segmentation fault in test run after building without bundled proj sources about pyproj

[jdkloe]

I have disabled the inverse hammer projection test in the rpm build for fedora, and after this all remaining tests run fine for rawhide (upcoming fedora 24). This one will have proj version 4.9.2
For fedora 23, which has proj version 4.9.1, there are 2 failing tests related to the hammer projection (both for python2 and python3). See the details in this log file:
https://kojipkgs.fedoraproject.org//work/tasks/1374/12551374/build.log
I guess this also is related to the missing inverse hammer in these proj versions.

The output for the failing tests is:
Trying:
p = Proj(proj='hammer') # hammer proj and inverse
Expecting nothing

---

File "/builddir/build/BUILDROOT/pyproj-1.9.5.1-1.fc23.x86_64/usr/lib64/python3.4/site-packages/pyproj/init.py", line 324, in pyproj.Proj.new
Failed example:
p = Proj(proj='hammer') # hammer proj and inverse
Exception raised:
Traceback (most recent call last):
File "/usr/lib64/python3.4/doctest.py", line 1318, in run
compileflags, 1), test.globs)
File "<doctest pyproj.Proj.__new[20]>", line 1, in
p = Proj(proj='hammer') # hammer proj and inverse
File "/builddir/build/BUILDROOT/pyproj-1.9.5.1-1.fc23.x86_64/usr/lib64/python3.4/site-packages/pyproj/init.py", line 355, in new
return _proj.Proj.new(self, projstring)
File "_proj.pyx", line 84, in _proj.Proj.cinit (_proj.c:1156)
raise RuntimeError(pj_strerrno(err))
RuntimeError: b'major axis or radius = 0 or not given'
Trying:
x,y = p(-30,40)
Expecting nothing
ok
Trying:
'x=%12.3f y=%12.3f' % (x,y)
Expecting:
'x=-2711575.083 y= 4395506.619'

---

File "/builddir/build/BUILDROOT/pyproj-1.9.5.1-1.fc23.x86_64/usr/lib64/python3.4/site-packages/pyproj/init.py", line 326, in pyproj.Proj.new
Failed example:
'x=%12.3f y=%12.3f' % (x,y)
Expected:
'x=-2711575.083 y= 4395506.619'
Got:
'x=16013975.139 y=19338537.497'

[micahcochran]

One option would be to move these inverse hammer tests to unittest and skip it the inverse hammer tests when ran like so:
python unittests/test.py --skip-inv-hammer

That would allow package maintainers to be able to ensure the rest of the functionality stays in the package. It would retain the useful unittest.

[jdkloe]

Yes, that would be a nice option (if the plan is to move all tests to the unittests anyway).
A more general concern I think is that it seems possible to have undefined inverse projections in proj. I have no detailed knowledge of the proj source code, but if more projections exist without their inverse, then this may trigger the same problem. First step would be to add unittests for all projections, and their inverse I think.

[micahcochran]

It appears 21 projections currently in PROJ.4 have no inverse function.

I've used unittests to test that bug fixes remain fixed. Unittest does have the ability to skip test if certain conditions aren't right.

[jdkloe]

Yes, and I see devel version of proj has added a check for NULL pointer of P->inv and P->fwd in the source files pj_inv.c and pj_fwd.c. These checks are present in the bundled c-code but not in proj releases 4.9.1 and 4.9.2. This will cause segmentation faults if the pyproj module is build without bundles sources with these proj versions. I just tested, and for example the 'airy' projection also shows this segmentation fault.

[micahcochran]

PR #44 was merged, which fixed this issue. The repository no longer has this issue. Closing.