Re: preprocessor/2948

[Sean McNeil <sean@mcneil.com>]

The following reply was made to PR c++/2948; it has been noted by GNATS.

From: Sean McNeil <sean@mcneil.com>
To: Neil Booth <neil@daikokuya.demon.co.uk>
Cc: nobody@gcc.gnu.org, gcc-gnats@gcc.gnu.org
Subject: Re: preprocessor/2948
Date: Tue, 29 May 2001 18:01:07 -0700

 I am unaware of the use of
 
 #pragma implementation "filename"
 
 all files have
 
 #pragma interface
 
 and
 
 #pragma implementation
 
 I just ran this through gcc 2.95.3 and now I'm confused even more.  It appears
 that the output of the normal gcc invokation is mostly correct and the 2 stage
 invokation with the -E flag is missing some compilation information.
 
 The largest difference and what triggered this whole investigation on my part
 has been the undefined reference to "operator delete (void*)" whereas in gcc
 2.95.3 it appears to be a reference to __builtin_delete.  There is also some
 reference to __builtin_vec_delete.  Yet the -E output clearly showed me no
 references to the delete operator.
 
 The more I look at this, the more I wonder what this additional code is and why
 there appears to be no template code in the cpp0 output yet there is output
 generated through cc1plus for the pragma implementation.  I'm affraid I know
 too little about any of this.
 
 Sean
 
 Neil Booth wrote:
 
 > Sean McNeil wrote:-
 >
 > >  The pragmas are indeed getting passed through by the cpp0.  So why would
 > >  they compile differently in the two cases?  This is very strange.  It is
 > >  related to that #pragma, though.  Perhaps the cpp is told to parse things
 > >  differently?
 >
 > Are the pragmas merely
 >
 > #pragma implementation
 >
 > or are they
 >
 > #pragma implementation "filename"
 >
 > ?  I think it's an issue with file names not comparing as strings
 > properly because of path information, or something like that.  Does
 > that give you a clue?
 >
 > Is there a
 >
 > #pragma interface
 >
 > somewhere to match the #pragma implementation?
 >
 > Is this a 2.95 regression?  I'm going to re-file this as a C++ bug.
 >
 > Neil.