Re: [igraph] make check fails on 20 tests (0.6.4 and 0.6.5)

[Gábor Csárdi]

Hi,

On Tue, Feb 4, 2014 at 11:20 AM, Dimitris Diochnos <address@hidden> wrote:

[...]

- Re-making the library in one of the versions 0.6.4/5 does allow me
to rerun successfully programs that I had written in the past and had
dependencies on igraph (I believe it was 0.6.3, or something like that
back then). However, after 'make' if one tries 'make check' in both
0.6.4 and 0.6.5 there are 20 tests where the library fails to pass.

This should not happen, at least surely not for 20 tests. I suspect it is a configuration issue on your part (e.g. warning messages of the compiler make them fail), or you are using an exotic platform. 

But if you don't tell us which ones fail and how, we cannot help you. Essentially we need the contents of tests/testsuite.dir/*/test* after running the tests. We also need information about your platform.

- On the other hand, the 0.7.0 library does pass on the same machine
the 'make check' part (0 failures), but at least one critical program
that I have ends up with runtime execution problems. For the same code
as in 0.6.4/5 I get warnings this time that I am using the
igraph_adjlist_get(&list, node) which now returns igraph_vector_int_t
* and not igraph_vector_t * as it used to return in the past (and in
the documentation online). I thought this was only typecasting but
apparently there has to be a more fundamental change in the code and
the way the adjacency lists are computed and extracted. Moreover, I
also checked the output of the following:
sizeof(      igraph_vector_t *    ) =  8
sizeof(const igraph_vector_t *    ) =  8
sizeof(      igraph_vector_int_t *) =  8
sizeof(const igraph_vector_int_t *) =  8
under 0.6.4/5 and 0.7.0 and in every case I get the answer above, so,
from a first glance this does not seem to be problem of alignment or a
new data type either. Long story sort, do you believe that the
igraph_adjlist_get(...) behaves correctly, or should I try to come up
with a minimal example where the output is not what it is expected? I
am pretty sure the work in 0.7.0 is unfinished though.

igraph 0.7.0 is pretty much finished, but it is not backwards compatible with 0.6.x, unfortunately.

Closing, I am also worried that 0.6.5 does not pass through 'make
check' some tests on cliques. I am not focusing on cliques right now,
but I would like to make sure that problems like this one from the
past (https://github.com/igraph/igraph/issues/123) has been resolved,
even though the library fails to pass 2 of the 3 make check tests for
cliques. So, is the most stable library currently online 0.6.5?

Issue 123 was fixed according to the linked page, but you can try it yourself, the bug report has a data set for which igraph failed back then. As for the failing test cases, again, if you don't show us what happens, what can we do?

Gabor

[...]