test2json.go

     1  // Copyright 2017 The Go Authors. All rights reserved.
     2  // Use of this source code is governed by a BSD-style
     3  // license that can be found in the LICENSE file.
     4  
     5  // Package test2json implements conversion of test binary output to JSON.
     6  // It is used by cmd/test2json and cmd/go.
     7  //
     8  // See the cmd/test2json documentation for details of the JSON encoding.
     9  package test2json
    10  
    11  import (
    12  	"bytes"
    13  	"encoding/json"
    14  	"fmt"
    15  	"io"
    16  	"strconv"
    17  	"strings"
    18  	"time"
    19  	"unicode"
    20  	"unicode/utf8"
    21  )
    22  
    23  // Mode controls details of the conversion.
    24  type Mode int
    25  
    26  const (
    27  	Timestamp Mode = 1 << iota // include Time in events
    28  )
    29  
    30  // event is the JSON struct we emit.
    31  type event struct {
    32  	Time        *time.Time `json:",omitempty"`
    33  	Action      string
    34  	Package     string     `json:",omitempty"`
    35  	Test        string     `json:",omitempty"`
    36  	Elapsed     *float64   `json:",omitempty"`
    37  	Output      *textBytes `json:",omitempty"`
    38  	FailedBuild string     `json:",omitempty"`
    39  }
    40  
    41  // textBytes is a hack to get JSON to emit a []byte as a string
    42  // without actually copying it to a string.
    43  // It implements encoding.TextMarshaler, which returns its text form as a []byte,
    44  // and then json encodes that text form as a string (which was our goal).
    45  type textBytes []byte
    46  
    47  func (b textBytes) MarshalText() ([]byte, error) { return b, nil }
    48  
    49  // A Converter holds the state of a test-to-JSON conversion.
    50  // It implements io.WriteCloser; the caller writes test output in,
    51  // and the converter writes JSON output to w.
    52  type Converter struct {
    53  	w          io.Writer  // JSON output stream
    54  	pkg        string     // package to name in events
    55  	mode       Mode       // mode bits
    56  	start      time.Time  // time converter started
    57  	testName   string     // name of current test, for output attribution
    58  	report     []*event   // pending test result reports (nested for subtests)
    59  	result     string     // overall test result if seen
    60  	input      lineBuffer // input buffer
    61  	output     lineBuffer // output buffer
    62  	needMarker bool       // require ^V marker to introduce test framing line
    63  
    64  	// failedBuild is set to the package ID of the cause of a build failure,
    65  	// if that's what caused this test to fail.
    66  	failedBuild string
    67  }
    68  
    69  // inBuffer and outBuffer are the input and output buffer sizes.
    70  // They're variables so that they can be reduced during testing.
    71  //
    72  // The input buffer needs to be able to hold any single test
    73  // directive line we want to recognize, like:
    74  //
    75  //	<many spaces> --- PASS: very/nested/s/u/b/t/e/s/t
    76  //
    77  // If anyone reports a test directive line > 4k not working, it will
    78  // be defensible to suggest they restructure their test or test names.
    79  //
    80  // The output buffer must be >= utf8.UTFMax, so that it can
    81  // accumulate any single UTF8 sequence. Lines that fit entirely
    82  // within the output buffer are emitted in single output events.
    83  // Otherwise they are split into multiple events.
    84  // The output buffer size therefore limits the size of the encoding
    85  // of a single JSON output event. 1k seems like a reasonable balance
    86  // between wanting to avoid splitting an output line and not wanting to
    87  // generate enormous output events.
    88  var (
    89  	inBuffer  = 4096
    90  	outBuffer = 1024
    91  )
    92  
    93  // NewConverter returns a "test to json" converter.
    94  // Writes on the returned writer are written as JSON to w,
    95  // with minimal delay.
    96  //
    97  // The writes to w are whole JSON events ending in \n,
    98  // so that it is safe to run multiple tests writing to multiple converters
    99  // writing to a single underlying output stream w.
   100  // As long as the underlying output w can handle concurrent writes
   101  // from multiple goroutines, the result will be a JSON stream
   102  // describing the relative ordering of execution in all the concurrent tests.
   103  //
   104  // The mode flag adjusts the behavior of the converter.
