datevec() and datenum() fail intermittently with custom format string (MATLAB R2010a)

Successive calls to datevec() and datenum() with the same argument alternately success and fail. Any ideas on what might be causing this or something I can do to help debug it?

Below is a command-line example, but it happens in functions and scripts too. Note: sometimes (e.g. after a fresh start of MATLAB) it works perfectly every time. (It will probably work perfectly for you, too.)

K>> start_timeTxt
start_timeTxt = 
2013-07-16 10:34:59

K>> class(start_timeTxt)
ans =
char

% Nothing abnormal in time string

K>> double(start_timeTxt)
ans =
  50    48    49    51    45    48    55    45    49    54    32    49    48    58    51    52    58    53    57

K>> datevec(start_timeTxt, 'yyyy-mm-dd HH:MM:SS')
??? Error using ==> dtstr2dtvecmx
Failed on converting date string to date number.

Error in ==> datevec at 124
          y = dtstr2dtvecmx(t,icu_dtformat);

% Try it again... 

K>> datevec(start_timeTxt, 'yyyy-mm-dd HH:MM:SS')
ans =
      2013           7          16          10          34          59
% And again...

K>> datevec(start_timeTxt, 'yyyy-mm-dd HH:MM:SS')
??? Error using ==> dtstr2dtvecmx
Failed on converting date string to date number.

Error in ==> datevec at 124
          y = dtstr2dtvecmx(t,icu_dtformat);
% And a third time...

K>> datevec(start_timeTxt, 'yyyy-mm-dd HH:MM:SS')
ans =
     2013           7          16          10          34          59

% Is datenum() broken too?  Expectedly, yes...:

K>> datenum(start_timeTxt, 'yyyy-mm-dd HH:MM:SS')
??? Error using ==> datenum at 182
DATENUM failed.

Caused by:
  Error using ==> dtstr2dtnummx
  Failed on converting date string to date number.

% but like datevec(), it alternately works too...

K>> datenum(start_timeTxt, 'yyyy-mm-dd HH:MM:SS')
ans =
7.3543e+005

K>> datenum(start_timeTxt, 'yyyy-mm-dd HH:MM:SS')
??? Error using ==> datenum at 182
DATENUM failed.

Caused by:
  Error using ==> dtstr2dtnummx
  Failed on converting date string to date number.

K>> datenum(start_timeTxt, 'yyyy-mm-dd HH:MM:SS')
ans =
7.3543e+005

% Try it with literals

>> datenum('2013-07-16 10:34:59', 'yyyy-mm-dd HH:MM:SS')
??? Error using ==> datenum at 182
DATENUM failed.

Caused by:
  Error using ==> dtstr2dtnummx
  Failed on converting date string to date number.

>> datenum('2013-07-16 10:34:59', 'yyyy-mm-dd HH:MM:SS')
ans =
7.3543e+005

% In case you are wondering:

K>> which datenum, which datevec
C:\Program Files\MATLAB\R2010a\toolbox\matlab\timefun\datenum.m
C:\Program Files\MATLAB\R2010a\toolbox\matlab\timefun\datevec.m

K>> ver

-------------------------------------------------------------------------------------

MATLAB Version 7.10.0.499 (R2010a) MATLAB License Number: ••••

Operating System: Microsoft Windows 7 Version 6.1 (Build 7601: Service Pack 1)

Java VM Version: Java 1.6.0_12-b04 with Sun Microsystems Inc. Java HotSpot™ 64-Bit Server VM mixed mode

-------------------------------------------------------------------------------------

K>> memory
Maximum possible array:               4553 MB (4.775e+009 bytes) *
Memory available for all arrays:      4553 MB (4.775e+009 bytes) *
Memory used by MATLAB:                 627 MB (6.578e+008 bytes)
Physical Memory (RAM):                4030 MB (4.225e+009 bytes)

*  Limited by System Memory (physical + swap file) available.

[dpb]

It's an earlier version--did you check for bug reports?

All I can suggest is to contact official TMW support; the response will be dependent upon your (or your sponsoring institution's) license status. I don't know when TMW considers EOL for earlier versions.

Since it's in the mex-file your options are somewhat limited...I don't suppose there's anything in common about the sequence of events prior to the time the failures start that you can point to, is there...that'd be too easy. :)

Anyway, since it is in date[num|vec], if after a restart it seems to function correctly for a repeated number of times on the test case, I'd tend to suspect there was a previous execution of a call that caused an internal problem that results in the symptom. That to say that perhaps somewhere earlier in production or development code there's a particular useage of datenum/datevec that is the preciptator--if you could identify such a case perhaps it could lead to some ideas as to cause and potential workaround. Long shot, I agree; unfortunately crystal orb is dark...

[per isakson]

Have you checked the format string before the call to dtstr2dtvecmx and dtstr2dtnummx, respectively? A mismatch in the format string causes this error:

    >> datevec(start_timeTxt, 'yyyy-mmdd HH:MM:SS')
    Error using dtstr2dtvecmx
    Failed on converting date string to date number.
    Error in datevec (line 118)
                y = dtstr2dtvecmx(t,icu_dtformat);

Answer 1

[Jan]

This is a known bug, see http://www.mathworks.com/support/bugreports/details.html?rp=622845. You find downloadable workarounds there also.

It is always worth to take a look into the list of known bugs in case of problems.

[Jim Tonti]

Jan- Thanks for the quick response. Since upgrading wasn't an option, and the bug was incredibly consistent, I had implemented the 'try-catch' workaround with success. It's good to see this rather strange bug has been fixed in future releases - it had me going for a while. Software is one of the places where insanity rules: doing the same thing and expecting (and getting!) different results. You have to love it!

[Jan]

@Jim: This bug has been fixed for the following releases and for your release also - you find the required new functions in the bug report.

It seems to be a simple problem with an uninitialized static variable and it is strange that such a bug was not found before the release.