fds icon indicating copy to clipboard operation
fds copied to clipboard

GitHub Actions for OSX build not working

Open mcgratta opened this issue 2 years ago • 7 comments

I am getting email warnings like this when making code changes.

mcgratta avatar Jan 14 '24 18:01 mcgratta

This may be a hiccup with GitHub. Looks like the commit sent to firemodels passed all checks fine, no? Seems your fork is up to date with firemodels (not ahead of it), so does not seem to be anything wrong with the code. You can rerun your Actions, like this:

Screenshot 2024-01-14 at 3 15 51 PM

rmcdermo avatar Jan 14 '24 20:01 rmcdermo

I'm seeing something odd with the code. With the ompi_gnu_osx_db executable, the code does not start if I have an input file. If I do not have an input file, the code starts (just giving the initial err file header). If I run compile and run _dv, it works to run with an input file. So, something is amiss. This looks like it is not just your recent commits, as I dialed back a couple of commits and see the same.

rmcdermo@PN122745 Junk (master) $ fds_db impinging_jet_Re_1e5_coarse.fds 

Program received signal SIGILL: Illegal instruction.

Backtrace for this error:

Could not print backtrace: executable file is not an executable

Could not print backtrace: executable file is not an executable
#0  0x1036cdaf7
#1  0x1036ccb43
#2  0x183d69a23
#3  0x1ed3d950b
#4  0x1ed3dd7eb
#5  0x18de97e4f
#6  0x18de97caf
#7  0x183b8890f
#8  0x183b97cc3
#9  0x18dfe5d93
#10  0x18dec18b3
#11  0x183b8890f
#12  0x183b8a14b
#13  0x18de97a23
#14  0x102e7a117
#15  0x102e7f9a3
#16  0x1ec4323af
Illegal instruction: 4

rmcdermo avatar Jan 14 '24 20:01 rmcdermo

When you run without an input file, FDS does not initiate MPI. It just writes version info and then exits. It's been this way for some time. There might be an issue with running fds from the command line. I don't know if the MPI initialization will be affected.

mcgratta avatar Jan 14 '24 20:01 mcgratta

_dv and release versions run fine form the command line. So, mpi may be involved, but there must be another compiler option that is at play here.

rmcdermo avatar Jan 14 '24 20:01 rmcdermo

We are running into something related to the floating-point traps like in this example, but I have not figured out the best work around.

rmcdermo avatar Jan 14 '24 21:01 rmcdermo

Related to what you were saying, the program dies in MPI_INIT_THREAD because I am not launching it with mpirun. It I use mpirun -n 1 <full path to exe> <input file> then it works fine. We need to be able to run these cases in serial.

rmcdermo avatar Jan 14 '24 21:01 rmcdermo

I am getting email warnings like this when making code changes.

As Randy commented, it seems the github server did not load the virtual machine in time to run the action. If anyone gets this actions failure please report. Lets see if it was random or related to the osx target somehow.

Screenshot from 2024-01-15 10-39-28

marcosvanella avatar Jan 15 '24 15:01 marcosvanella