gm: display summary of all results by ErrorType

gm/gmmain.cpp

Index: gm/gmmain.cpp
===================================================================
--- gm/gmmain.cpp	(revision 8399)
+++ gm/gmmain.cpp	(working copy)
@@ -34,6 +34,7 @@
 #include "SkRefCnt.h"
 #include "SkStream.h"
 #include "SkTArray.h"
+#include "SkTDict.h"
 #include "SkTileGridPicture.h"
 #include "SamplePipeControllers.h"
 
@@ -64,6 +65,7 @@
 #endif
 
 extern bool gSkSuppressFontCachePurgeSpew;
+const static int kAnyIntValue = 1; // when it doesn't matter what int value, just SOMETHING

[borenet, 2013/04/03 12:46:48]

What if we counted the number of times we encounter each render mode?  I'm not
sure we'll ever care what those numbers will be, but it seems more interesting
than a dummy int.

[epoger, 2013/04/03 14:28:25]

Done.  Note that this now takes twice as long, since it has to look up
renderModeDescriptor in the dictionary *twice*.  (It would perhaps be good for
SkTDict to provide a method that returns a pointer to the name/value pair, so
you could modify the value in place... I guess that could lead to some
"interesting" error cases, though.)

 
 #ifdef SK_SUPPORT_PDF
     #include "SkPDFDevice.h"
@@ -86,14 +88,6 @@
 
 using namespace skiagm;
 
-struct FailRec {
-    SkString    fName;
-    bool        fIsPixelError;
-
-    FailRec() : fIsPixelError(false) {}
-    FailRec(const SkString& name) : fName(name), fIsPixelError(false) {}
-};
-
 class Iter {
 public:
     Iter() {
@@ -184,12 +178,14 @@
 
 class GMMain {
 public:
-    GMMain() {
+    GMMain() : fRenderModesEncountered(1) {

[borenet, 2013/04/03 12:46:48]

Do you mind moving fUseFileHierarchy, fMismatchPath, and fTestsRuns into this
initialization list?

[epoger, 2013/04/03 14:28:25]

Done.

         // Set default values of member variables, which tool_main()
         // may override.
         fUseFileHierarchy = false;
         fIgnorableErrorCombination.add(kMissingExpectations_ErrorType);
         fMismatchPath = NULL;
+        fTestsRun = 0;
+        fRenderModesEncountered.reset();
     }
 
     SkString make_name(const char shortName[], const char configName[]) {
@@ -243,34 +239,100 @@
     }
 
     /**
-     * Records the errors encountered in fFailedTests, except for any error
-     * types we want to ignore.
+     * Add all render modes encountered thus far to the "modes" array.
      */
-    void RecordError(const ErrorCombination& errorCombination, const SkString& name,
-                     const char renderModeDescriptor []) {
-        // The common case: no error means nothing to record.
+    void GetRenderModesEncountered(SkTArray<SkString> &modes) {
+        SkTDict<int>::Iter iter(this->fRenderModesEncountered);
+        const char* mode;
+        while ((mode = iter.next(NULL)) != NULL) {
+            SkString modeAsString = SkString(mode);
+            // TODO(epoger): It seems a bit silly that all of these modes were
+            // recorded with a leading "-" which we have to remove here
+            // (except for mode "", which means plain old original mode).
+            // But that's how renderModeDescriptor has been passed into
+            // compare_test_results_to_reference_bitmap() historically,
+            // and changing that now may affect other parts of our code.
+            if (modeAsString.startsWith("-")) {
+                modeAsString.remove(0, 1);
+                modes.push_back(modeAsString);
+            }
+        }
+    }
+
+    /**
+     * Records the results of this test in fTestsRun and fFailedTests.
+     *
+     * We even record successes, and errors that we regard as
+     * "ignorable"; we can filter them out later.
+     */
+    void RecordTestResults(const ErrorCombination& errorCombination, const SkString& name,
+                           const char renderModeDescriptor []) {
+        // Things to do regardless of errorCombination.
+        fTestsRun++;
+        this->fRenderModesEncountered.set(renderModeDescriptor, kAnyIntValue);
+
         if (errorCombination.isEmpty()) {
             return;
         }
 
