Conference smurf::dec_mls_plus

494.0. "dec c compiler problem with -newc on MLS+ V4.0A" by SMURF::BAT (Segui la tua beatitudine) Fri Apr 25 1997 00:09

From:	US2RMC::"[email protected]" "CWOLF.US.ORACLE.COM" 24-APR-1997 22:33:52.86
To:	smurf::bat, [email protected]
CC:	[email protected]
Subj:	Compiler problem on MLS+ 4.0

Barbara, 
 
We're trying to port our product to MLS+ v4.0.  It seems that the new  
compiler has generated at least one case of an object file which causes 
a segmentation violation upon entering a function.  We found that if we 
compiled with the "-oldc" flag, everything is good to go.  We had  
found on MLS+ 3.1, that the following "cfe" flags were required: 
 
-Wf,-XNd15000,-XNp1500,-T100 
 
However, I don't think -XNd15000 and -XNp15000 do anything, since the  
V3.1 mand page for cfe doesn't mention them, also I can pass bogus 
flags to "cfe" as in: 
 
cc -Wf,-bogus test.c  
 
And no error messages get printed. 
 
But back to the point, which is why, when I compile with -oldc it  
works, but with -newc it core dumps.  BTW, since I started composing  
this email, I found why it core dumps, that if I take away a "register" 
qualifier from in front of the offending local variable which holds a  
pointer, it works with -newc or -oldc.  The question is are there  
reported problems with the "register" qualifier with -newc?   
 
Thanks, 
	-Chris 
 
 
+------------------------------------------------------------------------+ 
| ORACLE Corporation                                                     | 
|                                                  [email protected]   | 
| Chris Wolf, Senior Software Engineer             Voice +1.415.506.2529 | 
| World Wide Government & Education                Fax   +1.415.506.7800 | 
| Products Division                                                      | 
+------------------------------------------------------------------------+

% ====== Internet headers and postmarks (see DECWRL::GATEWAY.DOC) ======
% Received: from mail13.digital.com by us2rmc.zko.dec.com (5.65/rmc-22feb94) id AA19516; Thu, 24 Apr 97 22:21:37 -0400
% Received: from inet16.us.oracle.com by mail13.digital.com (8.7.5/UNX 1.5/1.0/WV) id WAA06366; Thu, 24 Apr 1997 22:15:13 -0400 (EDT)
% Received: from erseq6.us.oracle.com (erseq6.us.oracle.com [138.2.202.101]) by inet16.us.oracle.com (8.8.5/8.8.5) with SMTP id TAA01308; Thu, 24 Apr 1997 19:00:40 -0700 (PDT)
% Received:  by erseq6.us.oracle.com (8.6.13/37.7) id VAA08327; Thu, 24 Apr 1997 21:59:09 -0400
% Message-Id: <[email protected]>
% Date: 24 Apr 97 21:56:29 -0400
% From: "CWOLF.US.ORACLE.COM" <[email protected]>
% To: smurf::bat, [email protected]
% Subject: Compiler problem on MLS+ 4.0
% Cc: [email protected]
% Mime-Version: 1.0
% X-Mailer: Oracle InterOffice (version 4.0.4.0.26)
% Content-Transfer-Encoding: 7bit
% Content-Type: text/plain; charset="US-ASCII"
    

From:	SMURF::BAT          "Barbara A. Thomson ZKO3-2/X46 1-2955" 24-APR-1997 22:42:36.26
To:	US2RMC::"[email protected]"
CC:	BAT
Subj:	RE: Compiler problem on MLS+ 4.0


	I don't know if there are probs with register type,
	I'll start up a search on the cc notes file.
	Do you get the core dump if you use the -noansi_alias
	with -newc? (and use register type)
    

From:	SMURF::BAT          "Barbara A. Thomson ZKO3-2/X46 1-2955" 24-APR-1997 23:06:53.70
To:	US2RMC::"[email protected]"
CC:	BAT
Subj:	RE: Compiler problem on MLS+ 4.0


	Ok, I did a quick search of the notesfiles and QAR
	data bases, and turned up nothing (which isn't terribly
	significant, since I only did title searches).

	But, after looking at the few reports that had to do
	with registers and optimizations, I realized that the
	guys who got the answers had little test programs that
	proved their point.
	
	I'll post one for you if you got it.

	Later
	BAT
    

From:	US2RMC::"[email protected]" "CWOLF.US.ORACLE.COM" 25-APR-1997 13:57:56.78
To:	smurf::bat
CC:	
Subj:	Re: RE: Compiler problem on MLS+ 4.0

Barbara, 
 
-noansi_alias  did not help.  I will see if I can reproduce the problem 
in a small test program. 
 
	-Chris 
 

    Chris Wolf called, he says they are having rough time rebuilding
    Trusted Oracle7 on MLS+ V4.0A.
    
    There are not just header problems, there are programs that core dump
    and other things. (He's going to have Norman send me a list of errors.)
    They have many errors.
    
    This code has been built and is running on 12 different platforms --
    from Sun to HP -- including MLS+ V3.1A -- without any changes.
    
    It is worse if they use cc -new, and less bad, but still with errors,
    using cc -old
    
    Is there something obvious we are missing here?
    
    Is there some header file that is different?  Do MLS+ header files
    differ from Digital UNIX?  (accidentally?)
    
    I've asked him if these problems have happened on vanilla Digital UNIX,
    and he says he knows the people who did vanilla Oracle7, he'll ask them
    if they had troubles porting.
    
    What we want is a porting guide from OSF V2.1 to OSF V4.0A...
    are we using the wrong compiler, the wrong header files, what?

    See also notes 440 and 500