   105  // Passing ModeTime includes event timestamps and elapsed times.
   106  //
   107  // The pkg string, if present, specifies the import path to
   108  // report in the JSON stream.
   109  func NewConverter(w io.Writer, pkg string, mode Mode) *Converter {
   110  	c := new(Converter)
   111  	*c = Converter{
   112  		w:     w,
   113  		pkg:   pkg,
   114  		mode:  mode,
   115  		start: time.Now(),
   116  		input: lineBuffer{
   117  			b:    make([]byte, 0, inBuffer),
   118  			line: c.handleInputLine,
   119  			part: c.output.write,
   120  		},
   121  		output: lineBuffer{
   122  			b:    make([]byte, 0, outBuffer),
   123  			line: c.writeOutputEvent,
   124  			part: c.writeOutputEvent,
   125  		},
   126  	}
   127  	c.writeEvent(&event{Action: "start"})
   128  	return c
   129  }
   130  
   131  // Write writes the test input to the converter.
   132  func (c *Converter) Write(b []byte) (int, error) {
   133  	c.input.write(b)
   134  	return len(b), nil
   135  }
   136  
   137  // Exited marks the test process as having exited with the given error.
   138  func (c *Converter) Exited(err error) {
   139  	if err == nil {
   140  		if c.result != "skip" {
   141  			c.result = "pass"
   142  		}
   143  	} else {
   144  		c.result = "fail"
   145  	}
   146  }
   147  
   148  // SetFailedBuild sets the package ID that is the root cause of a build failure
   149  // for this test. This will be reported in the final "fail" event's FailedBuild
   150  // field.
   151  func (c *Converter) SetFailedBuild(pkgID string) {
   152  	c.failedBuild = pkgID
   153  }
   154  
   155  const marker = byte(0x16) // ^V
   156  
   157  var (
   158  	// printed by test on successful run.
   159  	bigPass = []byte("PASS")
   160  
   161  	// printed by test after a normal test failure.
   162  	bigFail = []byte("FAIL")
   163  
   164  	// printed by 'go test' along with an error if the test binary terminates
   165  	// with an error.
   166  	bigFailErrorPrefix = []byte("FAIL\t")
   167  
   168  	// an === NAME line with no test name, if trailing spaces are deleted
   169  	emptyName     = []byte("=== NAME")
   170  	emptyNameLine = []byte("=== NAME  \n")
   171  
   172  	updates = [][]byte{
   173  		[]byte("=== RUN   "),
   174  		[]byte("=== PAUSE "),
   175  		[]byte("=== CONT  "),
   176  		[]byte("=== NAME  "),
   177  		[]byte("=== PASS  "),
   178  		[]byte("=== FAIL  "),
   179  		[]byte("=== SKIP  "),
   180  	}
   181  
   182  	reports = [][]byte{
   183  		[]byte("--- PASS: "),
   184  		[]byte("--- FAIL: "),
   185  		[]byte("--- SKIP: "),
   186  		[]byte("--- BENCH: "),
   187  	}
   188  
   189  	fourSpace = []byte("    ")
   190  
   191  	skipLinePrefix = []byte("?   \t")
   192  	skipLineSuffix = []byte("\t[no test files]")
   193  )
   194  
   195  // handleInputLine handles a single whole test output line.
   196  // It must write the line to c.output but may choose to do so
   197  // before or after emitting other events.
   198  func (c *Converter) handleInputLine(line []byte) {
   199  	if len(line) == 0 {
   200  		return
   201  	}
   202  	sawMarker := false
   203  	if c.needMarker && line[0] != marker {
   204  		c.output.write(line)
   205  		return
   206  	}
   207  	if line[0] == marker {
   208  		c.output.flush()
   209  		sawMarker = true
   210  		line = line[1:]
   211  	}
   212  
   213  	// Trim is line without \n or \r\n.
   214  	trim := line
   215  	if len(trim) > 0 && trim[len(trim)-1] == '\n' {
   216  		trim = trim[:len(trim)-1]
   217  		if len(trim) > 0 && trim[len(trim)-1] == '\r' {
   218  			trim = trim[:len(trim)-1]
   219  		}
   220  	}
   221  
   222  	// === CONT followed by an empty test name can lose its trailing spaces.