-        // If only certain error type(s) were reported, we know we can ignore them.
-        if (errorCombination.minus(fIgnorableErrorCombination).isEmpty()) {
-            return;
+        // Things to do only if there is some error condition.
+        SkString fullName = make_name(name.c_str(), renderModeDescriptor);
+        for (int typeInt = 0; typeInt <= kLast_ErrorType; typeInt++) {
+            ErrorType type = static_cast<ErrorType>(typeInt);
+            if (errorCombination.includes(type)) {
+                fFailedTests[type].push_back(fullName);
+            }
         }
+    }
 
-        FailRec& rec = fFailedTests.push_back(make_name(name.c_str(), renderModeDescriptor));
-        rec.fIsPixelError = errorCombination.includes(kImageMismatch_ErrorType);
+    /**
+     * Return the number of significant (non-ignorable) errors we have
+     * encountered so far.
+     */
+    int NumSignificantErrors() {
+        int significantErrors = 0;
+        for (int typeInt = 0; typeInt <= kLast_ErrorType; typeInt++) {
+            ErrorType type = static_cast<ErrorType>(typeInt);
+            if (!fIgnorableErrorCombination.includes(type)) {
+                significantErrors += fFailedTests[type].count();
+            }
+        }
+        return significantErrors;
     }
 
-    // List contents of fFailedTests via SkDebug.
+    /**
+     * List contents of fFailedTests to stdout.
+     */
     void ListErrors() {
-        for (int i = 0; i < fFailedTests.count(); ++i) {
-            if (fFailedTests[i].fIsPixelError) {
-                gm_fprintf(stderr, "\t\t%s pixel_error\n", fFailedTests[i].fName.c_str());
+        // First, print a single summary line.
+        SkString summary;
+        summary.appendf("Ran %d tests:", fTestsRun);
+        for (int typeInt = 0; typeInt <= kLast_ErrorType; typeInt++) {
+            ErrorType type = static_cast<ErrorType>(typeInt);
+            summary.appendf(" %s=%d", getErrorTypeName(type), fFailedTests[type].count());
+        }
+        gm_fprintf(stdout, "%s\n", summary.c_str());
+
+        // Now, for each failure type, list the tests that failed that way.
+        for (int typeInt = 0; typeInt <= kLast_ErrorType; typeInt++) {
+            SkString line;
+            ErrorType type = static_cast<ErrorType>(typeInt);
+            if (fIgnorableErrorCombination.includes(type)) {
+                line.append("[ ] ");
             } else {
-                gm_fprintf(stderr, "\t\t%s\n", fFailedTests[i].fName.c_str());
+                line.append("[*] ");
             }
+
+            SkTArray<SkString> *failedTestsOfThisType = &fFailedTests[type];
+            int count = failedTestsOfThisType->count();
+            line.appendf("%d %s:", count, getErrorTypeName(type));
+            for (int i = 0; i < count; ++i) {
+                line.append(" ");
+                line.append((*failedTestsOfThisType)[i]);
+            }
+            gm_fprintf(stdout, "%s\n", line.c_str());
         }
+        gm_fprintf(stdout, "(results marked with [*] will cause nonzero return value)\n");
     }
 
     static bool write_document(const SkString& path,
@@ -510,7 +572,16 @@
         } else {
             gm_fprintf(stderr, "FAILED to write %s\n", path.c_str());
             ErrorCombination errors(kWritingReferenceImage_ErrorType);
-            RecordError(errors, name, renderModeDescriptor);
+            // TODO(epoger): Don't call RecordTestResults() here...
+            // Instead, we should make sure to call RecordTestResults
+            // exactly ONCE per test.  (Otherwise, gmmain.fTestsRun
+            // will be incremented twice for this test: once in
+            // compare_test_results_to_stored_expectations() before
+            // that method calls this one, and again here.)
+            //
+            // When we make that change, we should probably add a
+            // WritingReferenceImage test to the gm self-tests.)
+            RecordTestResults(errors, name, renderModeDescriptor);
             return errors;
         }
     }
@@ -622,7 +693,7 @@
                 report_bitmap_diffs(*expectedBitmapPtr, actualBitmap, completeName);
             }
         }
-        RecordError(errors, baseNameString, renderModeDescriptor);
+        RecordTestResults(errors, baseNameString, renderModeDescriptor);
 
