();
|
|
f->SetOp(FILTER_ENDS);
|
|
}
|
|
break;
|
|
case TOKEN_LIKE:
|
|
{
|
|
state = 2;
|
|
// set filter op
|
|
Filter *f = GetLastFilter();
|
|
f->SetOp(FILTER_LIKE);
|
|
break;
|
|
}
|
|
case TOKEN_BEGINSLIKE:
|
|
{
|
|
state = 2;
|
|
// set filter op
|
|
Filter *f = GetLastFilter();
|
|
f->SetOp(FILTER_BEGINSLIKE);
|
|
break;
|
|
}
|
|
case TOKEN_ISNOTEMPTY:
|
|
case TOKEN_ISEMPTY:
|
|
{
|
|
state = 3;
|
|
Filter *f = GetLastFilter();
|
|
f->SetOp(t==TOKEN_ISEMPTY ? FILTER_ISEMPTY : FILTER_ISNOTEMPTY);
|
|
// pop all operators in this level beginning by the last inserted
|
|
entry = (VListEntry<OpLevel> *)pstack.GetFoot();
|
|
while (entry)
|
|
{
|
|
if (entry->GetVal().Level == pcount)
|
|
{
|
|
AddFilterOp(entry->GetVal().Op);
|
|
VListEntry<OpLevel> *_entry = (VListEntry<OpLevel> *)entry->GetPrevious();
|
|
pstack.RemoveEntry(entry);
|
|
entry = _entry;
|
|
}
|
|
else
|
|
break;
|
|
}
|
|
}
|
|
break;
|
|
default:
|
|
Query_SyntaxError((int)(p-query));
|
|
return false;
|
|
}
|
|
break;
|
|
case 2:
|
|
if (t == TOKEN_SQBRACKETOPEN)
|
|
{
|
|
state = 3;
|
|
const char *s = strchr(p, ']');
|
|
if (!s)
|
|
{
|
|
Query_SyntaxError((int)(p-query));
|
|
return false;
|
|
}
|
|
p = Query_EatSpace(p);
|
|
if (*p == '[') p++;
|
|
|
|
char *format = ndestring_malloc((s-p+1)*sizeof(char));
|
|
strncpy(format, p, s-p);
|
|
format[s-p] = 0;
|
|
|
|
Filter *f = GetLastFilter();
|
|
int id = f->GetId();
|
|
ColumnField *c = GetColumnById(id);
|
|
int tt = c ? c->GetDataType() : -1;
|
|
if (disable_date_resolution || !c || (tt != FIELD_INTEGER && tt != FIELD_DATETIME && tt != FIELD_LENGTH && tt != FIELD_BOOLEAN))
|
|
{
|
|
|
|
if (disable_date_resolution)
|
|
{
|
|
StringField *field = (StringField *)f->Data();
|
|
|
|
if (!field)
|
|
{
|
|
// format was used without a value, assume value is 0
|
|
f->SetData(new StringField(""));
|
|
entry = (VListEntry<OpLevel> *)pstack.GetFoot();
|
|
while (entry)
|
|
{
|
|
if (entry->GetVal().Level == pcount)
|
|
{
|
|
AddFilterOp(entry->GetVal().Op);
|
|
VListEntry<OpLevel> *_entry = (VListEntry<OpLevel> *)entry->GetPrevious();
|
|
pstack.RemoveEntry(entry);
|
|
entry = _entry;
|
|
}
|
|
else
|
|
break;
|
|
}
|
|
field = (StringField*)f->Data();
|
|
}
|
|
|
|
field->SetNDEString(format);
|
|
ndestring_release(format);
|
|
p = s+1;
|
|
continue;
|
|
}
|
|
|
|
|
|
ndestring_release(format);
|
|
Query_SyntaxError((int)(p-query));
|
|
return false;
|
|
}
|
|
|
|
IntegerField *field = (IntegerField *)f->Data();
|
|
|
|
if (!field)
|
|
{
|
|
// format was used without a value, assume value is 0
