Skip to content
Projects
Groups
Snippets
Help
Loading...
Help
Support
Keyboard shortcuts
?
Submit feedback
Contribute to GitLab
Sign in / Register
Toggle navigation
S
sdkjs
Project overview
Project overview
Details
Activity
Releases
Repository
Repository
Files
Commits
Branches
Tags
Contributors
Graph
Compare
Issues
0
Issues
0
List
Boards
Labels
Milestones
Merge Requests
0
Merge Requests
0
Analytics
Analytics
Repository
Value Stream
Wiki
Wiki
Snippets
Snippets
Members
Members
Collapse sidebar
Close sidebar
Activity
Graph
Create a new issue
Commits
Issue Boards
Open sidebar
Boris Kocherov
sdkjs
Commits
67b55ee9
Commit
67b55ee9
authored
Jul 28, 2016
by
Alexander.Trofimov
Browse files
Options
Browse Files
Download
Email Patches
Plain Diff
add commented new code of formulas
parent
a8f12ac5
Changes
2
Hide whitespace changes
Inline
Side-by-side
Showing
2 changed files
with
622 additions
and
0 deletions
+622
-0
cell/model/FormulaObjects/parserFormula.js
cell/model/FormulaObjects/parserFormula.js
+621
-0
cell/model/Workbook.js
cell/model/Workbook.js
+1
-0
No files found.
cell/model/FormulaObjects/parserFormula.js
View file @
67b55ee9
...
@@ -51,6 +51,356 @@ function (window, undefined) {
...
@@ -51,6 +51,356 @@ function (window, undefined) {
var
c_oAscError
=
Asc
.
c_oAscError
;
var
c_oAscError
=
Asc
.
c_oAscError
;
var
TOK_TYPE_OPERAND
=
1
;
var
TOK_TYPE_FUNCTION
=
2
;
var
TOK_TYPE_SUBEXPR
=
3
;
var
TOK_TYPE_ARGUMENT
=
4
;
var
TOK_TYPE_OP_IN
=
5
;
var
TOK_TYPE_OP_POST
=
6
;
var
TOK_TYPE_WSPACE
=
7
;
var
TOK_TYPE_UNKNOWN
=
8
;
var
TOK_SUBTYPE_START
=
9
;
var
TOK_SUBTYPE_STOP
=
10
;
var
TOK_SUBTYPE_TEXT
=
11
;
var
TOK_SUBTYPE_LOGICAL
=
12
;
var
TOK_SUBTYPE_ERROR
=
14
;
var
TOK_SUBTYPE_UNION
=
15
;
function
ParsedThing
(
value
,
type
,
subtype
)
{
this
.
value
=
value
;
this
.
type
=
type
;
this
.
subtype
=
subtype
;
}
ParsedThing
.
prototype
.
getStop
=
function
()
{
return
new
ParsedThing
(
this
.
value
,
this
.
type
,
TOK_SUBTYPE_STOP
);
};
var
g_oCodeSpace
=
32
;
// Code of space
var
g_oCodeNumberSign
=
35
;
// Code of #
var
g_oCodeDQuote
=
34
;
// Code of "
var
g_oCodePercent
=
37
;
// Code of %
var
g_oCodeAmpersand
=
38
;
// Code of &
var
g_oCodeQuote
=
39
;
// Code of '
var
g_oCodeLeftParenthesis
=
40
;
// Code of (
var
g_oCodeRightParenthesis
=
41
;
// Code of )
var
g_oCodeMultiply
=
42
;
// Code of *
var
g_oCodePlus
=
43
;
// Code of +
var
g_oCodeComma
=
44
;
// Code of ,
var
g_oCodeMinus
=
45
;
// Code of -
var
g_oCodeDivision
=
47
;
// Code of /
var
g_oCodeSemicolon
=
59
;
// Code of ;
var
g_oCodeLessSign
=
60
;
// Code of <
var
g_oCodeEqualSign
=
61
;
// Code of =
var
g_oCodeGreaterSign
=
62
;
// Code of >
var
g_oCodeLeftSquareBracked
=
91
;
// Code of [
var
g_oCodeRightSquareBracked
=
93
;
// Code of ]
var
g_oCodeAccent
=
94
;
// Code of ^
var
g_oCodeLeftCurlyBracked
=
123
;
// Code of {
var
g_oCodeRightCurlyBracked
=
125
;
// Code of }
function
getTokens
(
formula
)
{
var
tokens
=
[];
var
tokenStack
=
[];
var
offset
=
0
;
var
length
=
formula
.