         if (addToJsonSummary) {
             add_actual_results_to_json_summary(completeName, actualChecksum, errors,
@@ -738,6 +809,7 @@
             add_actual_results_to_json_summary(name.c_str(), actualChecksum,
                                                ErrorCombination(kMissingExpectations_ErrorType),
                                                false);
+            RecordTestResults(ErrorCombination(kMissingExpectations_ErrorType), name, "");
         }
 
         // TODO: Consider moving this into compare_to_expectations(),
@@ -765,6 +837,13 @@
         GM* gm, const ConfigData& gRec, const char renderModeDescriptor [],
         SkBitmap& actualBitmap, const SkBitmap* referenceBitmap) {
 
+        // TODO(epoger): This method is run to compare results across
+        // different rendering modes (as opposed to
+        // compare_test_results_to_stored_expectations(), which
+        // compares results against expectations stored on disk).  If
+        // we would like the GenerateGMs step to distinguish between
+        // those two types of mismatches, we should report image
+        // mismatches in here with a different ErrorType.
         SkASSERT(referenceBitmap);
         SkString name = make_name(gm->shortName(), gRec.fName);
         Expectations expectations(*referenceBitmap);
@@ -880,6 +959,8 @@
             // ('image-surface gm test is failing in "deferred" mode,
             // and gm is not reporting the failure')
             if (errors.isEmpty()) {
+                // TODO(epoger): Report this as a new ErrorType,
+                // something like kImageGeneration_ErrorType?
                 return kEmpty_ErrorCombination;
             }
             return compare_test_results_to_reference_bitmap(
@@ -953,8 +1034,10 @@
 
     const char* fMismatchPath;
 
-    // information about all failed tests we have encountered so far
-    SkTArray<FailRec> fFailedTests;
+    // collection of tests that have failed with each ErrorType
+    SkTArray<SkString> fFailedTests[kLast_ErrorType+1];
+    int fTestsRun;
+    SkTDict<int> fRenderModesEncountered;
 
     // Where to read expectations (expected image checksums, etc.) from.
     // If unset, we don't do comparisons.
@@ -1265,7 +1348,9 @@
     ErrorCombination errorsForAllModes;
     uint32_t gmFlags = gm->getFlags();
 
-    // run the picture centric GM steps
+    // TODO(epoger): We should start recording any per-GM skipped
+    // modes (i.e. those we skipped due to gmFlags) with a new
+    // ErrorType, perhaps named kIntentionallySkipped_ErrorType.
     if (!(gmFlags & GM::kSkipPicture_Flag)) {
 
         ErrorCombination pictErrors;
@@ -1359,6 +1444,45 @@
     return errorsForAllModes;
 }
 
+/**
+ * Return a list of all entries in an array of strings as a single string
+ * of this form:
+ * "item1", "item2", "item3"
+ */
+SkString list_all(const SkTArray<SkString> &stringArray);
+SkString list_all(const SkTArray<SkString> &stringArray) {
+    SkString total;
+    for (int i = 0; i < stringArray.count(); i++) {
+        if (i > 0) {
+            total.append(", ");
+        }
+        total.append("\"");
+        total.append(stringArray[i]);
+        total.append("\"");
+    }
+    return total;
+}
+
+/**
+ * Return a list of configuration names, as a single string of this form:
+ * "item1", "item2", "item3"
+ *
+ * @param configs configurations, as a list of indices into gRec
+ */
+SkString list_all_config_names(const SkTDArray<size_t> &configs);
+SkString list_all_config_names(const SkTDArray<size_t> &configs) {
+    SkString total;
+    for (int i = 0; i < configs.count(); i++) {
+        if (i > 0) {
+            total.append(", ");
+        }
+        total.append("\"");
+        total.append(gRec[configs[i]].fName);
+        total.append("\"");
+    }
+    return total;
+}
+
 int tool_main(int argc, char** argv);
 int tool_main(int argc, char** argv) {
 
