Re: [xml] compiler warnings from HP ANSI C on HP-UX 11iv2 (11.23)

[Rick Jones <rick jones2 hp com>]

Bjorn Reese wrote:
> Rick Jones wrote:
>
> > Attached to this message is the typescript of a compile with +w1 and 
> > +M2 set to enable all the warnings the compiler sees fit to emit.  
> > They seem to be similar to those emitted by the HP-UX 11i v1 (11.11) 
> > compilers but with nicer message text.  This was also with +DD64 set 
> > to get a 64-bit compile.
>
>
> 6 of the warnings are for trionan.c, which is my domain.
>
> All 6 warnings can safely be ignored. I use explicit casts, but that
> does not appear to satisfy the compiler.

I suspect it may be getting overly paranoid about the possible alignment of the 
components.  IMO it seems silly to warn about casting a pointer from something 
with stricter alignment to one with looser alignment - such as the one for line 
181 in your code.  Of course, compilers can be strange and mysterious things. 
I'll have to tap my compiler contact's on the shoulder and ask them about it.

> I know that these warnings can be ignored by adding the +W1036 compiler
> option, but I am wondering if it possible to put this instruction into
> the source code, e.g. in the form of a pragma. I looked through the HP
> aC++ manual, but found nothing obvious.
>
> Then again, maybe I am overreacting here. Are the above warning level
> part of the normal build, or did you change it to see if the compiler
> would find some potential problems in the code? (Some of them, like
> the assignment of long to int, are definitely work looking at.)

The warnings are _not_ part of a normal build - I can see where I probably 
should have used a differently worded subject...  Without either of +w1 (more or 
less the equivalent of gcc -Wall) or +M2 they are not emitted by the compiler. 
So, when someone does ./configure and then make they will not see these.  If 
they do CFLAGS="+w1 +M2" ./configure and then make they will.  I suspect that 
only compulsive nitpickers :) will be setting those options.

I'm passing them along only for completeness and on the off chance they are 
pointing-out possible future trouble spots.  Some of the stuff about pointer 
math being given to an int rather than a long for example.

rick jones