length
;
var
currentChar
,
currentCharCode
,
nextCharCode
,
tmp
;
var
token
=
""
;
var
inString
=
false
;
var
inPath
=
false
;
var
inRange
=
false
;
var
inError
=
false
;
var
regexSN
=
/^
[
1-9
]{1}(\.[
0-9
]
+
)?
E
{1}
$/
;
nextCharCode
=
formula
.
charCodeAt
(
offset
);
while
(
offset
<
length
)
{
// state-dependent character evaluation (order is important)
// double-quoted strings
// embeds are doubled
// end marks token
currentChar
=
formula
[
offset
];
currentCharCode
=
nextCharCode
;
nextCharCode
=
formula
.
charCodeAt
(
offset
+
1
);
if
(
inString
)
{
if
(
currentCharCode
===
g_oCodeDQuote
)
{
if
(
nextCharCode
===
g_oCodeDQuote
)
{
token
+=
currentChar
;
offset
+=
1
;
}
else
{
inString
=
false
;
tokens
.
push
(
new
ParsedThing
(
token
,
TOK_TYPE_OPERAND
,
TOK_SUBTYPE_TEXT
));
token
=
""
;
}
}
else
{
token
+=
currentChar
;
}
offset
+=
1
;
continue
;
}
else
if
(
inPath
)
{
// single-quoted strings (links)
// embeds are double
// end does not mark a token
if
(
currentCharCode
===
g_oCodeQuote
)
{
if
(
nextCharCode
===
g_oCodeQuote
)
{
token
+=
currentChar
;
offset
+=
1
;
}
else
{
inPath
=
false
;
}
}
else
{
token
+=
currentChar
;
}
offset
+=
1
;
continue
;
}
else
if
(
inRange
)
{
// bracked strings (range offset or linked workbook name)
// no embeds (changed to "()" by Excel)
// end does not mark a token
if
(
currentCharCode
===
g_oCodeRightSquareBracked
)
{
inRange
=
false
;
}
token
+=
currentChar
;
offset
+=
1
;
continue
;
}
else
if
(
inError
)
{
// error values
// end marks a token, determined from absolute list of values
token
+=
currentChar
;
offset
+=
1
;
if
((
"
,#NULL!,#DIV/0!,#VALUE!,#REF!,#NAME?,#NUM!,#N/A,
"
).
indexOf
(
"
,
"
+
token
+
"
,
"
)
!=
-
1
)
{
inError
=
false
;
tokens
.
push
(
new
ParsedThing
(
token
,
TOK_TYPE_OPERAND
,
TOK_SUBTYPE_ERROR
));
token
=
""
;
}
continue
;
}
// trim white-space
if
(
currentCharCode
===
g_oCodeSpace
)
{
if
(
token
.
length
>
0
)
{
tokens
.
push
(
new
ParsedThing
(
token
,
TOK_TYPE_OPERAND
));
token
=
""
;
}
tokens
.
push
(
new
ParsedThing
(
""
,
TOK_TYPE_WSPACE
));
offset
+=
1
;
while
((
currentCharCode
=
formula
.
charCodeAt
(
offset
))
===
g_oCodeSpace
)
{
offset
+=
1
;
}
if
(
offset
>=
length
)
{
break
;
}
currentChar
=
formula
[
offset
];
nextCharCode
=
formula
.
charCodeAt
(
offset
+
1
);
}
// multi-character comparators (>= || <= || <>)
if
((
currentCharCode
===
g_oCodeLessSign
&&
(
nextCharCode
===
g_oCodeEqualSign
||
nextCharCode
===
g_oCodeGreaterSign
))
||
(
currentCharCode
===
g_oCodeGreaterSign
&&
nextCharCode
===
g_oCodeEqualSign
))
{
if
(
token
.
length
>
0
)
{
tokens
.