@@ -1540,12 +1664,7 @@
         moduloRemainder = -1;
     }
 
-    // Accumulate success of all tests.
-    int testsRun = 0;
-    int testsPassed = 0;
-    int testsFailed = 0;
-    int testsMissingReferenceImages = 0;
-
+    int gmsRun = 0;
     int gmIndex = -1;
     SkString moduloStr;
 
@@ -1585,43 +1704,44 @@
             continue;
         }
 
+        gmsRun++;
         SkISize size = gm->getISize();
         gm_fprintf(stdout, "%sdrawing... %s [%d %d]\n", moduloStr.c_str(), shortName,
                    size.width(), size.height());
 
-        ErrorCombination testErrors;
-        testErrors.add(run_multiple_configs(gmmain, gm, configs, grFactory));
+        run_multiple_configs(gmmain, gm, configs, grFactory);
 
         SkBitmap comparisonBitmap;
         const ConfigData compareConfig =
             { SkBitmap::kARGB_8888_Config, kRaster_Backend, kDontCare_GLContextType, 0, kRW_ConfigFlag, "comparison", false };
-        testErrors.add(gmmain.generate_image(
-            gm, compareConfig, NULL, NULL, &comparisonBitmap, false));
+        gmmain.generate_image(gm, compareConfig, NULL, NULL, &comparisonBitmap, false);
 
         // TODO(epoger): only run this if gmmain.generate_image() succeeded?
         // Otherwise, what are we comparing against?
-        testErrors.add(run_multiple_modes(gmmain, gm, compareConfig, comparisonBitmap,
-                                          tileGridReplayScales));
+        run_multiple_modes(gmmain, gm, compareConfig, comparisonBitmap, tileGridReplayScales);
 
-        // Update overall results.
-        // We only tabulate the particular error types that we currently
-        // care about (e.g., missing reference images). Later on, if we
-        // want to also tabulate other error types, we can do so.
-        testsRun++;
-        if (!gmmain.fExpectationsSource.get() ||
-            (testErrors.includes(kMissingExpectations_ErrorType))) {
-            testsMissingReferenceImages++;
-        }
-        if (testErrors.minus(gmmain.fIgnorableErrorCombination).isEmpty()) {
-            testsPassed++;
-        } else {
-            testsFailed++;
-        }
-
         SkDELETE(gm);
     }
-    gm_fprintf(stdout, "Ran %d tests: %d passed, %d failed, %d missing reference images\n",
-               testsRun, testsPassed, testsFailed, testsMissingReferenceImages);
+
+    SkTArray<SkString> modes;
+    gmmain.GetRenderModesEncountered(modes);
+
+    // Output summary to stdout.
+    gm_fprintf(stdout, "Ran %d GMs\n", gmsRun);
+    gm_fprintf(stdout, "... over %2d configs [%s]\n", configs.count(),
+               list_all_config_names(configs).c_str());
+    gm_fprintf(stdout, "...  and %2d modes [%s]\n", modes.count(), list_all(modes).c_str());
+    gm_fprintf(stdout, "... so there should be a total of %d tests.\n",
+               gmsRun * (configs.count() + modes.count()));
+
+    // TODO(epoger): Ultimately, we should signal an error if the
+    // expected total number of tests (displayed above) does not match
+    // gmmain.fTestsRun.  But for now, there are cases where those
+    // numbers won't match: specifically, if some configs/modes are
+    // skipped on a per-GM basis (due to gm->getFlags() for a specific
+    // GM).  Later on, we should record tests like that using some new
+    // ErrorType, like kIntentionallySkipped_ErrorType.  Then we could
+    // signal an error if the totals didn't match up.
     gmmain.ListErrors();
 
     if (FLAGS_writeJsonSummaryPath.count() == 1) {
@@ -1661,7 +1781,7 @@
 #endif
     SkGraphics::Term();
 
-    return (0 == testsFailed) ? 0 : -1;
+    return (0 == gmmain.NumSignificantErrors()) ? 0 : -1;
 }
 
 void GMMain::installFilter(SkCanvas* canvas) {