   223  	if bytes.Equal(trim, emptyName) {
   224  		line = emptyNameLine
   225  		trim = line[:len(line)-1]
   226  	}
   227  
   228  	// Final PASS or FAIL.
   229  	if bytes.Equal(trim, bigPass) || bytes.Equal(trim, bigFail) || bytes.HasPrefix(trim, bigFailErrorPrefix) {
   230  		c.flushReport(0)
   231  		c.testName = ""
   232  		c.needMarker = sawMarker
   233  		c.output.write(line)
   234  		if bytes.Equal(trim, bigPass) {
   235  			c.result = "pass"
   236  		} else {
   237  			c.result = "fail"
   238  		}
   239  		return
   240  	}
   241  
   242  	// Special case for entirely skipped test binary: "?   \tpkgname\t[no test files]\n" is only line.
   243  	// Report it as plain output but remember to say skip in the final summary.
   244  	if bytes.HasPrefix(line, skipLinePrefix) && bytes.HasSuffix(trim, skipLineSuffix) && len(c.report) == 0 {
   245  		c.result = "skip"
   246  	}
   247  
   248  	// "=== RUN   "
   249  	// "=== PAUSE "
   250  	// "=== CONT  "
   251  	actionColon := false
   252  	origLine := line
   253  	ok := false
   254  	indent := 0
   255  	for _, magic := range updates {
   256  		if bytes.HasPrefix(line, magic) {
   257  			ok = true
   258  			break
   259  		}
   260  	}
   261  	if !ok {
   262  		// "--- PASS: "
   263  		// "--- FAIL: "
   264  		// "--- SKIP: "
   265  		// "--- BENCH: "
   266  		// but possibly indented.
   267  		for bytes.HasPrefix(line, fourSpace) {
   268  			line = line[4:]
   269  			indent++
   270  		}
   271  		for _, magic := range reports {
   272  			if bytes.HasPrefix(line, magic) {
   273  				actionColon = true
   274  				ok = true
   275  				break
   276  			}
   277  		}
   278  	}
   279  
   280  	// Not a special test output line.
   281  	if !ok {
   282  		// Lookup the name of the test which produced the output using the
   283  		// indentation of the output as an index into the stack of the current
   284  		// subtests.
   285  		// If the indentation is greater than the number of current subtests
   286  		// then the output must have included extra indentation. We can't
   287  		// determine which subtest produced this output, so we default to the
   288  		// old behaviour of assuming the most recently run subtest produced it.
   289  		if indent > 0 && indent <= len(c.report) {
   290  			c.testName = c.report[indent-1].Test
   291  		}
   292  		c.output.write(origLine)
   293  		return
   294  	}
   295  
   296  	// Parse out action and test name.
   297  	i := 0
   298  	if actionColon {
   299  		i = bytes.IndexByte(line, ':') + 1
   300  	}
   301  	if i == 0 {
   302  		i = len(updates[0])
   303  	}
   304  	action := strings.ToLower(strings.TrimSuffix(strings.TrimSpace(string(line[4:i])), ":"))
   305  	name := strings.TrimSpace(string(line[i:]))
   306  
   307  	e := &event{Action: action}
   308  	if line[0] == '-' { // PASS or FAIL report
   309  		// Parse out elapsed time.
   310  		if i := strings.Index(name, " ("); i >= 0 {
   311  			if strings.HasSuffix(name, "s)") {
   312  				t, err := strconv.ParseFloat(name[i+2:len(name)-2], 64)
   313  				if err == nil {
   314  					if c.mode&Timestamp != 0 {
   315  						e.Elapsed = &t
   316  					}
   317  				}
   318  			}
   319  			name = name[:i]
   320  		}
   321  		if len(c.report) < indent {
   322  			// Nested deeper than expected.
   323  			// Treat this line as plain output.
   324  			c.output.write(origLine)
   325  			return
   326  		}
   327  		// Flush reports at this indentation level or deeper.
   328  		c.needMarker = sawMarker
   329  		c.flushReport(indent)
   330  		e.Test = name
   331  		c.testName = name
   332  		c.report = append(c.report, e)
   333  		c.output.write(origLine)
   334  		return
   335  	}
   336  	// === update.
