Re: [Mingw-users] Strategies for dealing with Unix-style files on Windows

[Keith M. <kei...@us...>]

On 20/03/11 17:02, Paul Leder wrote:
> I'd like to handle Unix-style as well as DOS-style line endings on the 
> MinGW version of my code. Does anyone have any thoughts on this?

FWIW, my 2p.

> The problem is that I'd prefer to open input files in text mode.

>From a POSIX perspective, the MS distinction between _O_BINARY and
_O_TEXT is an aberration; indeed these I/O modes aren't even defined for
POSIX versions of GCC. If you value portability of your code, (and your
desire to handle POSIX line endings seamlessly, alongside MS endings
suggests that you may), don't rely on MS text mode. Ever. Write your
code to explicitly handle CRLF as equivalent to NL instead.

> When  your open a NL-only file in text mode, however, fpos/fseek 
> don't work on the file. If you get a file position, and then seek to 
> it, you end up somewhere else in the file. This works for CR/NL 
> files, of course, but not NL-only files.

I'm confused. It has always been my understanding that fpos/fseek
cannot be used reliably with any file opened in text mode, because
your application never sees the file content as it really is; fseek
sets position in terms of raw byte count within the physical file, but
because not all of those bytes are actually seem by the application,
it is uncertain if fpos computes position on a consistent basis.


> I've also tried fgetpos/fsetpos, but they have exactly the same
> problem.

Since these presumably rely on the same underlying system code, that's
hardly surprising.

> I've also tried _setmode, to set the file mode to binary before the fpos 
> and back to text after the fseek. This works for NL-only files, but 
> fails for CR/NL files. MSDN also says that you should call _setmode 
> before doing any file I/O, so I wouldn't want to do this anyway.

No, you shouldn't do this.

> AFAICT, if I want to deal with "text" files that are NL-only, and I
> want to fpos/fseek on those files, then I must open them in binary
> mode. Unfortunately, when I do this, I get other errors in my code,
> which I haven't looked at yet. I guess I may just have to fix these
> instead.

I guess you should probably fix them regardless.

> I suppose one point of view is that you shouldn't open NL-only files
> in "text" mode on Windows, because they're not what Windows considers
> to be "text". Now that I've written it down, I suppose that's the
> only logical answer.
> 
> Thoughts?

My POV: don't ever rely on MS text mode. Write your code as if all files
are "binary", then you won't have to rewrite it, if/when you want to use
it on a POSIX system, where MS text mode is meaningless.

-- 
Regards,
Keith.