Request for a new release #394
Comments
Yes! @sylvestre: what do you say? |
|
sure, do you want to do it @fghoussen ? :) |
OK, I'll try to have a look by the week-end |
|
@sylvestre: release tag pushed on opencollab/master |
|
@fghoussen seems that we have some changes in this release that we should probably fix usr/include/arpack-ng/ when it was installed in usr/include/arpack/ before |
|
Honestly, i don't think mingw64 should be the cause of significant changes... I think we need to revert these changes asap and publish a new release. |
|
the different flavors should be done behind an option as it isn't a common case compared to the rest |
|
Let's see if #398 is enough to fix this |
|
@sylvestre: did you release 3.9.1? |
|
nope, I thought you were going to :) Should I ? |
Yes! :) |
|
Are you aware of any new issues with 3.9.0? I am currently trying to debug the following: igraph/igraph#2311 |
Do you reproduce the same issue with arpack-3.8.0 or lower? Using same version of compiler /sanitizers ? The only issue (which should not be an issue to me) I could think of would be #80 / JuliaLinearAlgebra/Arpack.jl#138. Not sure however, it's connected to your issue. In 3.9.0, the only known issue was #398 (fix merged in master) |
According to your stack (using sanitizers), I guess you use "your" hand-made eigen solve: did you try using https://github.com/opencollab/arpack-ng#eigen-support or https://github.com/opencollab/arpack-ng#iso_c_binding-support? |
This kind of thing (https://github.com/igraph/igraph/blob/d6796015974f66bcf0e3711593c64c55ff81fdef/src/linalg/arpack.c#L1065) always hide evil you don"t want to deal with thanks to ISO_C_BINDING |
|
I'm not sure if using ISO_C_BINDING is feasible for us, as igraph also bundles an older f2c'd version of ARPACK. This is important for our target audience, who are typically researchers inexperienced with programming. However, I finally managed to reproduce the issue completely independently of igraph, using only Octave. The problem is present only with 3.9.0, but not with 3.8.0. Could you please take another look, @fghoussen ? Note that I amended some of the earlier comments I made: igraph/igraph#2311 |
|
Does octave uses ICB? If igraph and octave does not use ICB, there is yet a possibility that both suffer from bad-ugly-old way to link from F77 to C using ugly tricks (link to "underscored" function symbols): then, compilers can not check anything but pass data blindly basically as (void*)... This is risky and hell to debug.
Using ICB should be seamless:
|
|
@fghoussen is it time to take 3.9.1 ? thanks |
|
Sure! Go for it :) |
|
It would be nice to reach a conclusion for #401. I hope that a minimal example written in C using ICB (coming soon) will be sufficient for you to continue investigating. |
|
It feels like we are not talking to each other, but in parallel. You have not tested the specific scenario I described, always ignoring critical details. You have claimed misuse several times, without pointing out any specific instance of misuse. As I said, a minimal example in C is incoming. I hope that we can use that as a baseline. |
@sylvestre: 3.9.1 does not show up here https://github.com/opencollab/arpack-ng/tags so we can't make a release entry here https://github.com/opencollab/arpack-ng/releases. #428 Is this normal? |
|
i don't think we tagged it, just push or create the tag :) |
Uuuh... Do you have/remind SHA1 of 3.9.1?... I don't!... |
|
just recreate it, we never released 3.9.1 afaik |
|
Can we have a 3.9.1 soon please? I'd like to have the fixes to the issue that affected igraph. There can always be a 3.9.2 that includes the improvements for which there are open PRs at the moment. |
|
@fghoussen would you like me to do it if you don't have time ? |
Sure. Currently between 2 confs/sites, not back home before again another week. Can't help before returning. |
|
Can we have this issue re-opened to make sure it won't get forgotten? |
|
Any update on a new release? |
|
Can't handle that at the time. @sylvestre: can you? |
|
done, sorry it took forever |
|
Note that we would be happy to have more help with arpack, review PR, preparing releases, answering to bug reports, etc I have 0 usage of it. I continue maintaining it because a lot of software rely on it. So, if you are a company / university relying on it, don't hesitate to contribute / help cc @SciTonio ;) |
Hi there, to finalize the merge of microsoft/vcpkg#29248, it would be better to have a new release so that the vcpkg port can follow readily the versioning scheme of the repo. Is it something possible ? Otherwise I would need to backport as patch all the recent work to make it work on windows too. Feasible but cumbersome (and anyway temporary)
The text was updated successfully, but these errors were encountered: