SQL format models¶
In Snowflake, SQL format models (i.e. literals containing format strings) are used to specify how numeric values are converted to text strings and vice versa. As such, they can be specified as arguments in the TO_CHAR , TO_VARCHAR and TO_DECIMAL , TO_NUMBER , TO_NUMERIC conversion functions.
Note
Snowflake also provides some limited SQL format model support for dates, times, and timestamps (see Date & time functions and Conversion functions). Full support for using SQL format models to format dates, times, and timestamps will be added in a future release.
Components of a format model¶
A format model consists of a string of format elements and literals.
Format elements¶
Format elements are sequences of digits and/or letters (mostly caseinsensitive), and, in some cases, symbols. Format elements can be directly concatenated to each other.
Some format elements are used commonly across all format models for controlling printing and matching input text. Other format elements have specific uses based on the type of values they are used to cast to/from. For more information, see the following sections in this topic:
Format literals¶
Format literals are sequences that can consist of combinations of:
Strings of arbitrary characters delimited by double quotes (a double quote is represented as two adjacent double quotes).
One or more of the following symbols:
Symbol/Character
Notes
.
(period)In fixed numeric models, treated as a format element when following
0
,9
, orX
; otherwise preserved asis.,
(comma)In numeric models, treated as a format element when following
0
,9
, orX
; otherwise preserved asis.;
(semicolon)Always preserved asis.
:
(colon)Always preserved asis.

(minus sign)Always preserved asis.
=
(equal sign)Always preserved asis.
/
(forward slash)Always preserved asis.
(
(left parenthesis)Always preserved asis.
)
(right parenthesis)Always preserved asis.
A literal is always printed asis, exactly where it was located in the format model.
Here is a brief example of using a SQL format model to print the minus
sign after a number rather than before it. The MI
indicates where to put the minus sign if the number is a negative number.
select to_varchar(123.45, '999.99MI') as EXAMPLE;
The output would look similar to 123.45
rather than the default 123.45
.
More examples are included at the end of this topic.
Format modifiers and generic space handling¶
The following table lists special format elements that control printing and matching input text, and are common to all format models:
Element 
Description 


Nothing printed; optional space on input. 

Fill mode modifier; toggles between compact and fill modes for any elements following the modifier in the model. 

