MATLAB: Test existence of files with EXIST

Actually the command exist(FileName, 'file') seems to be sufficient to check the existence of a file. Therefore I used this code to check, if the input of a function is an existing file (thanks to David who has found the bug):

function Hash = DataHash(Data)
...
if exist(Data, 'file') ~= 2
   error('File not found: %s.', Data);
end

The help text of exist explains, when the value 2 is replied:

 2 if A is an M-file on MATLAB's search path. It also returns 2 when A is
   the full pathname to a file or when A is the name of an ordinary file on
   MATLAB's search path

But "when A is the full pathname to a file" does not match, when A is a MEX-, MDL- or P-file, because in these cases 3, 4 or 6 is replied respectively. So let's try to improve the check:

if ~any(exist(Data, 'file') == [2, 3, 4, 6])
   error('File not found: %s.', Data);
end

But even then, exist() is smarter then expected:

File1 = fullfile(matlabroot, '\toolbox\matlab\graph2d\plot')
File2 = fullfile(matlabroot, '\toolbox\matlab\graph2d\plot.m')
File3 = fullfile(matlabroot, '\toolbox\signal\signal\@dspdata\plot')
File4 = fullfile(matlabroot, '\toolbox\signal\signal\@dspdata\plot.m')
exist(File1, 'file')  % 0 !
exist(File2, 'file')  % 2

exist(File3, 'file')  % 2 !
exist(File4, 'file')  % 2

I guess that File1 is not recognized, because plot is a built-in function, while @dspdata\plot (File3) is not a built-in function. But File3 is not an existing file:

fopen(File1, 'r')  % -1
fopen(File2, 'r')  %  3
fopen(File3, 'r')  % -1  !! inspite of: exist(File3, 'file') ~= 0
fopen(File4, 'r')  %  4
fclose('all')

So how can we check the existence of a file in a simple and reliable way?

function Ex = FileExist(FileName)
FID = fopen(FileName, 'r');
if FID == -1
  Ex = false;
else
  Ex = true;
  fclose(FID);
end

But there are still exceptions, because even fopen() is smart also:

cd(tempdir);
fopen('plot.m', 'r')  % 3, file is *found*!

Here fopen() searches in all folders of the Matlab PATH, but actually it should be searched in the current folder only. This has the side-effect, that fopen(name, 'r') is relatively slow. Another idea:

cd(tempdir);
fopen('plot.m', 'r+')  % -1, file is not found

This is faster than the 'r' mode, especially if folders of the PATH are stored on network drives. And requesting write access does restrict the search to the local folder only. But this fails, if the current user does not have write privileges to the file.

The next approach:

function Ex = FileExist(FileName)
dirFile = dir(FileName);
if length(dirFile) == 1
   Ex = ~(dirFile.isdir);
else
   Ex = false;
end

I could not find a file, where this test fails. It is very slow, if FileName is a folder on a network drive which contain very much files. But this is a rare case such that I prefer this test.

Finally a C-Mex using either GetFileAttributes under Windows or _open or _wopen under Linux/MacOS is faster: 10% for existing files, 90% for missing files. But the handling of the unicode strings is not trivial: 2 bytes per wchar under Windows, 4 bytes per wchar under Linux and MacOS, but under Linux wchar's are not used in common, but utf-8 encoded 1 byte per char strings. See Answers: Matlab string to wchar under Linux. I'm going to publish the Mex functions in the FEX, also a DirExist(), because exist(name, 'dir') has similar problems.

Did you consider such effects caused by the smartness of exist() in your programs?
Did a user of your programs run into troubles due to weak tests of file existence, e.g. when the resulting error messages are misleading?
Do your or your programs profit or suffer from the smartness of exist() and fopen()?
Do you think the behavior of these function is explained clearly enough in the help and doc text?
Do you want standard jobs solved reliably by simple commands in Matlab?

NOTE: Usage of the recursive font: I mean that smart is not smart.

MATLAB: Test existence of files with EXIST

Best Answer

Related Question

Best Answer

Related Solutions

MATLAB: Check if a file exists

MATLAB: Avoiding use of cd to resolve filenames

Related Question