push
(
new
ParsedThing
(
token
,
TOK_TYPE_OPERAND
));
token
=
""
;
}
tokens
.
push
(
new
ParsedThing
(
formula
.
substr
(
offset
,
2
),
TOK_TYPE_OP_IN
,
TOK_SUBTYPE_LOGICAL
));
offset
+=
2
;
nextCharCode
=
formula
.
charCodeAt
(
offset
);
continue
;
}
// scientific notation check
if
(
currentCharCode
===
g_oCodePlus
||
currentCharCode
===
g_oCodeMinus
)
{
if
(
token
.
length
>
1
)
{
if
(
token
.
match
(
regexSN
))
{
token
+=
currentChar
;
offset
+=
1
;
continue
;
}
}
}
// independent character evaulation (order not important)
// establish state-dependent character evaluations
switch
(
currentCharCode
)
{
case
g_oCodeDQuote
:
{
if
(
token
.
length
>
0
)
{
// not expected
tokens
.
push
(
new
ParsedThing
(
token
,
TOK_TYPE_UNKNOWN
));
token
=
""
;
}
inString
=
true
;
break
;
}
case
g_oCodeQuote
:
{
if
(
token
.
length
>
0
)
{
// not expected
tokens
.
push
(
new
ParsedThing
(
token
,
TOK_TYPE_UNKNOWN
));
token
=
""
;
}
inPath
=
true
;
break
;
}
case
g_oCodeLeftSquareBracked
:
{
inRange
=
true
;
token
+=
currentChar
;
break
;
}
case
g_oCodeNumberSign
:
{
if
(
token
.
length
>
0
)
{
// not expected
tokens
.
push
(
new
ParsedThing
(
token
,
TOK_TYPE_UNKNOWN
));
token
=
""
;
}
inError
=
true
;
token
+=
currentChar
;
break
;
}
case
g_oCodeLeftCurlyBracked
:
{
// mark start and end of arrays and array rows
if
(
token
.
length
>
0
)
{
// not expected
tokens
.
push
(
new
ParsedThing
(
token
,
TOK_TYPE_UNKNOWN
));
token
=
""
;
}
tmp
=
new
ParsedThing
(
'
ARRAY
'
,
TOK_TYPE_FUNCTION
,
TOK_SUBTYPE_START
);
tokens
.
push
(
tmp
);
tokenStack
.
push
(
tmp
.
getStop
());
tmp
=
new
ParsedThing
(
'
ARRAYROW
'
,
TOK_TYPE_FUNCTION
,
TOK_SUBTYPE_START
);
tokens
.
push
(
tmp
);
tokenStack
.
push
(
tmp
.
getStop
());
break
;
}
case
g_oCodeSemicolon
:
{
if
(
token
.
length
>
0
)
{
tokens
.
push
(
new
ParsedThing
(
token
,
TOK_TYPE_OPERAND
));
token
=
""
;
}
tmp
=
tokenStack
.
pop
();
if
(
tmp
&&
'
ARRAYROW
'
!==
tmp
.
value
)
{
return
null
;
}
tokens
.
push
(
tmp
);
tokens
.
push
(
new
ParsedThing
(
'
;
'
,
TOK_TYPE_ARGUMENT
));
tmp
=
new
ParsedThing
(
'
ARRAYROW
'
,
TOK_TYPE_FUNCTION
,
TOK_SUBTYPE_START
);
tokens
.
push
(
tmp
);
tokenStack
.
push
(
tmp
.
getStop
());
break
;
}
case
g_oCodeRightCurlyBracked
:
{
if
(
token
.
length
>
0
)
{
tokens
.
push
(
new
ParsedThing
(
token
,
TOK_TYPE_OPERAND
));
token
=
""
;
}
tokens
.
push
(
tokenStack
.
pop
());
tokens
.
push
(
tokenStack
.
pop
());
break
;
}
case
g_oCodePlus
:
case
g_oCodeMinus
:
case
g_oCodeMultiply
:
case
g_oCodeDivision
:
case
g_oCodeAccent
:
case
g_oCodeAmpersand
:
case
g_oCodeEqualSign
:
case
g_oCodeGreaterSign
:
case
g_oCodeLessSign
:
{
// standard infix operators
if
(
token
.
