From 5e3f5cb6a950b1cd46fe1cf172656b1bbe44a2e7 Mon Sep 17 00:00:00 2001 From: hinder Date: Wed, 30 Apr 2014 15:36:08 +0000 Subject: test.ccl: Do not require a very accurate convergence order The convergence order is extremely sensitive to roundoff, as its computation includes subtracting two very similar numbers. Instead, we output the integration results for the purpose of regression testing, as these are less sensitive to roundoff. git-svn-id: http://svn.einsteintoolkit.org/cactus/EinsteinAnalysis/Multipole/trunk@112 4f5cb9a8-4dd8-4c2d-9bbd-173fa4467843 --- test/test.ccl | 6 ++++++ 1 file changed, 6 insertions(+) (limited to 'test') diff --git a/test/test.ccl b/test/test.ccl index 6ccaab2..a53d4e6 100644 --- a/test/test.ccl +++ b/test/test.ccl @@ -1,2 +1,8 @@ ABSTOL 1e-8 RELTOL 1e-8 + +# The convergence order tests are very sensitive to roundoff, because +# the integration has a very small relative error (1e-7). Hence +# subtracting the numerical result from the exact result loses ~7 +# digits of precision. +ABSTOL 1e-3 ^test_.*_convergence_order\.asc$ -- cgit v1.2.3