Hi,
Admittedly, this is a rather heavy-handed approach, but changing epsilon
for the Collinear() calls to 1e-3 instead of 1e-4 lets the builds complete
successfully on all architectures, including i386.
This appears to be as being related to
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=323 , in other words, because
the CrossProduct as it is being calculated is probably having values
rounded multiple times.
Attached is a patch that has been applied on Ubuntu to change the epsilon
value used for the Line2 tests.
I think this warrants being brought up to the upstream too, perhaps they
have a preferred way of dealing with this.
Kindly,
Mathieu Trudel-Lapierre <mathieu...@ubuntu.com>
Freenode: cyphermox, Jabber: mathieu...@gmail.com
4096R/EE018C93 1967 8F7D 03A1 8F38 732E FF82 C126 33E1 EE01 8C93
From: Mathieu Trudel-Lapierre <mathieu.trudel-lapie...@canonical.com>
Subject: Allow for a bit more rounding when verifying collinearity
---
src/Line2_TEST.cc | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
Index: b/src/Line2_TEST.cc
===================================================================
--- a/src/Line2_TEST.cc
+++ b/src/Line2_TEST.cc
@@ -147,7 +147,7 @@ TEST(Line2Test, CollinearPoint)
pt.Set(0, 0.00001);
EXPECT_FALSE(lineA.Collinear(pt));
- EXPECT_TRUE(lineA.Collinear(pt, 1e-4));
+ EXPECT_TRUE(lineA.Collinear(pt, 1e-3));
{
math::Line2d ptLine(pt, pt);
EXPECT_FALSE(lineA.Collinear(ptLine));
@@ -159,7 +159,7 @@ TEST(Line2Test, CollinearPoint)
pt.Set(0, -0.00001);
EXPECT_FALSE(lineA.Collinear(pt));
- EXPECT_TRUE(lineA.Collinear(pt, 1e-4));
+ EXPECT_TRUE(lineA.Collinear(pt, 1e-3));
{
math::Line2d ptLine(pt, pt);
EXPECT_FALSE(lineA.Collinear(ptLine));
