Path: utzoo!utgpu!news-server.csri.toronto.edu!rutgers!usc!zaphod.mps.ohio-state.edu!ub!uhura.cc.rochester.edu!ee.rochester.edu!rochgate!f201.n260.z1.FIDONET.ORG!John.Passaniti
From: John.Passaniti@f201.n260.z1.FIDONET.ORG (John Passaniti)
Newsgroups: comp.lang.c++
Subject: TC++ review of MSDos compilers
Message-ID: <14.26E0AF3F@rochgate.FIDONET.ORG>
Date: 2 Sep 90 05:25:46 GMT
Sender: ufgate@rochgate.FIDONET.ORG (newsout1.26)
Organization: FidoNet node 1:260/201 - Age Of Chance, Rochester NY
Lines: 45


 > Also for anyone reading this message, if you are going
 > to compare sizes of generated .obj files - you have to
 > look into the object file itself to see how much code
 > and data was generated. There are lots of comment records
 > in the .obj file for debugging, path and timestamp of all
 > header files (autodependencies) used, and other information
 > for the linker. You can use our TDUMP.EXE utility that comes
 > with the professional package or the debugger and tools
 > product to dump out the sizes of code segments and data
 > segments for .obj and .exe files.
 >
 > david intersimone ("davidi")
 > director, developer relations
 > borland
 > davidi@well.sf.ca.us

     [This may have been discussed here in comp.lang.c++ 
before, but I just joined the newsgroup thanks to a local 
Fidonet gateway.]

     Something I have been curious about is if Borland pays 
attention to some of the hobbyist networks, such as 
Fidonet. Recently in the C_ECHO conference, there has been 
discussion about the quality of code generated from Turbo 
C++.

     I won't repost the entire message here, but the basic 
idea was that it appears as if Turbo C++ was derived from a 
compiler designed for a different processor than the Intel 
80x86 family.  The author claimed it looked as if the 
compiler was generating code using a Motorola mentality. He 
showed how a compiler for a 680x0 might optimize things, 
and then showed how the code generated from Turbo C++ 
showed those same kind of decisions.

     Another message dealt with a bug found in the code 
generated for switch statements.  It had to do with 
unsigned values not being compared in the right sequence, 
giving rise to subtle errors.

     I've kept these messages and would be happy to forward 
them to you.  But my real question is if Borland pays any 
attention to hobbyist networks like Fidonet, or even to 
Usenet.