Exact match modifier; toggles between lax and exact match modes for any elements following the modifier in the model. 
Note
The fill mode modifier has no effect on the textminimal numeric format elements (TM
, TM9
, and TME
).
Printing output strings using the fill mode modifier¶
By default, the fill mode is set to fill and the FM
fill mode modifier toggles it to compact; repeated use toggles it back to fill, etc.
In most cases, using fill mode on printing guarantees that format elements produce output of a fixed width by padding numbers on the left with leading zeros or spaces, and padding text with spaces on the right. This guarantees that columnar output in fixedwidth fonts will be aligned.
In compact mode, most format elements produce only minimumwidth output (i.e. leading zeros and spaces and trailing spaces are suppressed).
The format elements that don’t adhere to these rules are explicitly noted below.
The exact match modifier, FX
does not affect printing; the underscore format element prints nothing.
Parsing input strings using the modifiers¶
Parsing of input strings is affected by both the fill mode modifier, FM
, and the exact match modifier FX
. Initially:
Fill mode is set to fill and
FM
toggles it to compact and back.Exact match mode is set to lax and
FX
toggles it to exact and back.
All string matching against format elements and literals during parsing is caseinsensitive.
In lax mode, the first step of input parsing is skipping leading white space (a sequence of spaces, tabs, LF, CR, FF, and VT characters); the mode at the beginning input is strict if the first format
element is FX
, and lax otherwise.
Note
Only normal space characters are allowed within values to be parsed (i.e. components cannot be on different lines, separated by tabs, etc.).
In the lax match mode, spaces within literals are matched against any nonempty input sequence of spaces; nonspace characters are matched onetoone. In the exact mode, all characters in a literal must match the input characters onetoone.
The numeric format elements are matched against the corresponding digit sequences:
If both fill and exact modes are in effect, the number of digits must exactly correspond to the width of the corresponding numeric format elements (leading zeros are expected).
If compact or lax mode is in effect, a matching input number must have, at most, the number of digits equal to the maximal width of the format element, and at least one digit; leading zeros are ignored.
The textual format elements are matched caseinsensitively:
If both fill and exact modes are in effect, the number of trailing spaces, up to the max width of the element, is expected.
Otherwise, spaces after the variablelength textual elements are ignored in lax mode, and exact match to the actual word (without padding spaces) is expected in exact mode.
Finally, the trailing white space until the end of the input string is ignored if the current mode is lax.
Normally, both lax and exact modes do not allow matching spaces where spaces are not present in the format model or could not be generated by printing the content of format elements in fill mode.
Note
This behavior differs from Oracle lax match semantics, where spaces can be inserted in between any two format elements — Snowflake uses stricter matching semantics to avoid excessive false matches during automatic data type recognition.
Places where spaces should be ignored if present in both lax and exact modes can be explicitly marked using the _
(underscore) format element.
As a rule of thumb, a format in exact mode recognizes only input strings printed by the same format, while a format in lax mode recognizes input strings which were printed by the similar format with any fill mode modifiers added or removed.
Numeric format models¶
Numeric format models supports two types:
Fixedposition (with explicit placement of digits where the
0
,9
, orX
format elements are placed)Textminimal (
TM
,TME
, andTM9
format elements)
Note
These two types cannot be intermingled within the same model.
Fixedposition numeric formats¶
Note
This section discusses nonnegative fixedposition numbers; for more information about positioning of a number’s sign in the output for fixedposition numeric formats, see Sign Position for FixedPosition Formats.
Fixedposition numbers are represented using digit elements, 0
or 9
. For example, 999
holds numbers from 1 to 3 decimal digits. The fractional part of the numbers is delimited using separator
elements, .
(period) or D
:
.
is always rendered as a period.To use a different character for the
D
elements, modify the input string to replace all periods with commas and all commas with periods before applying the cast function.
Normally, the leading zeros in the integer part and trailing zeros in the fractional part are replaced with spaces (except when the value of the integer part is zero, in which case it is rendered as a
single 0
character). To suppress this behavior use the 0
format element in place of 9
; the corresponding positions have 0
characters preserved. The format element B
, when used before
the number, suppresses preserving the last 0
in the integer value (i.e. if you use B
and the value of the integer part of the number is zero, all digits are rendered as spaces).
The digit group separator ,
(comma) or G
results in the corresponding group separator character being printed if the number is big enough so the digits are on the both sides of group separator.
An example of a format model useful for printing currency sums would be 999,999.00
.
When there are more digits in the integer part of the number than there are digit positions in the format, all digits are printed as #
to indicate overflow.
The exponent element causes fixedposition numbers to be normalized so that the first digit in the integer part is 1 to 9 (unless the value of the number is zero, in which case the value of the exponent
is also zero). The EE
element automatically picks the right number of digits in the exponent, and does not print the +
sign, while EEE
, EEEE
, and EEEEE
always print the +
or 
sign for the exponent and the requested number of digits (leading zeros are not suppressed). Exponent overflow is indicated by #
in place of digits.
The exponent indicators print either capital E
or lowercase e
depending on the case of the first letter in the format element.
The X
format element works like 9
, except that hexadecimal digits 09AF
are printed. Currently, hexadecimal fractions are not supported. Similar to 9
, X
replaces leading zeros with
spaces. The 0
element, when used together with X
prints hexadecimal digits without leading zero suppression (thus use 000X
to print hex numbers that always contain 4 digits).
Note that X
prints hexadecimal digits with uppercase Latin letters, and lowercase x
prints lowercase Latin letters. The hexadecimal 0
format element uses the case of the subsequent X
format element.
Normally, hexadecimal numbers are printed as unsigned, i.e. negative numbers have all 1
’s in the most significant bit(s), but using the X
element together with an explicit sign (S
or MI
)
causes the 
sign to be printed along with the absolute value of the number.
Fixedposition numeric format models report overflow on special values (infinity or notanumber) of floating point numbers.
Fixedposition format elements¶
The following table lists the supported elements for fixedposition formats. Note the following:
The Repeatable column indicates whether an element can be repeated in a format model, otherwise the element can only be used once per format model.
The Casesensitive column indicates elements where the case of the element affects the format. For example:
EE
processes exponents with an uppercaseE
.ee
processes exponents with a lowercasee
.
All the other elements are caseinsensitive.
Element 
Repeatable 
Casesensitive 
Description 


