Skip to content

Update igraph from 0.6.4 to 0.7.1 #1146

Merged
merged 4 commits into from
Jun 27, 2019
Merged

Conversation

thomas
Copy link
Collaborator

@thomas thomas commented Jun 26, 2019

No description provided.

@pmenzel
Copy link
Collaborator

pmenzel commented Jun 26, 2019

What are the default CFLAGS?

Why not keep the shared library? We do not know, if some user linked against it in a personal project?

@thomas
Copy link
Collaborator Author

thomas commented Jun 26, 2019

The default CFLAGS are empty -> ''. The rest is set inside the Makefile.
I expect that noone is still using it. If someone does, its better to fix the issue onsite.

@thomas
Copy link
Collaborator Author

thomas commented Jun 27, 2019

Concerning the usage of libigraph.so.0, I found one from Sep-28-2015: /package/sequencer/zzz_OBSOLETE_2018/transabyss/deps/python27/lib/python2.7/site-packages/igraph/_igraph.so.

So, from now on let the people build their (python,perl,R,...) bindings from the lib-archive.
For reference see the discussion in PR #886.

@thomas thomas merged commit 857db24 into master Jun 27, 2019
@pmenzel
Copy link
Collaborator

pmenzel commented Jun 28, 2019

For the record, please do not remove shared libraries in the future. Upstream just tests with that, and not using it makes us and our users hit corner cases, cf #1151.

@thomas
Copy link
Collaborator Author

thomas commented Jun 28, 2019

Sorry Paul, but you are just presenting a 'FMIW' solution.
After some checking I've found an error when the 'static' igraph is imported by python (both pythons, it would have been helpfull if you would have mentionend it) undefined symbol: _ZTVN10__cxxabiv117__class_type_infoEthat points to some C++ issues (N.B. That igraph is using also c++ code is a point I've overlooked when building libigraph, meaning the current build is rubbish). The next test included two builds of libigraph and testing. Whereas the version with the shared lib passed all tests, the static-only variant happily fails:

ERROR: 227 tests were run,
56 failed unexpectedly.
7 tests were skipped.

To me this reads like libigraph has a little problem (that needs checking). But it clearly doesn't mean that we now should start over with shared libs again. Right?

@pmenzel
Copy link
Collaborator

pmenzel commented Jun 28, 2019 via email

@thomas
Copy link
Collaborator Author

thomas commented Jun 28, 2019

Since the currently installed libigraph.a has issues you may go ahead (maybe even rebuild with --enable-static=no ?). Also I do not opt for build 'everything' static. But as for 'everybody else', our system is not 'everybody elses single user apt update etc system' -- different story.

@donald
Copy link
Collaborator

donald commented Jun 29, 2019

FMIW ?

@thomas
Copy link
Collaborator Author

thomas commented Jun 29, 2019

FMIW -> For Me It Works

@thomas
Copy link
Collaborator Author

thomas commented Jun 29, 2019

Funfact beside: The igraph webpage is a bit cheeky when presenting the [download c library] grin

@thomas
Copy link
Collaborator Author

thomas commented Jun 29, 2019

Before some weekend efforts start...

The 'problem' with the testsuite is resolved, rather trivial: throw in some more -lxyz options.

Sign in to join this conversation on GitHub.
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

None yet

3 participants