From: Eric Backus <ericb AT lsid DOT hp DOT com>
Subject: Re: O_TEXT/O_BINARY grief
To: jes AT mbio DOT med DOT upenn DOT edu (Joe Smith) (Joe Smith)
Date: Tue, 21 Jun 94 14:27:59 PDT
Cc: djgpp AT sun DOT soe DOT clarkson DOT edu
Mailer: Elm [revision: 70.85]

Joe Smith wrote:
> OTOH, can anyone give me a for-instance where treating all six of the
> above characters as 'whitespace' leads to incorrect semantics?  And I
> don't think quoted strings that span source lines count, since there
> are perfectly portable ways to do that without embedding control
> characters in the source.  In the case mentioned (cpp macros), so you
> remove backslash-linefeed pairs and leave the carriage-return: the
> compiler will see the carriage-return as just another 'whitespace',
> the same as the tab(s) you might add at the beginning of the following
> line.
> 
> Is cpp *required* by the standard to remove backslash-linefeed pairs?
> It might conceivably keep the newline as part of the macro body.  If
> you write macros that *depend* on one behavior or another (say,
> splitting and identifier or operator), then you will justly suffer for
> it.

In the ANSI C standard, the first thing that the preprocessor does is
translate the physical file into the C character set.  This
translation includes fixing trigraphs, and converting the physical
file's end-of-line indications into single linefeeds.  After this
translation is done, everything is done UNIX-like with single
linefeeds.

The standard doesn't specify anything about what the physical file's
end-of-line indication must look like.  Presumably, this is system or
implementation dependant.  Presumably, if the physical file is read in
text mode, then everything works out.  This is why text mode *MUST*
understand the system's convention for text files.

Yes, the preprocessor is *required* to remove backslash-linefeed
pairs.  This happens after the above translation, but before almost
everything else.

Whitespace is NOT ignored inside of a quoted string.  So, if text mode
does not translate CRLF to LF, then the preprocessor will see
backslash-cr-lf rather than backslash-lf, and will not remove them.
This would be wrong.
--
				Eric Backus
				ericb AT lsid DOT hp DOT com
				(206) 335-2495