On Sat, July 13, 2013 13:26, Sven Barth wrote: > On 13.07.2013 01:29, Tomas Hajny wrote: >> On Fri, July 12, 2013 20:33, Sven Barth wrote: . . >>> If it is really the first case then I don't really see anything that >>> would not work with 2.6.2 that did work in 1.0.6... (except it's >>> something that relies on implementation details) >> >> Well, I could imagine some potentially relevant cases: . . >> 2) Code using lots of assembly and expecting the original calling >> convention (this falls into your category of "implementation details", >> but >> it's an understandable one if it involves large amount of code which >> would >> need to be not only modified but possibly also debugged thoroughly >> because >> there are more differences than just the way of passing variables - in >> particular ebx needs to be saved). Although in that case 1.0.10 should >> be >> a better option than 1.0.6 - unless the original poster has 1.0.6 only >> and >> not 1.0.10 (in the situation when we do not offer this version for >> download any longer, but I could imagine that finding both would be >> still >> possible). > > Isn't the old calling convention available as "oldfpccall" anyway?
It is, but this may not be sufficient. First, I don't know if ebx is saved automatically in such a case (it might be, but such a requirement didn't exist at 1.0.x times). Second, it might get even trickier if there is code in assembly calling Pascal code (e.g. standard RTL functions) again. Sorry, I know that it's possible to check what happens in such cases, but I'd need cross-compiler for i386 hosted on armel-qnx (anyone considering such target? ;-) ) right now. I guess that it doesn't matter too much since it was just speculation anyway. Tomas _______________________________________________ fpc-pascal maillist - fpc-pascal@lists.freepascal.org http://lists.freepascal.org/mailman/listinfo/fpc-pascal