|
|
f->SetData(new IntegerField(0));
|
|
entry = (VListEntry<OpLevel> *)pstack.GetFoot();
|
|
while (entry)
|
|
{
|
|
if (entry->GetVal().Level == pcount)
|
|
{
|
|
AddFilterOp(entry->GetVal().Op);
|
|
VListEntry<OpLevel> *_entry = (VListEntry<OpLevel> *)entry->GetPrevious();
|
|
pstack.RemoveEntry(entry);
|
|
entry = _entry;
|
|
}
|
|
else
|
|
break;
|
|
}
|
|
field = (IntegerField *)f->Data();
|
|
}
|
|
int r = field->ApplyConversion(format);
|
|
ndestring_release(format);
|
|
if (!r)
|
|
{
|
|
Query_SyntaxError((int)(p-query));
|
|
return false;
|
|
}
|
|
p = s+1;
|
|
continue;
|
|
}
|
|
// switch (t) {
|
|
// case TOKEN_IDENTIFIER:
|
|
else // JF> we make this relaxed, so anything is valid as a value
|
|
{
|
|
state = 3;
|
|
// set filter data
|
|
Filter *f = GetLastFilter();
|
|
int id = f->GetId();
|
|
ColumnField *c = GetColumnById(id);
|
|
switch (c ? c->GetDataType() : -1)
|
|
{
|
|
case FIELD_DATETIME:
|
|
if (disable_date_resolution)
|
|
goto field_string_override;
|
|
case FIELD_LENGTH:
|
|
{
|
|
int i;
|
|
IntegerField *i_f = new IntegerField();
|
|
i = Query_ParseLength(token);
|
|
i_f->SetValue(i);
|
|
f->SetData(i_f);
|
|
}
|
|
break;
|
|
|
|
case FIELD_BOOLEAN:
|
|
case FIELD_INTEGER:
|
|
{
|
|
int i;
|
|
IntegerField *i_f = new IntegerField();
|
|
i = atoi(token);
|
|
i_f->SetValue(i);
|
|
f->SetData(i_f);
|
|
}
|
|
break;
|
|
case FIELD_INT64:
|
|
{
|
|
int64_t i;
|
|
Int64Field *i_f = new Int64Field();
|
|
i = strtoull(token, 0, 10); // todo: Replace with own conversion and error checking
|
|
i_f->SetValue(i);
|
|
f->SetData(i_f);
|
|
}
|
|
break;
|
|
case FIELD_FILENAME:
|
|
{
|
|
FilenameField *s_f = new FilenameField();
|
|
s_f->SetNDEString(token);
|
|
f->SetData(s_f);
|
|
}
|
|
break;
|
|
case FIELD_STRING:
|
|
field_string_override:
|
|
{
|
|
StringField *s_f = new StringField();
|
|
s_f->SetNDEString(token);
|
|
f->SetData(s_f);
|
|
}
|
|
break;
|
|
default:
|
|
Query_SyntaxError((int)(p-query));
|
|
return false;
|
|
break;
|
|
|
|
}
|
|
// pop all operators in this level beginning by the last inserted
|
|
entry = (VListEntry<OpLevel> *)pstack.GetFoot();
|
|
while (entry)
|
|
{
|
|
if (entry->GetVal().Level == pcount)
|
|
{
|
|
AddFilterOp(entry->GetVal().Op);
|
|
VListEntry<OpLevel> *_entry = (VListEntry<OpLevel> *)entry->GetPrevious();
|
|
pstack.RemoveEntry(entry);
|
|
entry = _entry;
|
|
}
|
|
else
|
|
break;
|
|
}
|
|
break;
|
|
}
|
|
// default:
|
|
// Query_SyntaxError(p-query);
|
|
// return false;
|
|
// }
|
|
break;
|
|
case 3:
|
|
switch (t)
|
|
{
|
|
case TOKEN_SQBRACKETOPEN:
