skip checking categorization of libperl symbols for LTO builds

tonycoz · tonycoz · commit 69b4fa321bd1 · 2022-11-28T13:50:41.000+11:00
For LTO builds with gcc and clang the PL_no_mem symbol as listed by nm is flagged as "D" (writable data) and "T" (text, aka code) respectively. Looking at the final generated executable PL_no_mem does end up in the .rodata (read only data) section, so it might be worth adding a separate test for that. Fixes #20518
diff --git a/t/porting/libperl.t b/t/porting/libperl.t
@@ -9,6 +9,9 @@
 # - freebsd
 # and on other platforms, and if things seem odd, just give up (skip_all).
 #
+# Symbol types for LTO builds don't seem to match their final section, so
+# skip on LTO builds too.
+#
 # Debugging tip: nm output (this script's input) can be faked by
 # giving one command line argument for this script: it should be
 # either the filename to read, or "-" for STDIN.  You can also append
@@ -47,6 +50,15 @@ if ($Config{cc} =~ /g\+\+/) {
     skip_all "on g++";
 }
 
+# ccname is gcc for both gcc and clang
+if ($Config{ccname} eq "gcc" && $Config{ccflags} =~ /-flto\b/) {
+    # If we compile with gcc nm marks PL_no_mem as "D" (normal data) rather than a R (read only)
+    # but the symbol still ends up in the .rodata section of the image on linking.
+    # If we compile with clang 14, nm marks PL_no_mem as "T" (text, aka code) rather than R
+    # but the symbol still ends up in the .rodata section on linking.
+    skip_all "LTO libperl.a flags don't match the final linker sections";
+}
+
 my $libperl_a;
 
 for my $f (qw(../libperl.a libperl.a)) {
