[AstroPy] Fortran-style and large exponents in ascii.io (PR 3655)

Derek Homeier derek at astro.physik.uni-goettingen.de
Sat Jul 2 14:58:00 EDT 2016


Hi,

first, this is an invitation to review the pull request for Fortran-style exponent character support

https://github.com/astropy/astropy/pull/3655

which has been dormant for a fairly long time because of some test crashes on travis.
I’ve noticed far too late that long exponents (> 3 digits) already were flagged as a known
cause of segfaults on travis, and therefore have now isolated the corresponding tests
and marked them xfail.
Second, in the comments on that PR some existing inconsistencies between the handling
of ‘illegal’ exponents had been discussed; see my last comment for a kind of summary.
Summary of the summary, strings with exponents exceeding float64 range are e.g. read
in as inf and 0.0 in plain Python [float('1.423e388’)], while the ascii.io standard has generally
asked to return such entries as string fields. Numpy OTOH can parse such numbers as
extended precision if specifically asked:

np.float128('1.423e388')*10
1.423e+389

So I’d also like to poll if there is general interest in having ascii.io be able to handle such
numbers as well (set by an option, if necessary), which might then include four-digit
exponents.

Cheers,
					Derek




More information about the AstroPy mailing list