Textscan won't read the dates with spaces

Textscan doesn't work when date elements are separated by spaces. However the same format works when used with datetime function. Have a look at the following code:

textscan('1959 05 21','%{yyyy MM dd}D')                       % Doesn't work when there are spaces
datetime('now','Format','yyyy MM dd')                         % same format works with datetime function
textscan('1959-05-21','%{yyyy-MM-dd}D')                       % Works when when space is replaced with non letter character 
textscan('1959 05 21','%{yyyy MM dd}D','whitespace','')       % Works when whitespace is set to none
textscan('1959 05 21 567','%{yyyy MM dd}D%d','whitespace','') % Doesn't work

How can I make last line of the code to work?

Thanks

PS: Read Walter Roberson and per isakson's comments on the accepted answer

[per isakson]

"How can I make last line of the code to work?" I don't think it's possible. It seems that Matlab cannot handle the double use of space, as part of the date format and at the same time as list separator before the integer.

I assume that your problem is not to parse the string, but to find the limits of textscan.

Answer 1

[Jeremy Hughes]

Hi Per,

The issue is that textscan's delimiter is space by default. Parsing happens first, then datatype conversion. In this case, you're getting

"1959 05 21" -> "1959","05","21"

and trying to convert each of these into it own datetime. This is a pretty common confusion.

The trick to parsing this correctly is to supply 'Delimiter' to textscan. Try:

textscan('1959 05 21','%{yyyy MM dd}D','Delimiter',',')

Hope this helps, Jeremy

[Walter Roberson]

Note that the attempts

textscan('1959 05 21 567','%{yyyy MM dd}D%d','whitespace','')

or

    textscan('1959 05 21 567','%{yyyy MM dd}D%d','Delimiter',',')

will not work, and

    textscan('1959 05 21 567','%{yyyy MM dd}D%*[ ]%d','whitespace','')

will not work either. In each of those case, the entire group 1959 05 21 567 gets grabbed and passed to datetime for parsing. textscan parsing is greedy that way, just as is the case for numeric fields:

>> textscan('123456', '%d4%d')
ans =
  1×2 cell array
    {[123456]}    {0×1 int32}
>> textscan('123456', '%d%*[4]%d')
ans =
  1×2 cell array
    {[123456]}    {0×1 int32}
>> textscan('123456', '%d%d','delimiter','4')
ans =
  1×2 cell array
    {[123456]}    {0×1 int32}

>> textscan('123e56', '%de%d')
ans =
  1×2 cell array
    {[2147483647]}    {0×1 int32}
>> textscan('123e56', '%d%*[e]%d')
ans =
  1×2 cell array
    {[2147483647]}    {0×1 int32}
>> textscan('123e56', '%d%d','delimiter','e')
ans =
  1×2 cell array
    {[2147483647]}    {0×1 int32}
>> textscan('123e56', '%d%d','whitespace','e')
ans =
  1×2 cell array
    {[2147483647]}    {0×1 int32}

[per isakson]

Had the task been to read and parse the string, '1959 05 21 567', I would have tried

cac = textscan('1959 05 21 567','%10c%d');
dt  = datetime( cac{1},'Format','yyyy MM dd');

Note: this doesn't work with (a varying number of) leading spaces.

>> textscan('  1959 05 21 567','%12c%d')
ans = 
    '1959 05 21 5'    [67]

The Conversion Specifier, %c, lets me decide how many characters to read (except for leading (white)spaces) and these two lines are surprisingly efficient. ( %c is cheap.)

However, with 'whitespace','' it works

>> textscan('  1959 05 21 567','%12c%d', 'Whitespace','')
ans = 
    '  1959 05 21'    [567]

>> textscan('  1959 05 21 567 890','%12c%d%d', 'whitespace','')
ans = 
    '  1959 05 21'    [567]    [890]