   337  	// Finish any pending PASS/FAIL reports.
   338  	c.needMarker = sawMarker
   339  	c.flushReport(0)
   340  	c.testName = name
   341  
   342  	if action == "name" {
   343  		// This line is only generated to get c.testName right.
   344  		// Don't emit an event.
   345  		return
   346  	}
   347  
   348  	if action == "pause" {
   349  		// For a pause, we want to write the pause notification before
   350  		// delivering the pause event, just so it doesn't look like the test
   351  		// is generating output immediately after being paused.
   352  		c.output.write(origLine)
   353  	}
   354  	c.writeEvent(e)
   355  	if action != "pause" {
   356  		c.output.write(origLine)
   357  	}
   358  
   359  	return
   360  }
   361  
   362  // flushReport flushes all pending PASS/FAIL reports at levels >= depth.
   363  func (c *Converter) flushReport(depth int) {
   364  	c.testName = ""
   365  	for len(c.report) > depth {
   366  		e := c.report[len(c.report)-1]
   367  		c.report = c.report[:len(c.report)-1]
   368  		c.writeEvent(e)
   369  	}
   370  }
   371  
   372  // Close marks the end of the go test output.
   373  // It flushes any pending input and then output (only partial lines at this point)
   374  // and then emits the final overall package-level pass/fail event.
   375  func (c *Converter) Close() error {
   376  	c.input.flush()
   377  	c.output.flush()
   378  	if c.result != "" {
   379  		e := &event{Action: c.result}
   380  		if c.mode&Timestamp != 0 {
   381  			dt := time.Since(c.start).Round(1 * time.Millisecond).Seconds()
   382  			e.Elapsed = &dt
   383  		}
   384  		if c.result == "fail" {
   385  			e.FailedBuild = c.failedBuild
   386  		}
   387  		c.writeEvent(e)
   388  	}
   389  	return nil
   390  }
   391  
   392  // writeOutputEvent writes a single output event with the given bytes.
   393  func (c *Converter) writeOutputEvent(out []byte) {
   394  	c.writeEvent(&event{
   395  		Action: "output",
   396  		Output: (*textBytes)(&out),
   397  	})
   398  }
   399  
   400  // writeEvent writes a single event.
   401  // It adds the package, time (if requested), and test name (if needed).
   402  func (c *Converter) writeEvent(e *event) {
   403  	e.Package = c.pkg
   404  	if c.mode&Timestamp != 0 {
   405  		t := time.Now()
   406  		e.Time = &t
   407  	}
   408  	if e.Test == "" {
   409  		e.Test = c.testName
   410  	}
   411  	js, err := json.Marshal(e)
   412  	if err != nil {
   413  		// Should not happen - event is valid for json.Marshal.
   414  		fmt.Fprintf(c.w, "testjson internal error: %v\n", err)
   415  		return
   416  	}
   417  	js = append(js, '\n')
   418  	c.w.Write(js)
   419  }
   420  
   421  // A lineBuffer is an I/O buffer that reacts to writes by invoking
   422  // input-processing callbacks on whole lines or (for long lines that
   423  // have been split) line fragments.
   424  //
   425  // It should be initialized with b set to a buffer of length 0 but non-zero capacity,
   426  // and line and part set to the desired input processors.
   427  // The lineBuffer will call line(x) for any whole line x (including the final newline)
   428  // that fits entirely in cap(b). It will handle input lines longer than cap(b) by
   429  // calling part(x) for sections of the line. The line will be split at UTF8 boundaries,
   430  // and the final call to part for a long line includes the final newline.
   431  type lineBuffer struct {
   432  	b    []byte       // buffer
   433  	mid  bool         // whether we're in the middle of a long line
   434  	line func([]byte) // line callback
   435  	part func([]byte) // partial line callback
   436  }
   437  
   438  // write writes b to the buffer.
   439  func (l *lineBuffer) write(b []byte) {
   440  	for len(b) > 0 {
   441  		// Copy what we can into l.b.
   442  		m := copy(l.b[len(l.b):cap(l.b)], b)
   443  		l.b = l.b[:len(l.b)+m]
   444  		b = b[m:]
   445  
   446  		// Process lines in l.b.