|
|
{
|
|
const char *s = strchr(p, ']');
|
|
if (!s)
|
|
{
|
|
Query_SyntaxError((int)(p-query));
|
|
return false;
|
|
}
|
|
p = Query_EatSpace(p);
|
|
if (*p == '[') p++;
|
|
char *format = ndestring_malloc((s-p+1)*sizeof(char));
|
|
strncpy(format, p, s-p);
|
|
format[s-p] = 0;
|
|
Filter *f = GetLastFilter();
|
|
int id = f->GetId();
|
|
ColumnField *c = GetColumnById(id);
|
|
int tt = c ? c->GetDataType() : -1;
|
|
if (disable_date_resolution || !c || (tt != FIELD_INTEGER && tt != FIELD_DATETIME && tt != FIELD_LENGTH && tt != FIELD_BOOLEAN))
|
|
{
|
|
if (disable_date_resolution)
|
|
{
|
|
StringField *field = (StringField *)f->Data();
|
|
|
|
if (!field)
|
|
{
|
|
// format was used without a value, assume value is 0
|
|
f->SetData(new StringField(""));
|
|
entry = (VListEntry<OpLevel> *)pstack.GetFoot();
|
|
while (entry)
|
|
{
|
|
if (entry->GetVal().Level == pcount)
|
|
{
|
|
AddFilterOp(entry->GetVal().Op);
|
|
VListEntry<OpLevel> *_entry = (VListEntry<OpLevel> *)entry->GetPrevious();
|
|
pstack.RemoveEntry(entry);
|
|
entry = _entry;
|
|
}
|
|
else
|
|
break;
|
|
}
|
|
field = (StringField *)f->Data();
|
|
}
|
|
|
|
field->SetNDEString(format);
|
|
ndestring_release(format);
|
|
p = s+1;
|
|
continue;
|
|
}
|
|
ndestring_release(format);
|
|
Query_SyntaxError((int)(p-query));
|
|
return false;
|
|
}
|
|
|
|
IntegerField *field = (IntegerField *)f->Data();
|
|
|
|
if (!field)
|
|
{
|
|
// format was used without a value, assume value is 0
|
|
f->SetData(new IntegerField(0));
|
|
entry = (VListEntry<OpLevel> *)pstack.GetFoot();
|
|
while (entry)
|
|
{
|
|
if (entry->GetVal().Level == pcount)
|
|
{
|
|
AddFilterOp(entry->GetVal().Op);
|
|
VListEntry<OpLevel> *_entry = (VListEntry<OpLevel> *)entry->GetPrevious();
|
|
pstack.RemoveEntry(entry);
|
|
entry = _entry;
|
|
}
|
|
else
|
|
break;
|
|
}
|
|
field = (IntegerField *)f->Data();
|
|
}
|
|
int r = field->ApplyConversion(format);
|
|
ndestring_release(format);
|
|
if (!r)
|
|
{
|
|
Query_SyntaxError((int)(p-query));
|
|
return false;
|
|
}
|
|
p = s+1;
|
|
continue;
|
|
}
|
|
break;
|
|
|
|
case TOKEN_PARCLOSE:
|
|
state = 4;
|
|
// check parenthesis count
|
|
if (pcount == 0)
|
|
{
|
|
Query_SyntaxError((int)(p-query));
|
|
return false;
|
|
}
|
|
// down one level
|
|
pcount--;
|
|
// pop all operators in this level, beginning by the last inserted
|
|
while (entry)
|
|
{
|
|
if (entry->GetVal().Level == pcount)
|
|
{
|
|
AddFilterOp(entry->GetVal().Op);
|
|
VListEntry<OpLevel> *_entry = (VListEntry<OpLevel> *)entry->GetPrevious();
|
|
