diff --git a/test/helpers/precompile.jl b/test/helpers/precompile.jl
index daa1b57b..54fc58a9 100644
--- a/test/helpers/precompile.jl
+++ b/test/helpers/precompile.jl
@@ -11,20 +11,10 @@ function precompile_test_harness(@nospecialize(f), separate::Bool)
         pushfirst!(DEPOT_PATH, load_cache_path)
         f(load_path)
     finally
-        try
-            rm(load_path, force=true, recursive=true)
-        catch err
-            @show err
-        end
-        if separate
-            try
-                rm(load_cache_path, force=true, recursive=true)
-            catch err
-                @show err
-            end
-        end
-        filter!((≠)(load_path), LOAD_PATH)
-        separate && filter!((≠)(load_cache_path), DEPOT_PATH)
+        popfirst!(DEPOT_PATH)
+        popfirst!(LOAD_PATH)
+        rm(load_path, force=true, recursive=true)
+        separate && rm(load_cache_path, force=true, recursive=true)
     end
     nothing
 end
diff --git a/test/native/precompile.jl b/test/native/precompile.jl
index e9983a7e..c2648701 100644
--- a/test/native/precompile.jl
+++ b/test/native/precompile.jl
@@ -2,10 +2,11 @@
 
 precompile_test_harness("Inference caching") do load_path
     # Write out the Native test setup as a micro package
-    create_standalone(load_path, "TestCompiler", "native.jl")
+    create_standalone(load_path, "NativeCompiler", "native.jl")
 
-    write(joinpath(load_path, "InferenceCaching.jl"), :(module InferenceCaching
-        import TestCompiler
+    write(joinpath(load_path, "NativeBackend.jl"), :(
+        module NativeBackend
+        import NativeCompiler
         using PrecompileTools
 
         function kernel(A, x)
@@ -14,36 +15,36 @@ precompile_test_harness("Inference caching") do load_path
         end
 
         let
-            job, _ = TestCompiler.Native.create_job(kernel, (Vector{Int}, Int))
+            job, _ = NativeCompiler.Native.create_job(kernel, (Vector{Int}, Int))
             precompile(job)
         end
 
         # identity is foreign
         @setup_workload begin
-            job, _ = TestCompiler.Native.create_job(identity, (Int,))
+            job, _ = NativeCompiler.Native.create_job(identity, (Int,))
             @compile_workload begin
                 precompile(job)
             end
         end
     end) |> string)
 
-    Base.compilecache(Base.PkgId("InferenceCaching"))
+    Base.compilecache(Base.PkgId("NativeBackend"))
     @eval let
-        import TestCompiler
+        import NativeCompiler
 
         # Check that no cached entry is present
         identity_mi = GPUCompiler.methodinstance(typeof(identity), Tuple{Int})
 
         token = let
-            job, _ = TestCompiler.Native.create_job(identity, (Int,))
+            job, _ = NativeCompiler.Native.create_job(identity, (Int,))
             GPUCompiler.ci_cache_token(job)
         end
         @test !check_presence(identity_mi, token)
 
-        using InferenceCaching
+        using NativeBackend
 
         # Check that kernel survived
-        kernel_mi = GPUCompiler.methodinstance(typeof(InferenceCaching.kernel), Tuple{Vector{Int}, Int})
+        kernel_mi = GPUCompiler.methodinstance(typeof(NativeBackend.kernel), Tuple{Vector{Int}, Int})
         @test check_presence(kernel_mi, token)
 
         # check that identity survived
@@ -55,7 +56,7 @@ precompile_test_harness("Inference caching") do load_path
         GPUCompiler.enable_disk_cache!()
         @test GPUCompiler.disk_cache_enabled() == true
 
-        job, _ = TestCompiler.Native.create_job(InferenceCaching.kernel, (Vector{Int}, Int))
+        job, _ = NativeCompiler.Native.create_job(NativeBackend.kernel, (Vector{Int}, Int))
         @assert job.source == kernel_mi
         ci = GPUCompiler.ci_cache_lookup(GPUCompiler.ci_cache(job), job.source, job.world, job.world)
         @assert ci !== nothing
@@ -63,7 +64,7 @@ precompile_test_harness("Inference caching") do load_path
         path = GPUCompiler.cache_file(ci, job.config)
         @test path !== nothing
         @test !ispath(path)
-        TestCompiler.Native.cached_execution(InferenceCaching.kernel, (Vector{Int}, Int))
+        NativeCompiler.Native.cached_execution(NativeBackend.kernel, (Vector{Int}, Int))
         @test ispath(path)
         GPUCompiler.clear_disk_cache!()
         @test !ispath(path)
diff --git a/test/ptx/precompile.jl b/test/ptx/precompile.jl
index 86441c2c..b5f980c9 100644
--- a/test/ptx/precompile.jl
+++ b/test/ptx/precompile.jl
@@ -1,9 +1,10 @@
 precompile_test_harness("Inference caching") do load_path
     # Write out the PTX test helpers as a micro package
-    create_standalone(load_path, "TestCompiler", "ptx.jl")
+    create_standalone(load_path, "PTXCompiler", "ptx.jl")
 
-    write(joinpath(load_path, "InferenceCaching.jl"), :(module InferenceCaching
-        import TestCompiler
+    write(joinpath(load_path, "PTXBackend.jl"), :(
+        module PTXBackend
+        import PTXCompiler
         using PrecompileTools
 
         function kernel()
@@ -11,22 +12,22 @@ precompile_test_harness("Inference caching") do load_path
         end
 
         let
-            job, _ = TestCompiler.PTX.create_job(kernel, ())
+            job, _ = PTXCompiler.PTX.create_job(kernel, ())
             precompile(job)
         end
 
         # identity is foreign
         @setup_workload begin
-            job, _ = TestCompiler.PTX.create_job(identity, (Int,))
+            job, _ = PTXCompiler.PTX.create_job(identity, (Int,))
             @compile_workload begin
                 precompile(job)
             end
         end
     end) |> string)
 
-    Base.compilecache(Base.PkgId("InferenceCaching"))
+    Base.compilecache(Base.PkgId("PTXBackend"))
     @eval let
-        import TestCompiler
+        import PTXCompiler
 
         # Check that no cached entry is present
         identity_mi = GPUCompiler.methodinstance(typeof(identity), Tuple{Int})
@@ -41,10 +42,10 @@ precompile_test_harness("Inference caching") do load_path
             ci = isdefined(ci, :next) ? ci.next : nothing
         end
 
-        using InferenceCaching
+        using PTXBackend
 
         # Check that kernel survived
-        kernel_mi = GPUCompiler.methodinstance(typeof(InferenceCaching.kernel), Tuple{})
+        kernel_mi = GPUCompiler.methodinstance(typeof(PTXBackend.kernel), Tuple{})
         @test check_presence(kernel_mi, token)
 
         # check that identity survived