Bug#939048: Sextractor: improve tests (was: Bug#939048: transition: glibc)
Hi,
On 2019-09-05 13:30, Ole Streicher wrote:
> Package: sextractor
> Version: 2.19.5+dfsg-6
>
>
> Hi Paul, Aurelien,
>
> I maintain both packages (sextractor and iraf-fitsutil).
>
> 1st, quick answer for sextractor: You don't need to care; this is my
> fault/problem: The test is home-made and directly compares string-wise
> the result with a reference catalog, which fails due to minimal changes
> in the numbers -- from a quick view, that should be neglilible. That
I have just looked at the precision changes in math function in glibc
2.29. No degradation of precision is expected except for ccos, ccosh,
cexp, cos, cosh, cpow, csin, csinh, exp10, sin and sincos when they are
used with the non-default rounding mode (round to nearest). I doubt
sextractor changes the rounding mode, so the changes are probably slight
increase in precision due to the usage of new algorithms.
> also happened in the past, and I solved it by updating the reference
> (which is ofcourse not the smartest solution). If you have a good
> general way to handle this, I am happy to apply this; otherwise I will
> just read+compare the numbers with python3-numpy in future.
You might be able to filter the table with awk or other tool to round
the values a bit. But numpy might even be better in that case.
> For iraf-fitsutil, I still have no glue what happens there. That is also
> a bit painful to debug, since the package is mainly written in a domain
> specific language (IRAF SPP; Aurelien may know it from work). So I am
> afraid it may take a bit of time to analyze.
I have just submitted bug#939523 with a patch to fix the issue. I guess
that technically the next glibc upload should add a Breaks: on versions
lower than the fixed version.
Aurelien
--
Aurelien Jarno GPG: 4096R/1DDD8C9B
aurelien@aurel32.net http://www.aurel32.net
Reply to: