Seems Delphi XE has deprecated all separate variables, and after ten years did the same as FPC, use an absolute trick. And of course they named it differently (FPC: defaultformatsettings Delphi: formatsettings)

Like anything new in Delphi nowadays, it is a record-with-some-helpers

Sigh... always the same -.-

Like anything new in Delphi nowadays, it is a record-with-some-helpers

Just thinking: wouldn't objects in FPC achieve the same as records-with-methods in Delphi? If objects would be extended with "sealed" even the "sudden vmt" problems mentioned some time ago would be gone with simple objects-simulating-delphi-records-with-methods.

Regards,
Sven

In our previous episode, Sven Barth said:

I don't see the point of extending the life of the old syntax. Just let it be as legacy, and implement new functionality Delphi compatible.

With a risk?.
Wait until XF or something.
XE has a really elegant (well, sort of) vmt handling...

You are aware that I'm talking about "object" (the old TP style OOP) not "class" (the Delphi style OOP)? "object" is deprecated in Delphi since I don't know how long, but in FPC it's still maintained and even extended.

But what do you mean with "elegant vmt handling" in XE?

Regards,
Sven

Marco van de Voort wrote

Seems Delphi XE has deprecated all separate variables, and after ten years did the same as FPC, use an absolute trick. And of course they named it differently (FPC: defaultformatsettings Delphi: formatsettings)

Oh well, this is an easy one, for once:

var
   formatsettings : TXYZ; absolute defaultformatsettings;

Michael.

Sven Barth wrote

You are aware that I'm talking about "object" (the old TP style OOP) not "class" (the Delphi style OOP)? "object" is deprecated in Delphi since I don't know how long, but in FPC it's still maintained and even extended.

In delphi it is still maintained and even extended too.

For example objects got the extended syntax as well as classes:

   Obj1 = object
     type
       TF = integer;
     var
       F: TF;
   end;

Best regards,
Paul Ishenin

Marco van de Voort wrote

Like anything new in Delphi nowadays, it is a record-with-some-helpers

Time to implement this kind of records in FPC?

I tried once but failed because of big changes they require.

Best regards,
Paul Ishenin

In our previous episode, Paul Ishenin said:

If generics are more mature, we are going to need it.

E.g. I'm thinking of parameterizing some of the core fpimage readers/writers with stuff like

   YUV422 = record
             pixel : word;
             function red:integer;    inline;
             function green:integer;  inline;
             function blue:integer;   inline;
             function alpha:integer; inline;
            end;

for pixel value

Since r/g/b are only shift+ and mask, and alpha is even constant, that would speed up the reader etc.

I still plan to keep a basic polymorphic everything to 64-bit pixel readers, but have avenues to instantiate a couple of faster paths. I hope to reach speedups in the magnitude of 50 times. (currently my routines are 100 times faster, but that is handcoded for a couple of formats without any general purposeness)

I am referring to the reworked rtti handling and the TVirtualMethodInterceptor class in the rtti unit. (Delphi XE)

In our previous episode, Michael Van Canneyt said:

That + deprecate the old separate ones and updated user changes trunk? I'll do this today when I have time.

Then the only question is if we should deprecate defaultformatsettings. I decided to not do this today. Even if we do this, a lot of parameters for the overloaded "formatsettings" routines need to be renamed (to aformatsettings?) for this first to avoid accidental clashes.

I would never deprecate it. It was our solution, and there is no need to deprecate it. The other one is 'for delphi compatibility'.

Michael.

Ok, I hadn't known that.

At work we use Delphi 6 (since this month we are migrating to Delphi
2007) and I myself have only prevely used Delphi 2006 (I then switched to Lazarus).

Regards,
Sven

In our previous episode, said:

Never is a long time. Well, anyway even if we deprecated it in the 2.6 cycle, really removing it is then only in 2.8 or 2.10.

But I don't have really strong feelings about this. It's just that if the global vars are deprecated, and the defaultformatsettings are, we can kill them if the absolute tricks cause problems. We can leave them in indefinitely till there really are problems.

What about just using "sealed objects" for this in mode Delphi (maybe change record in that mode to mean "sealed object" by default)? I don't know how much sense that makes, but as both reside on the stack and both can contain methods (I'm talking about the extended records) I see no problem here.

Regards,
Sven

Ah ok. Thanks for clarifying.

Regards,
Sven

Objects don't know about "case" block inside.

Best regards,
Paul Ishenin

IMO not. It's simply an FPC construct.

Right. -.- I forgot about that, although I use it here and then...

Regards,
Sven