On 13.09.2017 15:33, Mattias Gaertner via Lazarus wrote:
Also were is there an ambiguity anyway (given all units are used with their
full name)? The compiler reads (in my above example) the token sequence ID
Point ID and thus can find the unit without any backtracking just as for
ordinary methods. Even once the default namespace option is added that
won't change as the compiler will simply test all provided namespaces as
prefixes and then the name as is. It's just a longer list of names than
usually to test for.
Let's say you have 'a.b.c.d'. Before namespaces the compiler could
simply resolve a, then b, then c and finally d.
With namespaces the compiler has to read the full 'a.b.c.d',
then search for a.b.c, then a.b, then a and then resolve the rest as
before.
I think you are wrong. I didn't study how the compiler resolves the
identifiers internally, but when I developed dotted unit support for
CodeTools I did it so: I register all namespaces and subnamespaces from
known units (for uses clause) and used units (for implementation code)
and resolve the identifiers from left to right - easy and how it's
always been.
In your example you always know if 'a' is a namespace or an identifier.
The identifier always wins agains a namespace:
unit ns1.Unit1;
interface
var
A: string;
ns1: Integer;
implementation
initialization
ns1.Unit1.A := 'a'; // << error illegal identifier
end.
unit ns1.Unit2;
interface
var
A: string;
ns1: Integer;
implementation
uses
ns1.Unit1; // error: duplicate identifier (bug in CodeTools: they
don't show an error here because they don't test agains identifier in
uses clause- but the compiler is correct)
initialization
ns1 := 1;
end.
Everything looks like the compiler uses the same algorithm -> it first
resolves a then b then c and so on, it doesn't need to go backwards.
Ondrej
--
_______________________________________________
Lazarus mailing list
Lazarus@lists.lazarus-ide.org
https://lists.lazarus-ide.org/listinfo/lazarus
