autodebugger tool

2 files changed, 83 insertions, 9 deletions

diff --git a/tests/runner.py b/tests/runner.py
index 35be05a6..27a939bd 100644
--- a/tests/runner.py
+++ b/tests/runner.py
@@ -27,6 +27,7 @@ EMSCRIPTEN = path_from_root('emscripten.py')
 DEMANGLER = path_from_root('third_party', 'demangler.py')
 NAMESPACER = path_from_root('tools', 'namespacer.py')
 EMMAKEN = path_from_root('tools', 'emmaken.py')
+AUTODEBUGGER = path_from_root('tools', 'autodebugger.py')
 
 # Global cache for tests (we have multiple TestCase instances; this object lets them share data)
 
@@ -183,6 +184,9 @@ class RunnerCore(unittest.TestCase):
     assert 'strict warning:' not in ret, 'We should pass all strict mode checks: ' + ret
     return ret
 
+  def run_llvm_interpreter(self, args):
+    return Popen([LLVM_INTERPRETER] + args, stdout=PIPE, stderr=STDOUT).communicate()[0]
+
   def assertContained(self, value, string):
     if value not in string:
       raise Exception("Expected to find '%s' in '%s'" % (value, string))
@@ -200,7 +204,7 @@ if 'benchmark' not in sys.argv:
 
   class T(RunnerCore): # Short name, to make it more fun to use manually on the commandline
     ## Does a complete test - builds, runs, checks output, etc.
-    def do_test(self, src, expected_output, args=[], output_nicerizer=None, output_processor=None, no_build=False, main_file=None, additional_files=[], js_engines=None, post_build=None, basename='src.cpp', libraries=[], includes=[], force_c=False):
+    def do_test(self, src, expected_output=None, args=[], output_nicerizer=None, output_processor=None, no_build=False, main_file=None, additional_files=[], js_engines=None, post_build=None, basename='src.cpp', libraries=[], includes=[], force_c=False):
         #print 'Running test:', inspect.stack()[1][3].replace('test_', ''), '[%s,%s,%s]' % (COMPILER.split(os.sep)[-1], 'llvm-optimizations' if LLVM_OPTS else '', 'reloop&optimize' if RELOOP else '')
         if force_c or (main_file is not None and main_file[-2:]) == '.c':
           basename = 'src.c'
@@ -215,6 +219,11 @@ if 'benchmark' not in sys.argv:
         if post_build is not None:
           post_build(filename + '.o.js')
 
+        # If not provided with expected output, then generate it right now, using lli
+        if expected_output is None:
+          expected_output = self.run_llvm_interpreter([filename + '.o'])
+          print '[autogenerated expected output: %20s]' % (expected_output[0:17].replace('\n', '')+'...')
+
         # Run in both JavaScript engines, if optimizing - significant differences there (typed arrays)
         if js_engines is None:
           js_engines = [V8_ENGINE, SPIDERMONKEY_ENGINE]
@@ -227,7 +236,7 @@ if 'benchmark' not in sys.argv:
 
         #shutil.rmtree(dirname) # TODO: leave no trace in memory. But for now nice for debugging
 
-    def prep_ll_test(self, filename, ll_file):
+    def prep_ll_test(self, filename, ll_file, force_recompile=False):
       if ll_file.endswith(('.bc', '.o')):
         shutil.copy(ll_file, filename + '.o')
         self.do_llvm_dis(filename)
@@ -235,24 +244,25 @@ if 'benchmark' not in sys.argv:
         os.remove(filename + '.o.ll')
         ll_file = filename + '.o.ll.in'
 
-      if LLVM_OPTS:
+      if LLVM_OPTS or force_recompile:
         shutil.copy(ll_file, filename + '.o.ll.pre')
-        Popen([LLVM_AS, filename + '.o.ll.pre'] + ['-o=' + filename + '.o'], stdout=PIPE, stderr=STDOUT).communicate()[0]
+        output = Popen([LLVM_AS, filename + '.o.ll.pre'] + ['-o=' + filename + '.o'], stdout=PIPE, stderr=STDOUT).communicate()[0]
+        assert 'error:' not in output, 'Error in llvm-as: ' + output
         self.do_llvm_opts(filename)
         Popen([LLVM_DIS, filename + '.o'] + LLVM_DIS_OPTS + ['-o=' + filename + '.o.ll'], stdout=PIPE, stderr=STDOUT).communicate()[0]
       else:
         shutil.copy(ll_file, filename + '.o.ll')
 
