[Debian-astro-maintainers] Bug#844490: FTBFS with boost1.62

[Thibaut Paumard]

Package: src:gyoto
Version: 1.1.1-2+b1
Severity: serious
Justification: fails to build from source (but built successfully in the
past)

Dear BTS,

gyoto FTBFS with boost1.62 on several release architectures:
https://buildd.debian.org/status/package.php?p=gyoto

It looks like it was known before the upload of boost1.62 to unstable:
http://people.canonical.com/~ubuntu-archive/transitions/html/boost1.62.html

The failure occurs in the test suite. Some debugging shows that the
culprit lies in lib/Screen.C, line 567. More precisely, the function
acos() overloaded for type boost::multiprecision::cpp_dec_float_100
sometimes never returns. It appears to be somewhat reproducible (I think
the test suite of gyoto fails always at the same point in its execution
on a given architecture, when it fails), but I was not able to isolate a
specific value of the argument that would hang acos().

The test suite can succeed if I decrease somewhat the number of digits,
e.g. using only 50 or even 90 decimal digits instead of 100. The build
failure occurs earlier if I use even more digits, so it looks like a
memory leak of some sort.

Several approaches could hide the problem under the carpet:
 - skipping the test suite on the affected architectures ;
 - decreasing the number of digits.

I'm very reluctant to do any of this because there, even if the test
suite goes through, there is no guarantee that the bug will not bite
users in the middle of a weeks-long simulation.

Besides, decreasing the number of digits means gyoto will loose in accuracy.

Looks like this needs to be fixed in boost.

Kind regards, Thibaut.