length
>
0
)
{
tokens
.
push
(
new
ParsedThing
(
token
,
TOK_TYPE_OPERAND
));
token
=
""
;
}
tokens
.
push
(
new
ParsedThing
(
currentChar
,
TOK_TYPE_OP_IN
));
break
;
}
case
g_oCodePercent
:
{
// standard postfix operators
if
(
token
.
length
>
0
)
{
tokens
.
push
(
new
ParsedThing
(
token
,
TOK_TYPE_OPERAND
));
token
=
""
;
}
tokens
.
push
(
new
ParsedThing
(
currentChar
,
TOK_TYPE_OP_POST
));
break
;
}
case
g_oCodeLeftParenthesis
:
{
// start subexpression or function
if
(
token
.
length
>
0
)
{
tmp
=
new
ParsedThing
(
token
,
TOK_TYPE_FUNCTION
,
TOK_SUBTYPE_START
);
tokens
.
push
(
tmp
);
tokenStack
.
push
(
tmp
.
getStop
());
token
=
""
;
}
else
{
tmp
=
new
ParsedThing
(
""
,
TOK_TYPE_SUBEXPR
,
TOK_SUBTYPE_START
);
tokens
.
push
(
tmp
);
tokenStack
.
push
(
tmp
.
getStop
());
}
break
;
}
case
g_oCodeComma
:
{
// function, subexpression, array parameters
if
(
token
.
length
>
0
)
{
tokens
.
push
(
new
ParsedThing
(
token
,
TOK_TYPE_OPERAND
));
token
=
""
;
}
tmp
=
(
0
!==
tokenStack
.
length
)
?
(
TOK_TYPE_FUNCTION
===
tokenStack
[
tokenStack
.
length
-
1
].
type
)
:
false
;
tokens
.
push
(
tmp
?
new
ParsedThing
(
currentChar
,
TOK_TYPE_ARGUMENT
)
:
new
ParsedThing
(
currentChar
,
TOK_TYPE_OP_IN
,
TOK_SUBTYPE_UNION
));
break
;
}
case
g_oCodeRightParenthesis
:
{
// stop subexpression
if
(
token
.
length
>
0
)
{
tokens
.
push
(
new
ParsedThing
(
token
,
TOK_TYPE_OPERAND
));
token
=
""
;
}
tokens
.
push
(
tokenStack
.
pop
());
break
;
}
default
:
{
// token accumulation
token
+=
currentChar
;
break
;
}
}
++
offset
;
}
// dump remaining accumulation
if
(
token
.
length
>
0
)
{
tokens
.
push
(
new
ParsedThing
(
token
,
TOK_TYPE_OPERAND
));
}
return
tokens
;
}
/** @enum */
/** @enum */
var
cElementType
=
{
var
cElementType
=
{
number
:
0
,
number
:
0
,
...
@@ -3586,6 +3936,277 @@ parserFormula.prototype.parse = function(local, digitDelim) {
...
@@ -3586,6 +3936,277 @@ parserFormula.prototype.parse = function(local, digitDelim) {
Что упрощает вычисление результата формулы.
Что упрощает вычисление результата формулы.
При разборе формулы важен порядок проверки очередной части выражения на принадлежность тому или иному типу.
При разборе формулы важен порядок проверки очередной части выражения на принадлежность тому или иному типу.
*/
*/
if
(
false
)
{
//console.log(this.Formula);
var
cFormulaList
=
(
local
&&
AscCommonExcel
.
cFormulaFunctionLocalized
)
?
AscCommonExcel
.
cFormulaFunctionLocalized
:
cFormulaFunction
;
var
aTokens
=
getTokens
(
this
.
Formula
);
if
(
null
===
aTokens
)
{
this
.
outStack
=
[];
this
.
error
.
push
(
c_oAscError
.
ID
.
FrmlWrongOperator
);
return
false
;
}
var
stack
=
[],
val
,
valUp
,
tmp
,
elem
,
len
,
indentCount
=
-
1
,
args
=
[],
prev
,
next
,
arr
=
null
,
bArrElemSign
=
false
;
for
(
var
i
=
0
,
nLength
=
aTokens
.
length
;
i
<
nLength
;
++
i
)
{
found_operand
=
null
;
val
=
aTokens
[
i
].
value
;
switch
(
aTokens
[
i
].
type
)
{
case
TOK_TYPE_OPERAND
:
{
if
(
TOK_SUBTYPE_TEXT
===
aTokens
[
i
].
subtype
)
{
elem
=
new
cString
(
val
);
}
else
{
tmp
=
parseFloat
(
val
);
if
(
isNaN
(
tmp
))
{
valUp
=
val
.
toUpperCase
();
if
(
'
TRUE
'
===
valUp
||
'
FALSE
'
===
valUp
)
{
elem
=
new
cBool
(
valUp
);
}
else
{
if
(
-
1
!==
val
.
indexOf
(
'
!
'
))
{
tmp
=
AscCommonExcel
.
g_oRangeCache
.
getRange3D
(
val
);
if
(
tmp
)
{
this
.
is3D
=
true
;
elem
=
(
tmp
.
isOneCell
()
&&
(
null
===
tmp
.
sheet2
||
tmp
.
sheet
===
tmp
.
sheet2
))
?
new
cRef3D
(
tmp
.
getName
(),
tmp
.
sheet
,
this
.
wb
)
:
new
cArea3D
(
tmp
.
getName
(),
tmp
.
sheet
,
null
!==
tmp
.
sheet2
?
tmp
.
sheet2
:
tmp
.
sheet
,
this
.
wb
);
}
else
{
this
.
error
.
push
(
c_oAscError
.
ID
.
FrmlWrongOperator
);
this
.
outStack
=
[];
return
false
;
}
}
else
{
tmp
=
AscCommonExcel
.
g_oRangeCache
.
getAscRange
(
valUp
);
if
(
tmp
)
{
elem
=
tmp
.
isOneCell
()
?
new
cRef
(
valUp
,
this
.
ws
)
:
new
cArea
(
valUp
,
this
.
ws
);
}
else
{
elem
=
new
cName
(
aTokens
[
i
].
value
,
this
.
wb
,
this
.
ws
);
}
}
}
}
else
{
elem
=
new
cNumber
(
tmp
);
}
}
if
(
arr
)
{
if
(
cElementType
.
number
!==
elem
.
type
&&
cElementType
.
bool
!==
elem
.
type
&&
cElementType
.
string
!==
elem
.
type
)
{
this
.
outStack
=
[];
this
.
error
.
push
(
c_oAscError
.
ID
.
FrmlAnotherParsingError
);
return
false
;
}
else
{
if
(
bArrElemSign
)
{
if
(
cElementType
.
number
!==
elem
.
type
)
{
this
.
outStack
=
[];
this
.
error
.
push
(
c_oAscError
.
ID
.
FrmlAnotherParsingError
);
return
false
;
}
elem
.
value
*=
-
1
;
bArrElemSign
=
false
;
}
arr
.
addElement
(
elem
);
}
}
else
{
this
.
outStack
.
push
(
elem
);
this
.
f
.
push
(
elem
);
}
break
;
}
case
TOK_TYPE_OP_POST
:
case
TOK_TYPE_OP_IN
:
{
if
(
TOK_SUBTYPE_UNION
===
aTokens
[
i
].
subtype
)
{
this
.
outStack
=
[];
this
.
error
.
push
(
c_oAscError
.
ID
.
FrmlWrongOperator
);
return
false
;
}
prev
=
aTokens
[
i
-
1
];
if
(
'
-
'
===
val
&&
(
0
===
i
||
(
TOK_TYPE_OPERAND
!==
prev
.
type
&&
TOK_TYPE_OP_POST
!==
prev
.
type
&&
(
TOK_SUBTYPE_STOP
!==
prev
.
subtype
||
(
TOK_TYPE_FUNCTION
!==
prev
.
type
&&
TOK_TYPE_SUBEXPR
!==
prev
.
type
)))))
{
elem
=
new
cFormulaOperators
[
'
un_minus
'
]();
}
else
{
elem
=
new
cFormulaOperators
[
val
]();
}
if
(
arr
)
{
if
(
bArrElemSign
||
'
un_minus
'
!==
elem
.
name
)
{
this
.
outStack
=
[];
this
.
error
.
push
(
c_oAscError
.
ID
.
FrmlWrongOperator
);
return
false
;
}
else
{
bArrElemSign
=
true
;
break
;
}
}
this
.
f
.
push
(
elem
);
len
=
stack
.
length
;
while
(
0
!==
len
)
{
tmp
=
stack
[
len
-
1
];
if
(
elem
.
isRightAssociative
?
(
elem
.
priority
<
tmp
.
priority
)
:
((
elem
.
priority
<=
tmp
.
priority
)))
{
this
.
outStack
.
push
(
tmp
);
--
len
;
}
else
{
break
;
}
}
stack
.
length
=
len
;
stack
.
push
(
elem
);
break
;
}
case
TOK_TYPE_FUNCTION
:
{
if
(
TOK_SUBTYPE_START
===
aTokens
[
i
].
subtype
)
{
val
=
val
.
toUpperCase
();
if
(
'
ARRAY
'
===
val
)
{
if
(
arr
)
{
this
.
outStack
=
[];
this
.
error
.
push
(
c_oAscError
.
ID
.
FrmlWrongOperator
);
return
false
;
}
arr
=
new
cArray
();
break
;
}
else
if
(
'
ARRAYROW
'
===
val
)
{
if
(
!
arr
)
{
this
.
outStack
=
[];
this
.
error
.
push
(
c_oAscError
.
ID
.
FrmlWrongOperator
);
return
false
;
}
arr
.
addRow
();
break
;
}
else
if
(
val
in
cFormulaList
)
{
elem
=
new
cFormulaList
[
val
]();
}
else
if
(
val
in
cAllFormulaFunction
)
{
elem
=
new
cAllFormulaFunction
[
val
]();
}
else
{
elem
=
new
cBaseFunction
(
val
);
elem
.
isXLFN
=
(
0
===
val
.
indexOf
(
"
_xlfn.
"
));
}
stack
.
push
(
elem
);
args
[
++
indentCount
]
=
1
;
}
else
{
if
(
arr
)
{
if
(
'
ARRAY
'
===
val
)
{
if
(
!
arr
.
isValidArray
())
{
this
.
outStack
=
[];
// размер массива не согласован
this
.
error
.
push
(
c_oAscError
.
ID
.
FrmlAnotherParsingError
);
return
false
;
}
this
.
outStack
.
push
(
arr
);
arr
=
null
;
}
else
if
(
'
ARRAYROW
'
!==
val
)
{
this
.
outStack
=
[];
this
.
error
.
push
(
c_oAscError
.
ID
.
FrmlAnotherParsingError
);
return
false
;
}
break
;
}
len
=
stack
.
length
;
while
(
0
!==
len
)
{
tmp
=
stack
[
len
-
1
];
--
len
;
this
.
outStack
.
push
(
tmp
);
if
(
cElementType
.
func
===
tmp
.
type
)
{
prev
=
aTokens
[
i
-
1
];
tmp
.
setArgumentsCount
(
args
[
indentCount
]
-
((
prev
&&
TOK_TYPE_FUNCTION
===
prev
.
type
&&
TOK_SUBTYPE_START
===
prev
.
subtype
)
?
1
:
0
));
if
(
!
tmp
.
checkArguments
())
{
this
.
outStack
=
[];
this
.
error
.
push
(
c_oAscError
.
ID
.
FrmlWrongMaxArgument
);
return
false
;
}
break
;
}
}
stack
.
length
=
len
;
--
indentCount
;
}
break
;
}
case
TOK_TYPE_ARGUMENT
:
{
if
(
arr
)
{
break
;
}
if
(
-
1
===
indentCount
)
{
throw
'
error!!!!!!!!!!!
