>>>>> "Robert" == Robert Dodier <robert.dodier at gmail.com> writes:
Robert> linear algebra code, and there is a persistent bug in one
Robert> of the test suite files.
Are you're talking about problem 65 and 68 in rtest_trig? These
failures are caused by differing ideas on where the branch cuts for
the special functions are. I think we agreed a little while ago that
we should use CL special functions as much as possible, which Barton
implemented.
If we use the branch cuts as defined in CL [1], then our rectform
doesn't get them right.
For example, we fail the test (with cmucl) for atanh(-2). According
to ANSI CL, atanh has a branch cut for on the real axis for x < -1
which is continuous with quadrant III. Thus, atanh(-2) should be
-log(3)/2 - %i*%pi/2. CMUCL gets this right. Clisp produces
-log(3)/2 + %i*%pi/2. rectform(atanh(-2)) produces the same as Clisp.
(I think Clisp is buggy here.)
Problem 68 is the bigfloat equivalent to problems 65. I think there
are some bugs here that I need to fix. I'll do that asap.
Robert> What is the status of the help system? If it is not working
Robert> at the moment, what can we do to bring it to a working state?
It seems to work for me, using maxima-local. (I rarely install the
cvs versions.) Is there something in particular you would like to see
tested?
Ray
Footnotes:
[1] (see
http://www.lispworks.com/documentation/HyperSpec/Body/f_asin_.htm and
http://www.lispworks.com/documentation/HyperSpec/Body/f_sinh_.htm)