I remember this test being fragile because it relies on output from a third-party application under a large amount of development and makes assumptions based on a hardcoded value which cannot be determined programmatically. This is something that I tried getting more context in the past, but the ball was dropped with this item. I very much understand how this is broken though, and I will try to think of a better way to check for this (the right solution might involve writing a C program which emulates a stripped down version of readelf, etc).

I don't think we should move forward with this change. In particular, your proposed approach will be hard to diagnose after the fact as well if the number of columns output or other formatting changes: the test will fail with difficult to understand diagnostic messages (in particular if the field is non-numeric); you can't test readelf to make sure it's executing cleanly; etc.