In the *program* file of my application, I coded
//....
Screen.Cursor := cdHourglass;
//....
ScreenCursor := crDefault;
//....

This at first generated two compile time errors :
  'Undeclared identifier : crXxxxxx' (or similar, my Delphi speaks
French)

I know the crXxxxx identifiers are defined in the Controls unit, so I
added it in the 'uses' clause and everything went Ok.

However, the unit Forms was already in the same 'uses' clause, and I
checked that Forms in turn 'uses' Controls...

So why must I explicitely specify it in *my* 'uses' clause?

Not really a problem, just curiosity...

Pierre Gemis

One exception is when you inherit a class or object from one unit to
another; all methods and properties (except the private ones) that are
part of the original class/object, are automatically propagated to the
new unit, because they are part of the inherited class/object which IS
declared in the new unit's interface.

-- Erwin Dokter
   mailto:edok...@home.nl
   http://members.home.nl/edokter

What Delphi wants to determine is the =minimal= number of ".PAS" files
that it must read in order to "understand" what your Delphi program is
saying.  If you refer to a constant ("crXxxxx"...) Delphi wants to know
where to find it.  It doesn't want to have to go chasing for it --
"lessee, we use Forms, and Forms uses X, Y, and Z, so maybe it's in X,
Y, or Z... or maybe it's in one of the units that X, Y, or Z uses..."
Pretty soon, you've got Delphi reading the source-code to the whole
<$%%@%!!> world trying to find a definition for "crXxxxx!"  :->

So Delphi obliges -you- to tell it where to find that definition, in a
"Uses" clause occurring either in the interface part or the
implementation part (and there's a subtle difference between the two).

Pragmatically speaking, this is not only more efficient and more
practical -- it is also more reliable.  It's better that you know
exactly what the compiler's doing ... what it's looking at, where it got
that definition for "crXxxxx" ... than to allow the compiler to "try to
be helpful."

Computers can't be "helpful" to humans.  There was once an attempt by
IBM to produce a compiler that would "automatically fix" errors in a
program.  The (probably apocryphal...) story goes that, as a final test,
IBM fed this compiler a copy of Abraham Lincoln's "Gettysburg Address,"
and it compiled with no errors.

Hmmm....

Thanks for the explanation, it was not only very clear but also very
'alive' (if one may say so in English) and funny to read.

Pierre Gemis