Re: Cross compilation, warnings and errors

On Mon, 24 Jan 2011 13:09:45 -0300
Luis Lavena <[email protected]> wrote:

> Thank you for your complete and thoughtful response, but seems there
> is a confusion.
> 
> Perhaps my english is not accurate, but what I pointed out is that
> during FreeTDS cross compilation I received the compilation error, not
> anything else.

There seems to be a rule on the Internet that anyone concerned that their English is unclear in fact writes perfectly clearly.  :-)

We're both a little confused.  I didn't realize bsqldb.c included sqlfront.h (it shouldn't).  

But, when I build bsqdb, I don't see the symbol redefinition problem on Linux, Win32, or Win64.  (The latter two built using Microsoft's compiler du jour.)  

> That means FreeTDS headers are not evaluating for missing or defined
> symbols prior to define their own.

Correct: LPCBYTE is defined unconditionally as a typedef.  Microsoft's sqlfront.h does the same thing.  I'm saying that's OK; if you have a symbol-redefinition error, it's better not to include the *other* definition.  This code would *not* be correct in sqlfront.h:

	#ifndef LPCBYTE
	#define LPCBYTE /* ... */
	#endif

because such code *assumes* the prior definition is equivalent to the one it would define itself.  That's not a safe assumption: LPCBYTE has no commonly recognized meaning; there's no guarantee that the prior definition conforms to FreeTDS's requirements.  

My advice, then:

1.  Substitute sybfront.h for sqlfront.h in src/apps/bsqldb.c.  I think that will fix the problem.  

2.  If not, look for a way to exclude unneeded defintions when including windows.h.  When using Microsoft's version, I define WINDOWS_LEAN_AND_MEAN before including windows.h.  What you'd need for mingw I don't know.  

HTH.  

--jkl