Re: new function: textscan.m

[Liam Groener]

On Oct 22, 2010, at 10:34 PM, Ben Abbott wrote:

> On Oct 23, 2010, at 10:14 AM, Liam Groener wrote:
> 
>> On Oct 22, 2010, at 6:46 PM, Ben Abbott wrote:
>> 
>>> On Oct 23, 2010, at 8:53 AM, John W. Eaton wrote:
>>> 
>>>> On 23-Oct-2010, Ben Abbott wrote:
>>>> 
>>>> | I've made an attempt to implement the missing function textscan.m
>>>> | 
>>>> | If there are no suggestions for improvement, I'll commit.
>>>> 
>>>> +  if (nargin > 2 && isnumeric (varargin{1}))
>>>> +    N = varargin{1};
>>>> 
>>>> I think it would help to quickly understand what N is if you used
>>>> nlines or similar instead of N.  Also, we generally try to avoid
>>>> uppercase variable names in Octave.
>>>> 
>>>> +  if ((! strcmp (class (fid), "double") || fid < 0) && ! ischar (fid))
>>>> +    error ("textscan: first input argument must be a valid file id, or 
>>>> string.");
>>>> +  endif
>>>> +
>>>> +  if (! ischar (formatstr) && ! isempty (formatstr))
>>>> +    error ("textscan: second input must be a format specification.");
>>>> +  endif
>>>> 
>>>> Maybe I'm just slow, but I have a harder time understanding negative
>>>> conditions like the ones above.  Instead of checking the conditions
>>>> that lead to errors, I find it simpler to write and easier to
>>>> understand code later if I test the conditions for success instead.
>>>> For example, instead of the above, I would write something like
>>>> 
>>>> if (isa (fid, "double") && fid > 0 || ischar (fid))
>>>> if (ischar (formatstr) || isempty (formatstr))
>>>>   ## ... code to do the real work here ...
>>>> else
>>>>   error ("textscan: second input must be a format specification");
>>>> endif
>>>> endif
>>>> else
>>>> error ("textscan: expecting first argument to be a file id or character 
>>>> string");
>>>> endif
>>>> 
>>>> Is that condition on formatstr correct?  Is it OK for it to be empty
>>>> if it is not a character string?
>>>> 
>>>> Note also that isa is probably better than class+strcmp.  But what
>>>> happens if fid is a matrix?  Should we check for that?  Should we
>>>> maybe have a is_valid_file_id function?  Maybe that would also be
>>>> useful in other places too.
>>>> 
>>>> jwe
>>> 
>> Hi Ben,
>> 
>> I thought that, in Matlab, N is the number of times that the format string 
>> is repeated (as in textread), not the number of lines to be read. Did you 
>> intend to make this change? (Or am I all wet?)
>> Liam
> 
> I have never used texscan before this week. It would be wise to be skeptical 
> of my understanding for how Matlab's version works.
> 
> Can you provide me an example that illustrates the difference between 
> repeating for format string, and reading the number of lines?
> 
> Ben
> 
Well, I haven't used textscan either. (I don't have Matlab.) I got my 
impressions of how textscan works from a Matlab book. I modified the example 
script I sent you the other day as follows:

B = [30 40 60 70 80];
fid = fopen('myoutput','w');
fprintf(fid,'%g miles %g kilometers\n',[B;8*B/5]);
fclose(fid);

[a,b,c,d] = textread('myoutput','%f %s',2)

fid=fopen('myoutput','r');
C = textscan(fid,'%f %s',2);
C{1}
C{2}
C{3}
C{4}
fclose(fid);

>From my understanding, both the textread and textscan parts of this script 
>should give more or less the same output. Note that, at least the textread 
>part, reads all five lines of the file, with four values per line, with N=2.

Liam G.