   447  		i := 0
   448  		for i < len(l.b) {
   449  			j, w := indexEOL(l.b[i:])
   450  			if j < 0 {
   451  				if !l.mid {
   452  					if j := bytes.IndexByte(l.b[i:], '\t'); j >= 0 {
   453  						if isBenchmarkName(bytes.TrimRight(l.b[i:i+j], " ")) {
   454  							l.part(l.b[i : i+j+1])
   455  							l.mid = true
   456  							i += j + 1
   457  						}
   458  					}
   459  				}
   460  				break
   461  			}
   462  			e := i + j + w
   463  			if l.mid {
   464  				// Found the end of a partial line.
   465  				l.part(l.b[i:e])
   466  				l.mid = false
   467  			} else {
   468  				// Found a whole line.
   469  				l.line(l.b[i:e])
   470  			}
   471  			i = e
   472  		}
   473  
   474  		// Whatever's left in l.b is a line fragment.
   475  		if i == 0 && len(l.b) == cap(l.b) {
   476  			// The whole buffer is a fragment.
   477  			// Emit it as the beginning (or continuation) of a partial line.
   478  			t := trimUTF8(l.b)
   479  			l.part(l.b[:t])
   480  			l.b = l.b[:copy(l.b, l.b[t:])]
   481  			l.mid = true
   482  		}
   483  
   484  		// There's room for more input.
   485  		// Slide it down in hope of completing the line.
   486  		if i > 0 {
   487  			l.b = l.b[:copy(l.b, l.b[i:])]
   488  		}
   489  	}
   490  }
   491  
   492  // indexEOL finds the index of a line ending,
   493  // returning its position and output width.
   494  // A line ending is either a \n or the empty string just before a ^V not beginning a line.
   495  // The output width for \n is 1 (meaning it should be printed)
   496  // but the output width for ^V is 0 (meaning it should be left to begin the next line).
   497  func indexEOL(b []byte) (pos, wid int) {
   498  	for i, c := range b {
   499  		if c == '\n' {
   500  			return i, 1
   501  		}
   502  		if c == marker && i > 0 { // test -v=json emits ^V at start of framing lines
   503  			return i, 0
   504  		}
   505  	}
   506  	return -1, 0
   507  }
   508  
   509  // flush flushes the line buffer.
   510  func (l *lineBuffer) flush() {
   511  	if len(l.b) > 0 {
   512  		// Must be a line without a \n, so a partial line.
   513  		l.part(l.b)
   514  		l.b = l.b[:0]
   515  	}
   516  }
   517  
   518  var benchmark = []byte("Benchmark")
   519  
   520  // isBenchmarkName reports whether b is a valid benchmark name
   521  // that might appear as the first field in a benchmark result line.
   522  func isBenchmarkName(b []byte) bool {
   523  	if !bytes.HasPrefix(b, benchmark) {
   524  		return false
   525  	}
   526  	if len(b) == len(benchmark) { // just "Benchmark"
   527  		return true
   528  	}
   529  	r, _ := utf8.DecodeRune(b[len(benchmark):])
   530  	return !unicode.IsLower(r)
   531  }
   532  
   533  // trimUTF8 returns a length t as close to len(b) as possible such that b[:t]
   534  // does not end in the middle of a possibly-valid UTF-8 sequence.
   535  //
   536  // If a large text buffer must be split before position i at the latest,
   537  // splitting at position trimUTF(b[:i]) avoids splitting a UTF-8 sequence.
   538  func trimUTF8(b []byte) int {
   539  	// Scan backward to find non-continuation byte.
   540  	for i := 1; i < utf8.UTFMax && i <= len(b); i++ {
   541  		if c := b[len(b)-i]; c&0xc0 != 0x80 {
   542  			switch {
   543  			case c&0xe0 == 0xc0:
   544  				if i < 2 {
   545  					return len(b) - i
   546  				}
   547  			case c&0xf0 == 0xe0:
   548  				if i < 3 {
   549  					return len(b) - i
   550  				}
   551  			case c&0xf8 == 0xf0:
   552  				if i < 4 {
   553  					return len(b) - i
   554  				}
   555  			}
   556  			break
   557  		}
   558  	}
   559  	return len(b)
   560  }
   561
View as plain text