Re: [PATCH] Improve compatibility between M4 and CPP.

Hi Eric,

Op 8 jul 2009, om 13:35 heeft Eric Blake het volgende geschreven:
> If our synclines are not suitable with the c99 preprocessor already,  
> then
> we are not compliant with POSIX.  But naming a new option -- 
> synclines-cpp
> makes it sound like synclines are not c99-ready unless you use this  
> new
> option.  That is, until I read your documentation of the option:

The intended meaning of the name "synclines-cpp" was "synclines as  
generated BY cpp", not "for cpp".

> So in effect, what it looks like you are
> trying to introduce is an option to make m4 line directives NOT  
> affect C99
> preprocessing (ie. they will be silently ignored by the c99  
> preprocessor,
> without affecting the c99 magic macros __LINE__ and __FILE__), but  
> still
> be recognizable enough for a human reading the file to still  
> correlate the
> lines back to previously preprocessed source code.

Either my attempt at documenting this option is poor, or I didn't  
explain myself properly the first time (or both), but you did not  
catch the intent behind this.

The purpose of these changes is to allow running M4 _between_ the C  
preprocessor and _compiler_, i.e. between "cpp" and "cc1":

(C code) -> CPP -> (preprocessed code) -> M4 -> CC1

 From that point, unfortunately one has to consider that the syncline  
format accepted by "cc1", when it is separated from the preprocessor,  
is not "#line N"; instead cc1's syncline information should be  
provided by the (archaic, K&R style) different "# N" format. See  
`do_linemarker' in GCC's libcpp/directives.c, compare with `do_line'.

So in order for CC1 to report messages/errors/warnings at the correct  
locations in the original source, M4 should output synclines using  
this format one way or another. Hence the new option.


> And if we do add that as an option, we should break this into multiple
> patches - one that adds an optional argument to __line__/__file__, and
> another patch that introduces the new option, rather than doing it  
> all in
> one patch.

My opinion is that they are related, i.e. provide the minimal set of  
features that allows to fit M4 between CPP and CC1. Now you are the  
maintainer and you probably now better, including how to apply these  
two changes on your branch separately from each other.

Regards,

-- 
k