pstack.RemoveEntry(entry);
|
|
entry = _entry;
|
|
}
|
|
else
|
|
break;
|
|
}
|
|
break;
|
|
case TOKEN_AND:
|
|
{
|
|
state = 0;
|
|
// push and
|
|
OpLevel o;
|
|
o.Op = FILTER_AND;
|
|
o.Level = pcount;
|
|
entry = new VListEntry<OpLevel>;
|
|
entry->SetVal(o);
|
|
pstack.AddEntry(entry, true);
|
|
break;
|
|
}
|
|
case TOKEN_OR:
|
|
{
|
|
state = 0;
|
|
// push or
|
|
OpLevel o;
|
|
o.Op = FILTER_OR;
|
|
o.Level = pcount;
|
|
entry = new VListEntry<OpLevel>;
|
|
entry->SetVal(o);
|
|
pstack.AddEntry(entry, true);
|
|
break;
|
|
}
|
|
default:
|
|
Query_SyntaxError((int)(p-query));
|
|
return false;
|
|
}
|
|
break;
|
|
case 4:
|
|
switch (t)
|
|
{
|
|
case TOKEN_AND:
|
|
{
|
|
state = 0;
|
|
// push and
|
|
OpLevel o;
|
|
o.Op = FILTER_AND;
|
|
o.Level = pcount;
|
|
entry = new VListEntry<OpLevel>;
|
|
entry->SetVal(o);
|
|
pstack.AddEntry(entry, true);
|
|
break;
|
|
}
|
|
case TOKEN_OR:
|
|
{
|
|
state = 0;
|
|
// push or
|
|
OpLevel o;
|
|
o.Op = FILTER_OR;
|
|
o.Level = pcount;
|
|
entry = new VListEntry<OpLevel>;
|
|
entry->SetVal(o);
|
|
pstack.AddEntry(entry, true);
|
|
break;
|
|
}
|
|
case TOKEN_PARCLOSE:
|
|
state = 4;
|
|
// check parenthesis count
|
|
if (pcount == 0)
|
|
{
|
|
Query_SyntaxError((int)(p-query));
|
|
return false;
|
|
}
|
|
// down one level
|
|
pcount--;
|
|
// pop all operators in this level, beginning by the last inserted
|
|
while (entry)
|
|
{
|
|
if (entry->GetVal().Level == pcount)
|
|
{
|
|
AddFilterOp(entry->GetVal().Op);
|
|
VListEntry<OpLevel> *_entry = (VListEntry<OpLevel> *)entry->GetPrevious();
|
|
pstack.RemoveEntry(entry);
|
|
entry = _entry;
|
|
}
|
|
else
|
|
break;
|
|
}
|
|
break;
|
|
default:
|
|
Query_SyntaxError((int)(p-query));
|
|
return false;
|
|
}
|
|
break;
|
|
default:
|
|
// Ahem... :/
|
|
break;
|
|
}
|
|
p += size;
|
|
}
|
|
if (pcount > 0)
|
|
{
|
|
Query_SyntaxError((int)(p-query));
|
|
return false;
|
|
}
|
|
return true;
|
|
}
|
|
|
|
//---------------------------------------------------------------------------
|
|
|
|
void Scanner::Query_SyntaxError(int c)
|
|
{
|
|
|
|
}
|
|
|
|
//---------------------------------------------------------------------------
|
|
void Scanner::Query_CleanUp()
|
|
{
|
|
while (pstack.GetNElements() > 0)
|
|
{
|
|
VListEntry<int> *e;
|
|
e = (VListEntry<int> *)pstack.GetHead();
|
|
pstack.RemoveEntry(e);
|
|
}
|
|
}
|
|
|
|
|
|
//---------------------------------------------------------------------------
|
|
const char *Scanner::Query_EatSpace(const char *p)
|
|
{
|
|
while (*p && *p == ' ') p++;
|
|
return p;
|
|
}
|
