Hi Paul,
Thank you for your prompt and extensive reply. I fully understand your
caution as regards the scientific correctness of Praat's results, even
though it would be better, from a system distribution point of view, to
avoid embedded libraries.
I will have a look at the specific GSL problem you mentioned.
Best,
Rafael
* Paul Boersma <paul.boer...@uva.nl> [2013-04-11 11:26]:
hi Rafael,
for several of these libraries we had to include corrections and
patches, e.g. to make them Unicode-compatible or to make error
reporting compatible with how it's done in Praat. As for GSL, it
contains bugs. For instance, the function fisherQ in Praat uses the GSL
version of incomplete beta, which is correct, but its inverse just
crashes on some not-so-uncommon combinations of parameters, so we have
an inversion computation that works better for this example than GSL
does (see test/num/fisherQ.praat for GSL's crash).
In other versions of GSL, these bugs might be reversed. This means we
cannot test correct execution of Praat with other versions of GSL. We
can test Praat only with the version(s) we have control over, so we
need to include the source code in our project. We just cannot afford
random bugs in the DSP code, because such bugs influence research
results. We therefore link with external libraries only in the case of
GUI and graphics; those libraries may contain bugs there as well (e.g.
Ubuntu 12.x broke the update of the Save menu), but those are easily
spotted and reported by users and don't effect scientific correctness.
best wishes,
Paul
_____
Paul Boersma
University of Amsterdam
Linguistics
Spuistraat 210, room 303
1012VT Amsterdam, The Netherlands
http://www.fon.hum.uva.nl/paul/
--
To UNSUBSCRIBE, email to debian-med-requ...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Archive: http://lists.debian.org/20130412034600.gk2...@laboissiere.net