Dollar sign printed before digits in the number (usually after the sign). 


Decimal fraction separator; always printed as a period. 


✔ 
Digit group separator; printed as a comma or blank space. 


✔ 
Position for a digit; leading/trailing zeros are explicitly printed. 


✔ 
Position for a digit; leading/trailing zeros are replaced with blank spaces. 


Forces representing a zero value as a space in the subsequent number. 


Decimal fraction separator; alternative for 


✔ 
Variablewidth exponent, from 2 to 5 characters, with no 


✔ 
Fixedwidth exponent (3 characters); range covers from 


✔ 
Fixedwidth exponent (4 characters); range covers from 


✔ 
Fixedwidth exponent (5 characters); range covers from 


✔ 
Digit group separator; alternative for 


Explicit numeric sign place holder; prints a space for positive numbers or a 


Explicit numeric sign place holder; prints a 


✔ 
✔ 
Hexadecimal digit. 
Sign position for fixedposition formats¶
By default, fixedposition formats always reserve a space for the number’s sign:
For nonnegative numbers, the default blank space is printed before the first digit.
For negative numbers, the default blank space and

sign are printed before the first digit (or decimal, when theB
format element is used for fractional numbers).
However, the S
, MI
, and $
format elements can be used to explicitly specify where the sign and/or blank space for the number are located.
For example (underscores, _
, are used in these examples to indicate where blank spaces are inserted):
Format Model 





















Printing numbers using fixedposition formats and the fill mode modifier¶
In fill mode, the variablelength format elements, such as EE
and MI
, are spacepadded on the right.
In compact mode, all spaces resulting from numeric format elements, including the variablelength elements, are removed, so the resulting strings are shorter and no longer aligned. For example (note the lack of blank spaces):
Format Model 






