Fix another batch of `./validate --slow` failures

A rather detailed summary can be found at:

    https://gist.github.com/alpmestan/be82b47bb88b7dc9ff84105af9b1bb82

This doesn't fix all expectation mismatches yet, but we're down to about
20 mismatches with my previous patch and this one, as opposed to ~150
when I got started.

Test Plan: ./validate --slow

Reviewers: bgamari, erikd, simonmar

Reviewed By: simonmar

Subscribers: thomie, carter

GHC Trac Issues: #14890

Differential Revision: https://phabricator.haskell.org/D4636

8 files changed, 35 insertions, 20 deletions

diff --git a/testsuite/tests/rts/Makefile b/testsuite/tests/rts/Makefile
index 630508542d..5077f325c1 100644
--- a/testsuite/tests/rts/Makefile
+++ b/testsuite/tests/rts/Makefile
@@ -64,7 +64,7 @@ T9405:
 
 define run_T5435_v
 $(RM) T5435_load_v_$(1) T5435_v_$(1)$(exeext)
-'$(TEST_HC)' $(TEST_HC_OPTS) -optc-D$(HostOS)_HOST_OS -v0 -c T5435_$(1).c -o T5435_load_v_$(1).o
+'$(TEST_HC)' $(TEST_HC_OPTS) -optc-D$(HostOS)_HOST_OS -optc-DLOAD_CONSTR=$(2) -v0 -c T5435_$(1).c -o T5435_load_v_$(1).o
 '$(TEST_HC)' $(TEST_HC_OPTS) -v0 T5435.hs -osuf main_v_$(1)_o -o T5435_v_$(1)$(exeext)
 ./T5435_v_$(1) v ./T5435_load_v_$(1).o
 endef
@@ -78,11 +78,13 @@ endef
 
 .PHONY: T5435_v_gcc
 T5435_v_gcc :
-	$(call run_T5435_v,gcc)
+	$(call run_T5435_v,gcc,0)
 
-.PHONY: T5435_v_asm
-T5435_v_asm :
-	$(call run_T5435_v,asm)
+.PHONY: T5435_v_asm_a T5435_v_asm_b
+T5435_v_asm_a :
+	$(call run_T5435_v,asm,0)
+T5435_v_asm_b :
+	$(call run_T5435_v,asm,1)
 
 .PHONY: T5435_dyn_gcc
 T5435_dyn_gcc :
diff --git a/testsuite/tests/rts/T5435_asm.c b/testsuite/tests/rts/T5435_asm.c
index 59b53b1c3e..90813aa839 100644
--- a/testsuite/tests/rts/T5435_asm.c
+++ b/testsuite/tests/rts/T5435_asm.c
@@ -33,17 +33,19 @@ static void (*mod_init_func[2])(void) __attribute__((
 
 #else /* ELF */
 
+#if LOAD_CONSTR == 0
 static void (*const init_array[2])(void) __attribute__((
             section(".init_array"), // put it in the right section
             used,                   // prevent GCC from optimizing this away
             aligned(sizeof(void*))  // avoid slop between GCC's preloaded initializers and ours
             ))
     = {initArray1, initArray2};
-
+#else
 static void (*ctors[2])(void) __attribute__((
             section(".ctors"),
             used,
             aligned(sizeof(void*))))
     = {ctors2, ctors1}; // ctors run in reverse
+#endif
 
