Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/18/84 (Fortune 01.1b1); site graffiti.UUCP
Path: utzoo!linus!decvax!ucbvax!ucdavis!lll-crg!seismo!ut-sally!ut-ngp!shell!graffiti!peter
From: peter@graffiti.UUCP (Peter da Silva)
Newsgroups: net.micro.atari
Subject: Re: ST programs without AES & VDI?
Message-ID: <312@graffiti.UUCP>
Date: Wed, 16-Oct-85 10:23:10 EDT
Article-I.D.: graffiti.312
Posted: Wed Oct 16 10:23:10 1985
Date-Received: Mon, 21-Oct-85 06:33:53 EDT
References: <10613@ucbvax.ARPA> <558@riccb.UUCP>
Distribution: net
Organization: The Power Elite, Houston, TX
Lines: 19

> You also have to be careful when passing arguments because the C compiler
> does not automatically convert everything to long words when passing arguments
> on the stack. chars and ints use MOV.W data,(SP)- and pointers and longs
> use MOV.L data,(SP)-. This can can a crash due to bus errors because if
> you're not careful you could pick up an invalid pointer.
> 
> Don't ask me why they decided to do it this way, probably just to cut down
> on generated code and improve speed, but it definately will cause problems
> with LOTS of software that assumes that arguments passed to function are
> the same size and resized to the proper size by the called function.

There is certainly precedent for this sort of thing: the original Ritchie
compiler did things this way. If your code assumes that int promotes to long
or to (void*) on a function call it will die on more machines than just this
one.

Now why they didn't follow the VAX' precedent and make ints 32 bits, and shorts
16 bits, I don't know. Are you sure that the compiler uses 16-bit ints? Strikes
me as silly on a 68000.