runtime/pprof TestConvert*Profile fails when dynamically linked

[David Benoit]

TestConvertCPUProfile and TestConvertMemProfile both fail when dynamically linked in Fedora and ubi9, because the executable segment (r-xp in /proc/self/maps) is mapped as the second section whereas with earlier versions of system libraries it is mapped as the first.

https://github.com/dbenoit17/openssl-fips/actions/runs/3146239220/jobs/5114463156#step:13:394

[David Benoit]

The following patch appears to resolve the issue, but it could potentially cause issues in different configurations. Might be a good starting point to work with:

diff --git a/src/runtime/pprof/proto_test.go b/src/runtime/pprof/proto_test.go
index 84a051a..d640250 100644
--- a/src/runtime/pprof/proto_test.go
+++ b/src/runtime/pprof/proto_test.go
@@ -95,11 +95,11 @@ func testPCs(t *testing.T) (addr1, addr2 uint64, map1, map2 *profile.Mapping) {
                        // region of memory.
                        t.Skipf("need 2 or more mappings, got %v", len(mprof.Mapping))
                }
-               addr1 = mprof.Mapping[0].Start
-               map1 = mprof.Mapping[0]
+               addr1 = mprof.Mapping[1].Start
+               map1 = mprof.Mapping[1]
                map1.BuildID, _ = elfBuildID(map1.File)
-               addr2 = mprof.Mapping[1].Start
-               map2 = mprof.Mapping[1]
+               addr2 = mprof.Mapping[2].Start
+               map2 = mprof.Mapping[2]
                map2.BuildID, _ = elfBuildID(map2.File)
        case "js":
                addr1 = uint64(abi.FuncPCABIInternal(f1))  

[Derek Parker]

I think we should skip this test in ubi9 for now as we work towards an upstream fix.

[Derek Parker]

So it looks like this was fixed upstream in the master branch via https://go-review.googlesource.com/c/go/+/424295. I suggest we carry this patch in our downstream 1.19 release branch. I'll open a PR for that.