Parsing numbers using fixedposition formats and the modifiers¶
Parsing strings containing numbers is affected by both the FX
and FM
modifiers:
In lax mode:
Digit group separators are optional (i.e. numbers with or without group separators match — though numbers of digits between respective group separators must match); it also permits
+
as a valid match for theMI
format element.The lax mode does not disable requirement that digits (even leading or trailing zeros) must be present to match
0
format elements.Spaces between the leading sign and the first digit are allowed in lax mode.
Also, in lax mode, all the exponent format elements (
EE
,EEE
,EEEE
, andEEEEE
) are treated asEE
, and match an exponent specification with 1 to 3 digits and optional+
or
sign.Use
B
to allow matching numbers with no digits in the integer part. The decimal dot before an empty fractional part is optional in lax mode.
In exact mode:
The number must have a proper number of spaces in place of omitted digits to match the format (i.e. in fill mode, it is spaces and, in compact mode, it is a lack of spaces).
Omitting group separators is not allowed under exact mode, and
MI
won’t match the+
sign.The exponent format elements other than
EE
must match the sign place and the exact number of digits required by the format element.The decimal dot in the place specified by the format model is mandatory.
Textminimal numeric formats¶
While fixedposition numeric format models always explicitly specify the number of digits, the textminimal format elements use a minimal number of digits based on the value of the number. The TM*
format
elements always produce variablelength output with no spaces, regardless of the fill mode modifier (fill or compact).
TM9
prints the number as an integer or decimal fraction, based on the value of the number. Any decimal fixedpoint number value is printed precisely with the number of digits in the fractional part determined by the scale of the number (trailing zeros are preserved in fill mode).For floatingpoint numbers,
TM9
picks the number of fractional digits based on the number’s exponent (note that precise binary to decimal fraction conversion is not possible). If the floatingpoint number’s magnitude is too large, causing the positional notation to be too long, it switches to scientific notation (seeTME
below). If the floatingpoint number is too small,TM9
prints zero.TME
prints the number in scientific notation, i.e. with exponent (same asEE
) and one digit in the integer position of the fractional part. The case of the exponent indicator (E
ore
) matches the case of the first letter (T
ort
) in the format element.TM
chooses eitherTM9
orTME
depending on the magnitude of the number, to minimize the length of the text while preserving precision.
Textminimal format elements¶
The following table lists the supported elements for textminimal formats. Note the following:
No elements can be repeated within a textminimal format string.
The Casesensitive column indicates elements where the case of the element affects the format. For example:
TME
processes exponents with an uppercaseE
.tme
processes exponents with a lowercasee
.
All the other elements are caseinsensitive.
Element 
Repeatable 
Casesensitive 
Description 


Dollar sign is inserted before digits in the number (usually after sign). 


✔ 
Textminimal number, either 


✔ 
Textminimal number in positional notation. 


✔ 
Textminimal number in scientific notation (with exponent). 


Forces representing a zero value as a space in the subsequent number. 


Explicit numeric sign place holder; becomes either 


Explicit numeric sign place holder; becomes either 
Sign position for textminimal formats¶
By default, the sign for textminimal formats is either:

for negative numbers, prepended to the number.Omitted for nonnegative numbers.
The $
, S
, and MI
elements have the same effect as with fixedposition format models. Note that floatingpoint numbers have two distinct zero values (+0.
and 0.
) which represent
infinitesimal positive and negative values, respectively.
Parsing numbers using textminimal formats and the modifiers¶
Parsing with the textminimal format models is not affected by the FX
or FM
modifiers; however, the explicit sign elements, S
and MI
are affected, as described above.
TM9
matches any decimal number (integer or fractional) in positional notation; it does not match numbers in scientific notation (i.e. with exponent). Conversely:
TME
matches only scientific notation.TM
matches both.
Numbers matched by textminimal elements cannot have spaces or digit group separators within them.
Letters within exponent elements and hexadecimal digits are always matched without regard to case (lower or upper).
Alternate, automatic, and default formats¶
Element 
Description 


Separates alternative formats. 

Automatic format(s). 
When parsing strings, it is possible to specify multiple alternative formats by separating format strings with the 
character. The string is successfully parsed if it matches any one format. If the
input string matches multiple formats, any format will be used for the conversion.
An entire format used for parsing can be replaced with the keyword AUTO
; this inserts one or more alternative automatic formats depending on the type of the source or result value. Adding a custom format
to the automatic format(s) can be done using AUTO
as one of the alternatives.
Default formats are used when formats are not explicitly specified in cast functions, for parsing input values (i.e. in CSV files), and for printing results.
Default formats for printing¶
The following table lists the default formats for printing:
SQL Data Type 
Parameter 
Default Format 

DECIMAL 
none 

DOUBLE 
none 

Default formats for parsing¶
The following table lists the default formats for parsing:
SQL Data Type 
Parameter 
Default 

DECIMAL 
None 

DOUBLE 
None 

The list of formats used for automatic optimistic string conversion (i.e. for strings which are automatically recognized as numeric) is the union of all the formats in the above table of default input formats.
Examples¶
Output examples¶
This example shows how to display numbers with leading zeros:
create table sample_numbers (f float); insert into sample_numbers (f) values (1.2); insert into sample_numbers (f) values (123.456); insert into sample_numbers (f) values (1234.56); insert into sample_numbers (f) values (123456.789); select to_varchar(f, '999,999.999'), to_varchar(f, 'S000,000.000') from sample_numbers;
The output will look similar to:
+++  TO_VARCHAR(F, '999,999.999')  TO_VARCHAR(F, 'S000,000.000')  +==============================+===============================+  1.2  +000,001.200  +++  123.456  +000,123.456  +++  1,234.56  +001,234.560  +++  123,456.789  123,456.789  +++
You don’t need leading zeros in order to align numbers. The default fill mode is “fill”, which means that leading blanks are used to align numbers based on the positions of the decimal points.
select to_varchar(f, '999,999.999'), to_varchar(f, 'S999,999.999') from sample_numbers;
The output will look similar to:
+++  TO_VARCHAR(F, '999,999.999')  TO_VARCHAR(F, 'S999,999.999')  +==============================+===============================+  1.2  +1.2  +++  123.456  +123.456  +++  1,234.56  +1,234.56  +++  123,456.789  123,456.789  +++
This example shows what happens if you use the FM (Fill Mode) modifier to switch from “fill” mode to “compact” mode, i.e. to remove leading characters that would align the numbers:
select to_varchar(f, '999,999.999'), to_varchar(f, 'FM999,999.999') from sample_numbers;
The output will look similar to:
+++  TO_VARCHAR(F, '999,999.999')  TO_VARCHAR(F, 'FM999,999.999')  +==============================+================================+  1.2  1.2  +++  123.456  123.456  +++  1,234.56  1,234.56  +++  123,456.789  123,456.789  +++
This example shows how to display numbers in exponential notation:
select to_char(1234, '9d999EE'), 'will look like', '1.234E3';
The output will look similar to:
++++  TO_CHAR(1234, '9D999EE')  'WILL LOOK LIKE'  '1.234E3'  +==========================+==================+===========+  1.234E3  will look like  1.234E3  ++++
This shows how to include literals in the output. The literal portions are enclosed within double quotes (which, in turn, are inside the single quotes that delimit the string).
select to_char(12, '">"99"<"');
The output will look similar to:
++  > 12<  ++
Input examples¶
These examples demonstrate the use of format models for inputs.
The following example shows some simple input operations, with an emphasis on showing the difference between using “0” and “9” to specify format of digits.
The digit “9” as a formatter will accept blanks or “missing” leading digits. The digit “0” as a formatter will not accept blanks or missing leading zeros.
 All of the following convert the input to the number 12,345.67. SELECT TO_NUMBER('012,345.67', '999,999.99', 8, 2); SELECT TO_NUMBER('12,345.67', '999,999.99', 8, 2); SELECT TO_NUMBER(' 12,345.67', '999,999.99', 8, 2);  The first of the following works, but the others will not convert.  (They are not supposed to convert, so "failure" is correct.) SELECT TO_NUMBER('012,345.67', '000,000.00', 8, 2); SELECT TO_NUMBER('12,345.67', '000,000.00', 8, 2); SELECT TO_NUMBER(' 12,345.67', '000,000.00', 8, 2);This shows how to accept either of two numeric formats (
###
or###
). Create the table and insert data. create table format1 (v varchar, i integer); insert into format1 (v) values ('101'); insert into format1 (v) values ('102'); insert into format1 (v) values ('103');  Try to convert varchar to integer without a  format model. This fails (as expected)  with a message similar to:  "Numeric value '102' is not recognized" update format1 set i = TO_NUMBER(v);  Now try again with a format specifier that allows the minus sign  to be at either the beginning or the end of the number.  Note the use of the vertical bar ("") to indicate that  either format is acceptable. update format1 set i = TO_NUMBER(v, 'MI999999MI'); select i from format1;