|
|
|
//---------------------------------------------------------------------------
|
|
const char *Scanner::Query_ProbeNonAlphaNum(const char *p)
|
|
{
|
|
int inquote=0;
|
|
while (*p && (!Query_isControlChar(*p) || (inquote)))
|
|
{
|
|
if (*p == '\"')
|
|
{
|
|
if (!inquote)
|
|
inquote = 1;
|
|
else
|
|
return p+1;
|
|
}
|
|
p++;
|
|
}
|
|
return p;
|
|
}
|
|
|
|
//---------------------------------------------------------------------------
|
|
int Scanner::Query_isControlChar(char p)
|
|
{
|
|
switch (p)
|
|
{
|
|
case '&':
|
|
case '|':
|
|
case '!':
|
|
case '(':
|
|
case '[':
|
|
case ')':
|
|
case ']':
|
|
case '>':
|
|
case '<':
|
|
case '=':
|
|
case ',':
|
|
case ' ':
|
|
return true;
|
|
}
|
|
return false;
|
|
}
|
|
|
|
//---------------------------------------------------------------------------
|
|
char *Scanner::Query_ProbeAlphaNum(char *p)
|
|
{
|
|
while (*p && Query_isControlChar(*p)) p++;
|
|
return p;
|
|
}
|
|
|
|
//---------------------------------------------------------------------------
|
|
char *Scanner::Query_ProbeSpace(char *p)
|
|
{
|
|
while (*p && *p != ' ') p++;
|
|
return p;
|
|
}
|
|
|
|
//---------------------------------------------------------------------------
|
|
int Scanner::Query_LookupToken(const char *t)
|
|
{
|
|
for (int i=0;i<sizeof(Tokens)/sizeof(tokenstruct);i++)
|
|
{
|
|
if (!_stricmp(Tokens[i].token, t))
|
|
return Tokens[i].tid;
|
|
}
|
|
return TOKEN_IDENTIFIER;
|
|
}
|
|
|
|
//---------------------------------------------------------------------------
|
|
|
|
int Scanner::Query_GetNextToken(const char *p, int *size, char **_token, int tokentable)
|
|
{
|
|
|
|
int t = TOKEN_EOQ;
|
|
const char *startptr = p;
|
|
|
|
if (!*p) return TOKEN_EOQ;
|
|
|
|
p = Query_EatSpace(p);
|
|
|
|
const char *e = Query_ProbeNonAlphaNum(p);
|
|
|
|
if (e != p) // We have a word
|
|
{
|
|
size_t token_length = e-p;
|
|
if (*_token) ndestring_release(*_token);
|
|
*_token = ndestring_wcsndup(p, token_length);
|
|
if (*(*_token) == '\"' && (*_token)[token_length-1] == '\"') // check for quoted string
|
|
{
|
|
size_t l=token_length-2;
|
|
if (l>0)
|
|
{
|
|
memcpy(*_token,(*_token)+1,l*sizeof(char));
|
|
(*_token)[l]=0;
|
|
Query_Unescape(*_token);
|
|
}
|
|
else
|
|
(*_token)[0]=0;// we have an empty string
|
|
}
|
|
|
|
switch (tokentable)
|
|
{
|
|
case -1:
|
|
t = TOKEN_IDENTIFIER;
|
|
break;
|
|
case 0:
|
|
t = Query_LookupToken(*_token);
|
|
break;
|
|
case 1:
|
|
t = IntegerField::LookupToken(*_token);
|
|
}
|
|
p = e;
|
|
}
|
|
else // We have a symbol
|
|
{
|
|
switch (*p)
