Bug#749365: lucene2: FTBFS too much often

[Santiago Vila]

retitle 749365 lucene2: FTBFS too much often
severity 749365 serious
thanks

While checking for "dpkg-buildpackage -A", I built this package twice
on testing and it failed once. See attach.

It also fails on reproducible builds:

https://tests.reproducible-builds.org/debian/rb-pkg/unstable/amd64/lucene2.html

It currently fails in unstable but not in testing.


So this fails approximately half of the time. That's not good enough
for a releaseable package. Being able to build a Debian package from
source should not be a lottery or depend on a coin being flipped.


Emmanuel Bourg wrote:

> The lucene tests have random parameters that are probably the cause of
> the instability observed

Seems likely, yes, but then what the tests do is basically like this:

override_dh_auto_test:
	Roll a dice. Exit with error if the result is 4, 5 or 6.

A package which do tests based on random parameters is a nightmare.
Failure of such tests are unreproducible by nature and bug reports
about the tests are almost useless because of them being unreproducible.


Please consider modifying the source so that those "random parameters"
are always the same.

For example, in C, random() always gives the same random numbers, so
that whatever thing which is done with those numbers is reproducible.

If you still want some randomness, use a seed which is based on the
version number, so that every time lucene2 version 2.9.4+ds1-5 is
built, the same parameters are used for the tests.


OTOH, if the tests are not to be trusted (and statistics tell us that
they should not be), please consider disabling the tests completely,
or at least the failing ones.

Thanks.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: lucene2_2.9.4+ds1-5_amd64-20160812-2024.gz
Type: application/gzip
Size: 34411 bytes
Desc: 
URL: <http://lists.alioth.debian.org/pipermail/pkg-java-maintainers/attachments/20160817/a9c42af9/attachment-0001.bin>