'
;
}
args
[
indentCount
]
+=
1
;
len
=
stack
.
length
;
while
(
0
!==
len
)
{
tmp
=
stack
[
len
-
1
];
if
(
cElementType
.
func
===
tmp
.
type
)
{
break
;
}
this
.
outStack
.
push
(
tmp
);
--
len
;
}
stack
.
length
=
len
;
next
=
aTokens
[
i
+
1
];
if
(
next
&&
(
TOK_TYPE_ARGUMENT
===
next
.
type
||
(
TOK_TYPE_FUNCTION
===
next
.
type
&&
TOK_SUBTYPE_START
!==
next
.
subtype
)))
{
this
.
outStack
.
push
(
new
cEmpty
());
break
;
}
break
;
}
case
TOK_TYPE_SUBEXPR
:
{
if
(
TOK_SUBTYPE_START
===
aTokens
[
i
].
subtype
)
{
elem
=
new
parentLeft
();
stack
.
push
(
elem
);
}
else
{
elem
=
new
parentRight
();
len
=
stack
.
length
;
while
(
0
!==
len
)
{
tmp
=
stack
[
len
-
1
];
--
len
;
this
.
outStack
.
push
(
tmp
);
if
(
tmp
instanceof
parentLeft
)
{
break
;
}
}
stack
.
length
=
len
;
}
this
.
f
.
push
(
elem
);
break
;
}
case
TOK_TYPE_WSPACE
:
{
if
(
0
!==
i
&&
i
!==
nLength
-
1
)
{
prev
=
aTokens
[
i
-
1
];
next
=
aTokens
[
i
+
1
];
if
((
TOK_TYPE_OPERAND
===
prev
.
type
||
((
TOK_TYPE_FUNCTION
===
prev
.
type
||
TOK_TYPE_SUBEXPR
===
prev
.
type
)
&&
TOK_SUBTYPE_STOP
===
prev
.
subtype
))
&&
((
TOK_TYPE_OPERAND
===
next
.
type
)
||
((
TOK_TYPE_FUNCTION
===
next
.
type
||
TOK_TYPE_SUBEXPR
===
next
.
type
)
&&
TOK_SUBTYPE_START
===
next
.
subtype
)))
{
aTokens
[
i
].
type
=
TOK_TYPE_OP_IN
;
aTokens
[
i
].
value
=
'
'
;
--
i
;
}
}
break
;
}
}
}
while
(
stack
.
length
!==
0
)
{
this
.
outStack
.
push
(
stack
.
pop
());
}
if
(
this
.
outStack
.
length
!=
0
)
{
return
this
.
isParsed
=
true
;
}
else
{
return
this
.
isParsed
=
false
;
}
}
this
.
operand_expected
=
true
;
this
.
operand_expected
=
true
;
var
wasLeftParentheses
=
false
,
wasRigthParentheses
=
false
,
found_operand
=
null
,
_3DRefTmp
=
null
,
_tableTMP
=
null
;
var
wasLeftParentheses
=
false
,
wasRigthParentheses
=
false
,
found_operand
=
null
,
_3DRefTmp
=
null
,
_tableTMP
=
null
;
var
cFormulaList
=
(
local
&&
AscCommonExcel
.
cFormulaFunctionLocalized
)
?
AscCommonExcel
.
cFormulaFunctionLocalized
:
cFormulaFunction
;
var
cFormulaList
=
(
local
&&
AscCommonExcel
.
cFormulaFunctionLocalized
)
?
AscCommonExcel
.
cFormulaFunctionLocalized
:
cFormulaFunction
;
...
...
cell/model/Workbook.js
View file @
67b55ee9
...
@@ -5302,6 +5302,7 @@ Woorksheet.prototype._BuildDependencies=function(cellRange){
...
@@ -5302,6 +5302,7 @@ Woorksheet.prototype._BuildDependencies=function(cellRange){
cellId
=
g_oCellAddressUtils
.
getCellId
(
c
.
nRow
,
c
.
nCol
),
cellId
=
g_oCellAddressUtils
.
getCellId
(
c
.
nRow
,
c
.
nCol
),
aRefs
=
[],
aRefs
=
[],
cache
=
inCache
(
aCache
,
c
.
sFormula
,
aRefs
),
cache
=
inCache
(
aCache
,
c
.
sFormula
,
aRefs
),
//cache = false,
bInCache
=
false
;
bInCache
=
false
;
if
(
cache
)
if
(
cache
)
{
{
...
...
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment