Not really a bug but...

["Peter Rockett" <p.rockett@...............>]

Not really a bug but I can find nowhere else to feed back this information:
 
I have been trying to use the QR solver (gsl_linalg_QR_solve) function. A
small piece of feedback: The manual does not make it clear that you have to
explicitly call the QR decomposition before calling gsl_linalg_QR_solve.
Probably one of these things that is blindingly obvious if you know but not
clear if you don't! (More disconcertingly, if you fail to precompute the QR
decomposition, you get a very misleading error message about a vector having
a negative length!) I suspect the same thing goes for LU decomposition as
well?
 
So next time somebody is revising the manual, could I suggest that the need
to precompute the QR decomposition is mentioned. Or I could suggest a minor
reword? But I don't want to sound like I'm moaning - just trying to feed
back some small improvement. Great library that has been the mainstay of my
research for a number of years! 
 
BTW: I have been using it with MinGW on WinXP since around version 1.6. It
just builds 'out of the box'.
 
Best wishes
 
Peter Rockett