Hi, not sure if this is a bug: <hansl> matrix a = zeros(3,1) matrix b = ones(2,3) matrix hu = a' * b' hu = a'(b') hu = a'b' # fails </hansl> Apparently gretl interprets the last line as "(a'b)'" which kind of implies that the implicit multiplication with the transpose sign has higher precedence than the actual transposition of b.

> "Kind of". But couldn't this be interpreted as a case of simple > left-to-right evaluation in the absence of another deciding criterion? Well, I admit that according to table 3.1 in the command reference the unary transpose and the binary transpose-multiply have the same priority, and so it would boil down to left-to-right. But I'm not sure that's right. Semantically it's clear that transpose-multiply just exists to mimick mathematical handwriting (on the blackboard or on paper). So wherever it's meaningful, "'" must be replaced by "'*". And then in the expression "a'*b'" the right precedence kicks in. Or to put it differently: Nobody would ever write "(a'b)'" on the blackboard as "a'b'", because I think it's a broad consensus that the two expressions are different.

On Sat, 2 Jul 2016, Sven Schreiber wrote: >> "Kind of". But couldn't this be interpreted as a case of simple >> left-to-right evaluation in the absence of another deciding criterion? > > Well, I admit that according to table 3.1 in the command reference the > unary transpose and the binary transpose-multiply have the same priority, > and so it would boil down to left-to-right. > > But I'm not sure that's right. Semantically it's clear that > transpose-multiply just exists to mimick mathematical handwriting (on the > blackboard or on paper). So wherever it's meaningful, "'" must be replaced > by "'*". And then in the expression "a'*b'" the right precedence kicks in. > > Or to put it differently: Nobody would ever write "(a'b)'" on the > blackboard as "a'b'", because I think it's a broad consensus that the two > expressions are different. I agree with Sven on this, but OTOH what people expect on the basis of their mathematical background counts up to a point. The important thing is to stick as closely as possible to the principle of least surprise, which is, in turn, dictated by what other matrix-oriented languages do: as far as I can tell, the expression a'b' * is equivalent to (a') * (b') in Ox, Gauss and Julia * is illegal in Octave (which doesn't have, I think, the binary ' operator) so we're the exception. So I guess it's just another case of choice between backward-compatibility and pragamtism.

Am 02.07.2016 um 19:45 schrieb Allin Cottrell: > On Sat, 2 Jul 2016, Riccardo (Jack) Lucchetti wrote: > >> a'b' >> >> * is equivalent to (a') * (b') in Ox, Gauss and Julia >> * is illegal in Octave (which doesn't have, I think, the binary ' >> operator) >> >> so we're the exception. So I guess it's just another case of choice >> between backward-compatibility and pragamtism. > > I take the point here, but changing gretl's behavior in this regard will > not be trivial. Unary transpose is unique as an operator that appears to > the right of its operand, and we cannot in general tell unary transpose > from binary transpose-multiply without "look behind", which we don't > have in our parser at present. > > I'll think about this some more, but I'm not going to attempt a change > before the forthcoming release. > Absolutely it's better not to rush it. About the "look behind" stuff I'm not sure I understand. Currently I guess the parser already needs to decide whether any "'" is unary or binary. Or do you mean that it's always treated as binary unless the next thing is a closing parentheses or the end of line? (Although even then the "next thing" situation would imply some looking behind.)