     # No building - just process an existing .ll file (or .bc, which we turn into .ll)
-    def do_ll_test(self, ll_file, output, args=[], js_engines=None, output_nicerizer=None, post_build=None):
+    def do_ll_test(self, ll_file, expected_output=None, args=[], js_engines=None, output_nicerizer=None, post_build=None, force_recompile=False):
       if COMPILER != LLVM_GCC: return # We use existing .ll, so which compiler is unimportant
 
       filename = os.path.join(self.get_dir(), 'src.cpp')
 
-      self.prep_ll_test(filename, ll_file)
+      self.prep_ll_test(filename, ll_file, force_recompile)
       self.do_emscripten(filename)
       self.do_test(None,
-                   output,
+                   expected_output,
                    args,
                    no_build=True,
                    js_engines=js_engines,
@@ -1663,8 +1673,7 @@ if 'benchmark' not in sys.argv:
 
         # Generate the native code output using lli
         lli_file = os.path.join(self.get_dir(), 'lli.raw')
-        stdout = Popen([LLVM_INTERPRETER, os.path.join(self.get_dir(), 'src.c.o'), '-i', original_j2k, '-o', lli_file],
-              stdout=PIPE, stderr=STDOUT).communicate()[0]
+        stdout = self.run_llvm_interpreter([os.path.join(self.get_dir(), 'src.c.o'), '-i', original_j2k, '-o', lli_file])
         assert 'Successfully generated' in stdout, 'Error in lli run: ' + stdout
         lli_data = open(lli_file, 'rb').read()
 
@@ -1743,6 +1752,22 @@ if 'benchmark' not in sys.argv:
           output = 'hello, world!'
         self.do_ll_test(path_from_root('tests', 'cases', name), output)
 
+    def test_autodebug(self):
+      if COMPILER != LLVM_GCC: return # TODO: Check both
+      if LLVM_OPTS: return # They mess us up
+
+      # Run a test that should work, generating some code
+      self.test_structs()
+
+      # Autodebug the code
+      filename = os.path.join(self.get_dir(), 'src.cpp.o.ll')
+      output = Popen(['python', AUTODEBUGGER, filename, filename+'.ll'], stdout=PIPE, stderr=STDOUT).communicate()[0]
+      assert 'Success.' in output
+
+      # Compare to each other, and to expected output
+      self.do_ll_test(path_from_root('tests', filename+'.ll'), force_recompile=True)
+      self.do_ll_test(path_from_root('tests', filename+'.ll'), '34 : 10\n42 : 7008\n51 : 7018') # No need to force recompile twice - already autodebugged
+
     ### Integration tests
 
     def test_scriptaclass(self):
diff --git a/tools/autodebugger.py b/tools/autodebugger.py
new file mode 100644
index 00000000..b90501cc
--- /dev/null
+++ b/tools/autodebugger.py
@@ -0,0 +1,49 @@
+'''
+Processes an LLVM assembly (.ll) file, adding debugging information.
+
+You can then run the .ll file in the LLVM interpreter (lli) and 
+compare that to the output when compiled using emscripten.
+'''
+
+import os, sys, re
+
+POSTAMBLE = '''
+@.emscripten.autodebug.str = private constant [9 x i8] c"%d : %d\\0A\\00", align 1 ; [#uses=1]
+
+; [#uses=1]
+define void @emscripten_autodebug(i32 %line, i32 %value) {
+entry:
+  %0 = call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([9 x i8]* @.emscripten.autodebug.str, i32 0, i32 0), i32 %line, i32 %value) ; [#uses=0]
+  br label %return
+
+return:                                           ; preds = %entry
+  ret void
+}
+'''
+
+filename, ofilename = sys.argv[1], sys.argv[2]
+f = open(filename, 'r')
+data = f.read()
+f.close()
+
+if 'declare i32 @printf(' not in data:
+  POSTAMBLE += '''
+; [#uses=1]
+declare i32 @printf(i8*, ...)
+'''
+
+lines = data.split('\n')
+for i in range(len(lines)):
+  #if i == 5:
+  #  lines[i] += '\n
+
+  m = re.match('  store (?P<type>i64|i32|i16|i8) %(?P<var>[\w.]+), .*', lines[i])
+  if m and m.group('type') == 'i32': # TODO: Other types
+    lines[i] += '\n  call void @emscripten_autodebug(i32 %d, i32 %%%s)' % (i+1, m.group('var'))
+
+f = open(ofilename, 'w')
+f.write('\n'.join(lines) + '\n' + POSTAMBLE + '\n')
+f.close()
+
+print 'Success.'
+