|
|
{
|
|
case '&':
|
|
if (*(p+1) == '&') p++;
|
|
t = TOKEN_AND;
|
|
break;
|
|
case '|':
|
|
if (*(p+1) == '|') p++;
|
|
t = TOKEN_OR;
|
|
break;
|
|
case '!':
|
|
if (*(p+1) == '=')
|
|
{
|
|
p++;
|
|
t = TOKEN_NOTEQUAL;
|
|
break;
|
|
}
|
|
t = TOKEN_NOT;
|
|
break;
|
|
case '(':
|
|
t = TOKEN_PAROPEN;
|
|
break;
|
|
case ')':
|
|
t = TOKEN_PARCLOSE;
|
|
break;
|
|
case '[':
|
|
t = TOKEN_SQBRACKETOPEN;
|
|
break;
|
|
case ']':
|
|
t = TOKEN_SQBRACKETCLOSE;
|
|
break;
|
|
case ',':
|
|
t = TOKEN_COMMA;
|
|
break;
|
|
case '>':
|
|
if (*(p+1) == '=')
|
|
{
|
|
p++;
|
|
t = TOKEN_AOREQUAL;
|
|
break;
|
|
}
|
|
if (*(p+1) == '<')
|
|
{
|
|
p++;
|
|
t = TOKEN_NOTEQUAL;
|
|
break;
|
|
}
|
|
t = TOKEN_ABOVE;
|
|
break;
|
|
case '<':
|
|
if (*(p+1) == '=')
|
|
{
|
|
p++;
|
|
t = TOKEN_BOREQUAL;
|
|
break;
|
|
}
|
|
if (*(p+1) == '>')
|
|
{
|
|
p++;
|
|
t = TOKEN_NOTEQUAL;
|
|
break;
|
|
}
|
|
t = TOKEN_BELOW;
|
|
break;
|
|
case '=':
|
|
if (*(p+1) == '>')
|
|
{
|
|
p++;
|
|
t = TOKEN_AOREQUAL;
|
|
break;
|
|
}
|
|
if (*(p+1) == '<')
|
|
{
|
|
p++;
|
|
t = TOKEN_BOREQUAL;
|
|
break;
|
|
}
|
|
if (*(p+1) == '!')
|
|
{
|
|
p++;
|
|
t = TOKEN_NOTEQUAL;
|
|
break;
|
|
}
|
|
if (*(p+1) == '=') p++;
|
|
t = TOKEN_EQUAL;
|
|
break;
|
|
default:
|
|
t = TOKEN_UNKNOWN;
|
|
break;
|
|
}
|
|
p++;
|
|
}
|
|
|
|
*size = (int)(p - startptr);
|
|
return t;
|
|
}
|
|
|
|
static uint8_t quickhex(char c)
|
|
{
|
|
int hexvalue = c;
|
|
if (hexvalue & 0x10)
|
|
hexvalue &= ~0x30;
|
|
else
|
|
{
|
|
hexvalue &= 0xF;
|
|
hexvalue += 9;
|
|
}
|
|
return hexvalue;
|
|
}
|
|
|
|
static uint8_t DecodeEscape(const char *&str)
|
|
{
|
|
uint8_t a = quickhex(*++str);
|
|
uint8_t b = quickhex(*++str);
|
|
str++;
|
|
return a * 16 + b;
|
|
}
|
|
|
|
static void DecodeEscapedUTF8(char *&output, const char *&input)
|
|
{
|
|
bool error=false;
|
|
|
|
while (*input == '%')
|
|
{
|
|
if (isxdigit(input[1]) && isxdigit(input[2]))
|
|
{
|
|
*output++=DecodeEscape(input);
|
|
}
|
|
else if (input[1] == '%')
|
|
{
|
|
input+=2;
|
|
*output++='%';
|
|
}
|
|
else
|
|
{
|
|
error = true;
|
|
break;
|
|
}
|
|
}
|
|
|
|
if (error)
|
|
{
|
|
*output++ = *input++;
|
|
}
|
|
}
|
|
|
|
// benski> We have the luxury of knowing that decoding will ALWAYS produce smaller strings
|
|
// so we can do it in-place
|
|
void Query_Unescape(char *str)
|
|
{
|
|
const char *itr = str;
|
|
while (*itr)
|
|
{
|
|
switch (*itr)
|
|
{
|
|
case '%':
|
|
DecodeEscapedUTF8(str, itr);
|
|
break;
|
|
default:
|
|
*str++ = *itr++;
|
|
break;
|
|
}
|
|
}
|
|
*str = 0;
|
|
}
|
|
|