From d9330988e6f23f929e99c30eb597e69a6f67dd5e Mon Sep 17 00:00:00 2001
From: "Steven G. Johnson" <stevenj@alum.mit.edu>
Date: Tue, 16 May 2023 14:03:13 -0400
Subject: [PATCH] update CI badge (#15)

* update CI badge

* fix test failure on Julia 1.3
---
 .travis.yml      | 10 ----------
 README.md        |  4 ++--
 test/runtests.jl |  4 ++--
 3 files changed, 4 insertions(+), 14 deletions(-)
 delete mode 100644 .travis.yml

diff --git a/.travis.yml b/.travis.yml
deleted file mode 100644
index 94d99ba..0000000
--- a/.travis.yml
+++ /dev/null
@@ -1,10 +0,0 @@
-language: julia
-os:
-    - linux
-julia:
-    - 1.3
-    - 1.4
-    - 1.5
-    - nightly
-notifications:
-  email: false
diff --git a/README.md b/README.md
index 51959d7..93b9414 100644
--- a/README.md
+++ b/README.md
@@ -1,6 +1,6 @@
 # FastChebInterp
 
-[![Build Status](https://travis-ci.org/stevengj/FastChebInterp.jl.svg?branch=master)](https://travis-ci.org/stevengj/FastChebInterp.jl)
+[![CI](https://github.com/JuliaMath/FastChebInterp.jl/workflows/CI/badge.svg)](https://github.com/JuliaMath/FastChebInterp.jl/actions?query=workflow%3ACI)
 
 Fast multidimensional Chebyshev interpolation on a hypercube (Cartesian-product)
 domain, using a separable (tensor-product) grid of Chebyshev interpolation points, as well as Chebyshev regression (least-square fits) from an arbitrary set of points.   In both cases we support arbitrary dimensionality, complex and vector-valued functions, and fast derivative and Jacobian computation.
@@ -63,7 +63,7 @@ julia> cos(2x + 3cos(4x)) * (2 - 12sin(4x)) # exact derivative
 ```
 
 Interpolation is most efficient and accurate if we evaluate our function at the points given by `chebpoints`.   However, we can also perform least-square polynomial fitting (in the Chebyshev basis, which is well behaved even at high degree) from an *arbitrary* set of points — this is useful if the points were specified externally, or if we want to "smooth" the data by fitting to a polynomial of lower degree than for interpolation.    For example, we can fit the same function above, again to a degree-200 Chebyshev polynomial, using 10000 *random* points in the domain:
-```jl 
+```jl
 xr = rand(10000) * 10 # 10000 uniform random points in [0, 10]
 c = chebregression(xr, f.(xr), 0, 10, 200) # fit to a degree-200 polynomial
 ```jl
diff --git a/test/runtests.jl b/test/runtests.jl
index 3e64b99..6d96b74 100644
--- a/test/runtests.jl
+++ b/test/runtests.jl
@@ -37,7 +37,7 @@ end
         interp0 = chebinterp(f.(x), lb, ub, tol=0)
         @test repr("text/plain", interp0) == "ChebPoly{2,$T,$T} order (48, 39) polynomial on [-0.3,0.9] × [0.1,1.2]"
         @test ndims(interp) == 2
-        x1 = T[0.2, 0.3]
+        x1 = T[0.2, 0.8] # not too close to boundary or test_frule can fail by taking a big FD step
         @test interp(x1) ≈ f(x1)
         @test interp(x1) ≈ interp0(x1) rtol=10eps(T)
         @test all(n -> n[1] < n[2], zip(size(interp.coefs), size(interp0.coefs)))
@@ -148,4 +148,4 @@ end
     @test interp([0,0]) ≈ 3*cos(0.55)
     @test interp([0.1,0.22]) ≈ 3.2*cos(0.55)
     @test chebgradient(interp, [0.1,0.2])[2] == [2*cos(0.55), 0]
-end
\ No newline at end of file
+end