Will Honea wrote in a message to Mike Bilow:

MB> Watcom C++ definitely aligns by default on doubleword boundaries. 
MB> I checked  the documentation and you are right that it claims to
MB> align  by default on byte boundaries, but this seems to be wrong 
MB> when checked against generated code.  I suspect that this  changed
MB> at some point in the compiler and was forgotten in  the
MB> documentation.  In my opinion, any 32-bit compiler  should default
MB> to doubleword alignment, and GCC gets a fair  amount of speed from
MB> doing so.  

 WH> Mike, Watcom (10b) defaults to byte bounds for 16 bit code -
 WH> both C and C++ - as I found out the hard way porting some
 WH> MSC 6.0 sources.  I never even looked at the 32 bit so you
 WH> probably saved me some head scratching when I try to access
 WH> the old data files from the OS/2 port :<}

I've worked extensively with Watcom 16-bit code, but I can't say that I
ever tried using the compiler default on this.  My standard makefile always
sets the packing explicitly, so I have not run into any problems.  Where
Watcom will burn you badly when moving from Microsoft C 6.0 is in segment
alignment, since Watcom's 16-bit OS/2 compiler expects "para"
aligned segments and MS C6 requires only "dword" aligned
segments.  If you try to link a MS C6 library, such as IBM's DHCALLS.LIB
for device drivers, with code generated by Watcom, then the odds are that
the linker will mess up (although it can work by coincidence).
 
-- Mike


---