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
Porting to "El Capitan" (OS X 10.11) #19370
Comments
comment:1
Some of the bits included in the patch are clean up:
|
Branch pushed to git repo; I updated commit sha1. New commits:
|
comment:4
Building with |
comment:5
Replying to @kiwifb:
And same thing with |
comment:7
Do you mean |
comment:8
I thought the important bit was for |
comment:9
things get out of hands with
and
Unfortunately libomalloc is not linked into sage as far as I can see. |
comment:10
Run the command On OS X 10.10 with |
comment:11
OK retrying with |
comment:12
For the record, the symbol is indeed in libsingular on OSX10.10:
|
comment:13
Thanks for the info Volker, I will have to check if it is, in a non-debug build of singular. |
comment:14
PS: The libsingular compile log (logs/pkgs/singular-3.1.7p1.p0.log) shows that it links against omalloc. Maybe double check that its not the Singular build system's fault
|
comment:15
PPS: let me know if you want an account on the OSX 10.10 buildbot to compare with... |
comment:16
I'll think about the offer but I may be too split for time to run things on the two fronts at the same time. I will definitely check what happens here regarding I really need to get a debug build in shape if I want to track that segfault. |
comment:17
OK I think the But debugging doesn't come easy
anyone has a clue? |
comment:18
Thats normal, a self-compiled gdb lacks permissions on OSX to debug. Run gdb as root or figure out how to properly codesign it (the latter didn't work for me when I tried once it but is supposed to work). |
comment:19
feeling really silly I should know that. May have to cut some optimization and avoid stripping.
at least we are starting to get somewhere. |
comment:20
Looks like it uses the system clang _Unwind_RaiseException, whereas it should be using the one from SAGE_ROOT/local/lib/libgcc |
comment:21
Yes I can find the symbol in |
comment:22
I just got El Capitan running on a virtual machine! To celebrate, I tried a sage-build based on this ticket's branch. I get segfaults in zn_poly and maxima (among other things). Both seem to be alignment-problems; here is the backtrace for maxima, running ecl:
I think there is a chance these might be misconfigurations as a consequence of the virtualisation, but I thought I'd share this anyway... |
comment:24
Replying to @cnassau: With SAGE_CHECK=yes I get two test failures (segfaults from t-divrem_1, t-fat) from gmp. These seem to be similar to the ones described here for gmp-6.0.0 on Yosemite: https://gmplib.org/list-archives/gmp-bugs/2014-June/003499.html Debugging this with gdb seems difficult; all I manage to get (from running test t-divrem_1) is
|
comment:25
Replying to @cnassau:
Well I had to build |
comment:26
I think we are hit by this incredibly trivial (and incredibly powerful) bug in the gmp build-system: https://lists.macosforge.org/pipermail/macports-tickets/2014-September/170442.html The problem is that OS 10.10 and OS 10.11 are erroneously recognized as OSX 10.1 (!) Among other things this leads to a corrupted gcc which is hit by internal compiler errors when optimization is turned on. I have added a (somewhat hasty) patch for gmp which at least fixes the two test failures t-divrem_1 and t-fat. This patch is on the trac branch u/cnassau/capitan . The gmp test-suite still fails, but now at a later stage (cxx). |
comment:27
That bug is old. It can be worked around by setting MACOSX_DEPLOYMENT_TARGET=10.9, see src/bin/sage-env. The latter file almost certainly needs to be updated to also set it for 10.11 |
comment:28
Indeed, with the obvious "fix" in sage-env I could now build all of sage and start running the doctests. My 40G harddrive overflowed during the build because I kept the full installation logs (sigh), and I now get plenty of doctest failures, apparently due to missing packages (sympy, conway_polynomials, ...). Will have to rebuild, it seems... |
comment:36
Replying to @kiwifb:
Andrew's last changes for I have a couple of issues on my mind that I think should be addressed. Pls feel free to convince me that these are non-issues and/or better addressed in a different ticket:
This second issue is not fixable by changing compilation options, I believe, so we should just forbid relocation. I think something like this is required:
|
comment:37
Thanks for your comments Christian. I believe they raise very important point. In this ticket I focused on getting sage to build, but we may need to emphasise that it is not currently relocatable at the end of the build process somewhere. There may be ways to fix that situation in the future but unless someone builds a special tool the task is complicated. The suggestion of building in a standard location to produce redistributable binaries is interesting. Some optional packages may have similar issues with But I think we have an important milestone, Issues about producing binaries, optional package and so on can be put in a follow up ticket, let people be able to build the thing. |
comment:38
No problem. Could you maybe create a new ticket for the follow up issues? I'd be happy to set this ticket back to positive-review then. |
comment:39
The branch didn't work for me:
I may have done something wrong however. |
comment:40
Replying to @sagetrac-yomcat:
If this also fails it might help to post the full build log somewhere. |
comment:41
Replying to @cnassau: I redid the build with a |
comment:42
Replying to @sagetrac-yomcat:
Wonderful! I'll give this ticket its positive review back then. |
Changed reviewer from François Bissey, Andrew Ohana to François Bissey, Andrew Ohana, Christian Nassau |
comment:43
Replying to @cnassau:
I have now created #19410 for this. |
comment:44
We could support "upgrading" to |
This comment has been minimized.
This comment has been minimized.
comment:45
With this branch on OS X 10.11, Cython failed its test suite for me. See [log file http://www.math.washington.edu/~palmieri/Sage/cython-0.23.1.p0.log]. Should we set this back to "needs work"? (I also had a failure with gf2x, but I think that was an instance of #18882. It passed its test suite when I ran it a second time.) |
comment:46
I am quite unwilling to put it back to work personally but this needs investigating. Would upgrading |
comment:47
Just do it on a separate ticket. |
comment:48
I see the same problem on OS X 10.10, so let's keep this as "positive review". |
comment:49
So would you feel that #16044 is now invalid/fixed? Certainly that would seem to be the case, though the relocation issue isn't fixed, I guess. Anyway I have now finally given up on building anything after Sage 6.4.1 on PPC Macs because of a few different issues, not least of which is the continued upward spiral of gcc versions. (Naturally one definitely can build it! But not so easily from scratch, you must first build a newer gcc to build the newest gcc, and that's not automatic enough for me to bother with it.) |
comment:50
I suggest we mark it won't fix. I have no machine and no intentions to test this on Tiger. But yes in principle it would solve the issues. Funny that I was kind of force to push a branch to get sage to build on a new version of OS X with fix that were spotted that far back (and for which there was solutions that far back too). |
Changed branch from u/ohanar/elcapitan to |
Changed commit from |
Changed author from François Bissey, Andrew Ohana to François Bissey, R. Andrew Ohana |
comment:52
for uniqueness of names |
Currently sage is not working at all on OS X 10.11. The aim of this ticket is to remedy to that.
Points identified so far:
install_name
fail to loadR
) should have been applied more widely on future releasegit
withoutopenssl
on OS X in Update to git 2.1.2 #17091 is not working anymore and needs replacing with something less subtleWill all these problems fix
sage
can build an run on OS X 10.11. This ticket is only concerned with buildingsage
, issues about binary distribution and checking of optional packages is left for other tickets.CC: @jhpalmieri @sagetrac-yomcat @cnassau
Component: porting
Author: François Bissey, R. Andrew Ohana
Branch:
d3f5e99
Reviewer: François Bissey, Andrew Ohana, Christian Nassau
Issue created by migration from https://trac.sagemath.org/ticket/19370
The text was updated successfully, but these errors were encountered: