ModelField Segfaults for Some Calls

[ddasilva]

Hi Matt,

I've been using this library for a number of years and it's great. I recently released my own package rbinvariantslib, which uses PyGeopack.

Recently I started getting segfaults for calls which I think should be simple and (if I'm not crazy) worked before. This snippet below makes a regular grid of spacing 0.1 in xyz between -10 and 10, and evalutes TS05 on it. It segfaults at the last call. Unusually, when the grid spacing is 0.5, it does not segfault.

Can you take a look? If there's any workaround before the bug is fixed please let me know.

Best,
Daniel

import os

# Don't print warnings from pygeopack about data not found
os.environ['GEOPACK_NOWARN'] = '1'
import PyGeopack as gp

from rbinvariantslib import models, invariants
from datetime import datetime
import numpy as np

# Setup time
time = datetime(2015, 10, 2)
gp_date = int(time.strftime("%Y%m%d"))
gp_ut = int(time.strftime("%H")) + time.minute / 60

# Setup params
params = {'Pdyn': 2.609999974394062, 'SymH': -14.0, 'By': 10.9399995803833, 'Bz': 1.7699999809265137, 'W1': 0.0, 'W2': 0.0, 'W3': 0.0, 'W4': 0.0, 'W5': 0.0, 'W6': 0.0}

# Setup x, y, z inputs
axis = np.arange(-10, 10, 0.1)
x, y, z = np.meshgrid(axis, axis, axis)
x = x.flatten()
y = y.flatten()
z = z.flatten()

# Make call (this segfaults)
Bx, By, Bz = gp.ModelField(
     x,
     y,
     z,
     Date=gp_date,
     ut=gp_ut,
     Model='TS05',
     CoordIn='SM',
     CoordOut='SM',
     **params
)

[ddasilva]

It turns out that if I call gp.ModelField() in smaller batches (e.g., 10,000 elements at a time) the segfault doesn't occur and there is no major difference in speed. I will use this workaround in my library.

[mattkjames7]

Hi Daniel,
Sorry it has taken me so long to look into this.
Judging by the code which you have sent in the first message where you attempt to get 8,000,000 vectors at a time and the fact that it works fine in smaller batches says to me that there might be a stack overflow causing the segmentation fault. I'll have to check the code to be certain, but it's possible that I need to change they way some of the variables are defined - i.e. by allocating them in the heap instead of the stack.
I'll try to have a look into it over the next week or so...
Cheers,
Matt.

[ddasilva]

Thanks @mattkjames7 . No rush because I have my workaround!