Re: Text tools, line endings, and O_TEXT

[Eric Blake]

John Cowan <cowan <at> ccil.org> writes:

> Are you saying that an O_TEXT open under Cygwin, given a binary mount,
> will cause newline translation?  I had assumed that O_TEXT would be
> ignored under Cygwin just as it is under Linux.

Yes.  The cygwin behavior is as follows:

open(O_RDONLY), fopen("r") - open the file using the mount point rules to 
decide whether the file is text or binary

open(O_RDONLY|O_BINARY), fopen("rb") - open the file forced in binary, 
regardless of mount point

open(O_RDONLY|O_TEXT), fopen("rt") - open the file forced in text, regardless 
of mount point

pipe() - create pipe with mode determined by CYGWIN env variable (default setup 
is binary, but if $CYGWIN includes "nobinmode", pipe becomes binary)

binary mode reads the file as is, text mode strips all \r and ends the file at 
the first ^z.

freopen(NULL, mode, stream) - works to swap existing mode of an open fd-based 
stream between text and binary (but currently fails on special streams, such as 
from fmemopen)

#include <io.h>
setmode(fd, <0, O_BINARY, or O_TEXT>) - works to swap existing mode of an open 
fd, as well as any open stream based on that fd (using 0 restores mount point 
default mode)

> 
> That said, I believe it is good for portability to use O_TEXT
> and O_BINARY (and their stdio equivalents) in all software intended
> to be portable.

Using O_BINARY is important, however, I claim that using O_TEXT in portable 
code should be a very rare circumstance.  There's a reason that C99 only calls 
out fopen("r") and fopen("rb"), but not fopen("rt").

-- 
Eric Blake