[fpc-pascal] FPC 3.0.4 DOS GO32V2

[Nikolay Nikolov]

On 12/03/2017 11:44 PM, Tomas Hajny wrote:
> On Sun, December 3, 2017 22:10, Nikolay Nikolov wrote:
>> On 12/02/2017 10:50 PM, Lubomír Čabla wrote:
>
> Hi,
>
>>> FPC 3.0.2
>>>
>>> function get_segment_base_address(d : word) : longint;
>>> function set_segment_base_address(d : word;s : longint) : boolean;
>>>
>>> FPC 3.0.4
>>>
>>> function get_segment_base_address(d : word) : longint;
>>> function set_segment_base_address(d : word;s : dword) : boolean;
>>>
>>> For get_segment_base_address is correct longint or dword?
>> I think (and I'm the one, who made the change) dword is more correct,
>> because segment bases can be >=2GB and longint is signed, so results in
>> a negative number. In theory, it should only matter if you have range
>> checking turned on, but I think 32-bit linear addresses should only be
>> declared as dword (or longword) and previous code, that used longint, or
>> added longint typecasts to avoid range check errors should be fixed.
> That makes sense (and DJGPP sources assume the same), but
> get_segment_base_address should probably return a dword as well then,
> right?
That is true. I've missed that one, unfortunately :( I'll probably fix 
this in trunk tomorrow.
>
>
>> Maybe I should document it in:
>> http://wiki.freepascal.org/User_Changes_3.0.4 ? I kinda skimped on that,
>> because it only affects existing code, when range checking is on, and
>> usually range checking is turned off for low level selector manipulating
>> code.
> Yes, adding a note there would be better.
Ok, added it.

Nikolay