On Tue, Apr 23, 2013 at 09:29:55AM +0200, Andreas Tille wrote:
On Tue, Apr 23, 2013 at 08:39:11AM +0200, Rafael Laboissiere wrote:
I forgot to say two things:
1) I reverted my previous changes in Git regarding the linking
against the native GSL library. The version in HEAD uses now the
GSL sources shipped in the pristine tarball, since this is the
desire of the upstream author.
Uhmmm, so your negotiations about using packaged libraries did not
helped? That's sad. Was it a pure "desire" or did he gave reasons for
this. I mean: I could perfectly follow good reasons based on a diff or
a test case that might fail when using option B instead of option A.
However, following pure desire in contrast to my technical understanding
is not really convincing to me.
Or maybe there was no new statement from upstream about this? Rafael?
Op 13 apr. 2013, om 22:24 heeft Rafael Laboissiere het volgende geschreven:
It fails, unfortunately:
Error: Script assertion fails in line 32 (undefined):
abs (fisherQ (invFisherQ (i/1000, df1, df2), 1, 100000) - 3/1000) < 1e-11 ; 3 1 100000
OK, this may be due to only a minor precision problem in GSL or to
another set of NaN values in GSL, so determining whether a dynamic GSL
library would work for Praat at all will require more testing on our
part. For the time being, and perhaps indefinitely, we'll stay with
statically linking our libraries, at least for all the editions we
produce from here.
best wishes,
Paul