Problem (bug?) with dotted unit names and default name spaces.
lazarus / 64-FPC development

----------------------------------------------------------------------
glorfin · 2020-08-21 23:56

I have two units, LMath.uPointsVec.pas in a package LMComponents, and
FitLogExp in the project (ConstrFit).
Both of them use the same type TRealPoint, defined in lmath.uTypes
unit.

In LMath.uPointsVec:
uses  LMath.uTypes; // complete identifier

In FitLogExp:
uses uTypes; // partial identifier, but project contain -FNlmath
switch.

Line in FitLogExp:
DataPoints.Append(Pt); //DataPoints is of type TPoints, defined in
uPointsVec

leads to following error message:
In FitLogExp.pas(62,25) Error: Incompatible type for arg no. 1: Got
"LMATH.UTYPES.TRealPoint", expected "LMATH.UTYPES.TRealPoint"

If I use complete identifier in uses clause of FitLogExp, everything
works fine.

----------------------------------------------------------------------
TRon · 2020-08-22 00:11

Quote from: glorfin on August 22, 2020, 01:56:09 amIn
FitLogExp.pas(62,25) Error: Incompatible type for arg no. 1: Got
"LMATH.UTYPES.TRealPoint", expected "LMATH.UTYPES.TRealPoint"
You got to love such 'explanatory' errors 

I do not know for sure if it was intended to have the distinction, but
no matter what: it is a bug. Either the distinction is intended, in
which case the error message is ambiguous or it is not intended to
have the distinction and that fact in itself is the bug.

----------------------------------------------------------------------
glorfin · 2020-08-22 00:32

Well, if a type is incompatible with itself, it looks very much like a
bug. Most probably, compiler routine which resolves unit names, first
creates a list of units, found directly, afterwards it searches once
more for units, combining name space provided in -FNxxxx and unit name
in "uses" clause, but then forgets to test if this unit was already
found, and ends up with two distinct items for the same unit.

----------------------------------------------------------------------
TRon · 2020-08-22 00:40

Your analysis reads very plausible, but fact is that it doesn't matter
what caused it. It is just plain wrong (any which way you view/explain
it, if even for undocumented behaviour)

Nice find btw, even knowing that what you are working on made it just
a matter of time.

Please file a report if you are able to.

----------------------------------------------------------------------
PascalDragon · 2020-08-22 10:29

Quote from: TRon on August 22, 2020, 02:40:39 amPlease file a report
if you are able to.

And please provide a complete, but small and self contained example.

----------------------------------------------------------------------
glorfin · 2020-08-22 17:36

Done. Issue 0037622.

----------------------------------------------------------------------
PascalDragon · 2020-08-23 11:45

Thanks