 #endif
diff --git a/testsuite/tests/rts/T5435_dyn_asm.stdout b/testsuite/tests/rts/T5435_dyn_asm.stdout
index 1893d0f56a..429c314309 100644
--- a/testsuite/tests/rts/T5435_dyn_asm.stdout
+++ b/testsuite/tests/rts/T5435_dyn_asm.stdout
@@ -1,5 +1,3 @@
 initArray1
 initArray2
-ctors1
-ctors2
 success
diff --git a/testsuite/tests/rts/T5435_v_asm_a.stdout b/testsuite/tests/rts/T5435_v_asm_a.stdout
new file mode 100644
index 0000000000..3124fa8559
--- /dev/null
+++ b/testsuite/tests/rts/T5435_v_asm_a.stdout
@@ -0,0 +1,3 @@
+initArray1
+initArray2
+success
\ No newline at end of file
diff --git a/testsuite/tests/rts/T5435_v_asm.stdout-darwin b/testsuite/tests/rts/T5435_v_asm_a.stdout-darwin
index 8827792585..8827792585 100644
--- a/testsuite/tests/rts/T5435_v_asm.stdout-darwin
+++ b/testsuite/tests/rts/T5435_v_asm_a.stdout-darwin
diff --git a/testsuite/tests/rts/T5435_v_asm.stdout-mingw32 b/testsuite/tests/rts/T5435_v_asm_a.stdout-mingw32
index 293bd12fb0..293bd12fb0 100644
--- a/testsuite/tests/rts/T5435_v_asm.stdout-mingw32
+++ b/testsuite/tests/rts/T5435_v_asm_a.stdout-mingw32
diff --git a/testsuite/tests/rts/T5435_v_asm_b.stdout b/testsuite/tests/rts/T5435_v_asm_b.stdout
new file mode 100644
index 0000000000..318d66e12b
--- /dev/null
+++ b/testsuite/tests/rts/T5435_v_asm_b.stdout
@@ -0,0 +1,3 @@
+ctors1
+ctors2
+success
\ No newline at end of file
diff --git a/testsuite/tests/rts/all.T b/testsuite/tests/rts/all.T
index 8fa2257b1c..a07179a4df 100644
--- a/testsuite/tests/rts/all.T
+++ b/testsuite/tests/rts/all.T
@@ -174,14 +174,7 @@ def checkDynAsm(actual_file, normaliser):
     actual_raw = read_no_crs(actual_file)
     actual_str = normaliser(actual_raw)
     actual = actual_str.split()
-    if actual == ['initArray1', 'initArray2', 'ctors1', 'ctors2', 'success']:
-        return 1
-    elif actual == ['initArray1', 'initArray2', 'ctors2', 'ctors1', 'success']:
-        # gold seems to produce this ordering; this is slightly odd but if it's
-        # wrong it's not our fault. See #13883.
-        return 1
-    elif actual == ['ctors1', 'ctors2', 'initArray1', 'initArray2', 'success']:
-        if_verbose(1, 'T5435_dyn_asm detected old-style dlopen, see #8458')
+    if actual == ['initArray1', 'initArray2', 'success']:
         return 1
     elif opsys('darwin') and actual == ['modInitFunc1', 'modInitFunc2', 'success']:
         return 1
@@ -191,13 +184,22 @@ def checkDynAsm(actual_file, normaliser):
         if_verbose(1, 'T5435_dyn_asm failed with %s, see all.T for details' % actual)
         return 0
 
+# T5435_v_asm got split into two tests because depending
+# on the linker, .init_array and .ctors sections are loaded
+# in a different order (but all entries within a section
+# do get loaded in a deterministic order). So we test each
+# separately now.
 # These should have extra_clean() arguments, but I need
 # to somehow extract out the name of DLLs to do that
-
-test('T5435_v_asm', [extra_files(['T5435.hs', 'T5435_asm.c']),
+test('T5435_v_asm_a', [extra_files(['T5435.hs', 'T5435_asm.c']),
                      when(arch('powerpc64') or arch('powerpc64le'),
                           expect_broken(11259))],
-     run_command, ['$MAKE -s --no-print-directory T5435_v_asm'])
+     run_command, ['$MAKE -s --no-print-directory T5435_v_asm_a'])
+# this one just needs to run on linux, as darwin/mingw32 are covered
+# by the _a test already.
+test('T5435_v_asm_b', [extra_files(['T5435.hs', 'T5435_asm.c']),
+		      when(opsys('darwin') or opsys('mingw32'), skip)],
+     run_command, ['$MAKE -s --no-print-directory T5435_v_asm_b'])
 test('T5435_v_gcc', [extra_files(['T5435.hs', 'T5435_gcc.c']),
                      when(arch('powerpc64') or arch('powerpc64le'),
                           expect_broken(11259))],
@@ -382,7 +384,12 @@ test('T12497', [ unless(opsys('mingw32'), skip)
                ],
                run_command, ['$MAKE -s --no-print-directory T12497'])
 
-test('T12903', [when(opsys('mingw32'), skip)], compile_and_run, [''])
+test('T12903', [ when(opsys('mingw32'), skip)
+               # produces out of sequence samples in the profasm way
+	       , expect_broken_for(15065, ['profasm'])
+	       , omit_ways(['ghci'])
+	       ], compile_and_run, [''])
+
 test('T13832', exit_code(1), compile_and_run, ['-threaded'])
 test('T13894', normal, compile_and_run, [''])
 test('T14497', normal, compile_and_run, ['-O'])