-
Notifications
You must be signed in to change notification settings - Fork 26
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
ImportError: failed to map segment from shared object #610
Comments
Strange stuff, version of pari, cypari2, singular and flint. |
|
That's a very weird message too. What does |
On Prefix the failure is with a different lib
Upgrading to |
Wondering if this is something like this https://stackoverflow.com/questions/58975839/dopen-failed-to-map-segment-from-shared-object-when-not-running-as-root in which case changing ulimit may do something. The settings for that stuff sits in Interesting point that I have seen sometimes when ssh-ing in a box, some extra limits may be applied. |
What do you have for
I changed |
I have, when ssh-ing
core file size puts limits on the size of coredumps. Leaving it at 0 means no coredump. |
My |
On prefix I have
I'm unable to change
|
You may have to be root to change it. Otherwise, you can change things in /etc/security/limits.conf, I don't know if you need a reboot or a new login for it to be effective. |
Same failure on Prefix as above comment after upgrading to |
I am seeing this too on vanilla: it's Python 3.7 for me, though. For me
works, but omitting
and in the output I see this
|
I certainly recall being able to build and run Sage 9.2 on the box just fine. I'm rebuilding 9.2 now to see if it exhibits the same oddity. |
I see that Gentoo now provides Pari 2.13. I wonder if it makes a difference. |
Sage 9.2 built from source: deja vu, same |
pari 2.13 does not get installed due to a weird texlive error. No idea why it is trying to do something with
|
Someone else reported a problem with the documentation build in pari a while ago, but neither Francois nor I could reproduce it. If you ever figure out what's going wrong with |
pdftex is just broken on my system for no good reason
|
it all works if I use xetex and xelatex instead. |
basically you can patch the doc/Makefile in pari to use xe(la)tex and be done with. |
https://wiki.gentoo.org/wiki/TeX_Live#eselect_pdftex makes me wonder whether special steps are needed to enable/disable pdftex |
No, that page is just outdated... eselect-pdftex was removed back in August 2020, and was obsolete before then. There are various hacks floating around to fix fmt issues, but I don't really understand this stuff. Regardless.. if you haven't done anything to break your latex installation, pdftex should just work. (I don't think pari's documentation requires xetex, but I could be wrong.) Maybe @Zlogene can suggest something? |
I enabled xetex as it has better utf-8 support, I |
OK, I reinstalled |
Doctesting vanilla 9.3.beta1 using all available Gentoo components as
gives the failures
The failures in |
upgrade to Pari 2.13 does not change the picture. |
in the 1st coredump in src/sage/doctest/test.py I see openblas, so it could be the Gentoo openblas upgrade to 3.12 (Sage is still on 3.9) that is responsible for this mess. |
On gentoo prefix, where I have symbols compiled in, I have
So maybe openblas? |
In Gentoo prefix this
now passes with |
I can confirm that going back to openblas-0.3.9-r1 in my "vanilla" case fixes the problem. |
OK, so openblas has historically been playing with cgroups for a while to deal with resource issues. May be something has been switched on. Do both of you, have openmp enabled for openblas? |
The documentation issue with pari seems to be a migration problem between versions of texlive. Some rebuilding may need to be enforced. |
My openblas
|
OK, that may cause some rebuild. Can I ask you guys to try |
this seems to be platform-dependent; on Homebrew one can use its OpenBLAS 3.12 with openmp in Sage just fine. |
But did you try with 0.3.12 with pthread instead of openmp? |
All tests passed for
in prefix with |
There may be something subtle at play because of ldso switching. Or something in Gentoo patch set. That would explain that homebrew does work. Although I guess homebrew could have a critical patch :) |
https://github.com/Homebrew/homebrew-core/blob/master/Formula/openblas.rb no patches as far as I can see |
So, maybe there is a hardware/platform component. I believe if the |
On Mon, 23 Nov 2020 12:46:35 -0800 Dima Pasechnik ***@***.***> wrote:
https://github.com/Homebrew/homebrew-core/blob/master/Formula/openblas.rb
no patches as far as I can see
Do you have `libgomp` linked into your build of `openblas` on homebrew?
|
yes. It comes from gcc 10, by the way. I wonder whether your Gentoo tries come with gcc 10, or a "normal" gcc 8 from the system? |
I am all on gcc 10 personally. I usually upgrade early to figure problems out. |
On Gentoo, I'm using |
I've never seen this before. It's present with vanilla and also on Prefix. The failure on Prefix is basically the same but with a different library that can't be imported. The following is on Gentoo.
But the following works:
The failure on Prefix does not exhibit the
PARI
issue, but adding the--memlimit=0
does allow the doctest to pass.The text was updated successfully, but these errors were encountered: