path: root/src/cmd/go/testdata/script/README

This directory holds test scripts *.txt run during 'go test cmd/go'.
To run a specific script foo.txt

	go test cmd/go -run=Script/^foo$

In general script files should have short names: a few words, not whole sentences.
The first word should be the general category of behavior being tested,
often the name of a go subcommand (list, build, test, ...) or concept (vendor, pattern).

Each script is a text archive (go doc cmd/go/internal/txtar).
The script begins with an actual command script to run
followed by the content of zero or more supporting files to
create in the script's temporary file system before it starts executing.

As an example, run_hello.txt says:

	# hello world
	go run hello.go
	stderr 'hello world'
	! stdout .

	-- hello.go --
	package main
	func main() { println("hello world") }

Each script runs in a fresh temporary work directory tree, available to scripts as $WORK.
Scripts also have access to these other environment variables:

	GOARCH=<target GOARCH>
	GOCACHE=<actual GOCACHE being used outside the test>
	GOEXE=<executable file suffix: .exe on Windows, empty on other systems>
	GOOS=<target GOOS>
	GOPATH=$WORK/gopath
	GOPROXY=<local module proxy serving from cmd/go/testdata/mod>
	GOROOT=<actual GOROOT>
	GOROOT_FINAL=<actual GOROOT_FINAL>
	TESTGO_GOROOT=<GOROOT used to build cmd/go, for use in tests that may change GOROOT>
	HOME=/no-home
	PATH=<actual PATH>
	TMPDIR=$WORK/tmp
	GODEBUG=<actual GODEBUG>
	devnull=<value of os.DevNull>
	goversion=<current Go version; for example, 1.12>
	:=<OS-specific path list separator>

The scripts' supporting files are unpacked relative to $GOPATH/src (aka $WORK/gopath/src)
and then the script begins execution in that directory as well. Thus the example above runs
in $WORK/gopath/src with GOPATH=$WORK/gopath and $WORK/gopath/src/hello.go
containing the listed contents.

The lines at the top of the script are a sequence of commands to be executed
by a tiny script engine in ../../script_test.go (not the system shell).
The script stops and the overall test fails if any particular command fails.

Each line is parsed into a sequence of space-separated command words,
with environment variable expansion and # marking an end-of-line comment.
Adding single quotes around text keeps spaces in that text from being treated
as word separators and also disables environment variable expansion.
Inside a single-quoted block of text, a repeated single quote indicates
a literal single quote, as in:

	'Don''t communicate by sharing memory.'

A line beginning with # is a comment and conventionally explains what is
being done or tested at the start of a new phase in the script.

The command prefix ! indicates that the command on the rest of the line
(typically go or a matching predicate) must fail, not succeed. Only certain
commands support this prefix. They are indicated below by [!] in the synopsis.

The command prefix ? indicates that the command on the rest of the line
may or may not succeed, but the test should continue regardless.
Commands that support this prefix are indicated by [?].

The command prefix [cond] indicates that the command on the rest of the line
should only run when the condition is satisfied. The available conditions are:

 - GOOS and GOARCH values, like [386], [windows], and so on.
 - Compiler names, like [gccgo], [gc].
 - Test environment details:
   - [short] for testing.Short()
   - [cgo], [msan], [race] for whether cgo, msan, and the race detector can be used
   - [net] for whether the external network can be used
   - [link] for testenv.HasLink()
   - [root] for os.Geteuid() == 0
   - [symlink] for testenv.HasSymlink()
   - [case-sensitive] for whether the file system is case-sensitive
   - [exec:prog] for whether prog is available for execution (found by exec.LookPath)
   - [GODEBUG:value] for whether value is one of the comma-separated entries in the GODEBUG variable
   - [buildmode:value] for whether -buildmode=value is supported

A condition can be negated: [!short] means to run the rest of the line
when testing.Short() is false. Multiple conditions may be given for a single
command, for example, '[linux] [amd64] skip'. The command will run if all conditions
are satisfied.

The commands are:

- [! | ?] cc args... [&]
  Run the C compiler, the platform specific flags (i.e. `go env GOGCCFLAGS`) will be
  added automatically before args.

- cd dir
  Change to the given directory for future commands.

- chmod perm path...
  Change the permissions of the files or directories named by the path arguments
  to be equal to perm. Only numerical permissions are supported.

- cmp file1 file2
  Check that the named files have the same content.
  By convention, file1 is the actual data and file2 the expected data.
  File1 can be "stdout" or "stderr" to use the standard output or standard error
  from the most recent exec or go command.
  (If the files have differing content, the failure prints a diff.)

- cmpenv file1 file2
  Like cmp, but environment variables are substituted in the file contents
  before the comparison. For example, $GOOS is replaced by the target GOOS.

- [! | ?] cp src... dst
  Copy the listed files to the target file or existing directory.
  src can include "stdout" or "stderr" to use the standard output or standard error
  from the most recent exec or go command.

- env [-r] [key=value...]
  With no arguments, print the environment to stdout
  (useful for debugging and for verifying initial state).
  Otherwise add the listed key=value pairs to the environment.
  The -r flag causes the values to be escaped using regexp.QuoteMeta
  before being recorded.

- [! | ?] exec program [args...] [&]
  Run the given executable program with the arguments.
  It must (or must not) succeed.
  Note that 'exec' does not terminate the script (unlike in Unix shells).

  If the last token is '&', the program executes in the background. The standard
  output and standard error of the previous command is cleared, but the output
  of the background process is buffered — and checking of its exit status is
  delayed — until the next call to 'wait', 'skip', or 'stop' or the end of the
  test. If any background processes remain at the end of the test, they
  are terminated using os.Interrupt (if supported) or os.Kill and the test
  must not depend upon their exit status.

- [!] exists [-readonly] [-exec] file...
  Each of the listed files or directories must (or must not) exist.
  If -readonly is given, the files or directories must be unwritable.
  If -exec is given, the files or directories must be executable.

- [! | ?] go args... [&]
  Run the (test copy of the) go command with the given arguments.
  It must (or must not) succeed.

- [!] grep [-count=N] [-q] pattern file
  The file's content must (or must not) match the regular expression pattern.
  For positive matches, -count=N specifies an exact number of matches to require.
  The -q flag disables printing the file content on a mismatch.

- mkdir path...
  Create the listed directories, if they do not already exists.

- rm file...
  Remove the listed files or directories.

- skip [message]
  Mark the test skipped, including the message if given.

- [!] stale path...
  The packages named by the path arguments must (or must not)
  be reported as "stale" by the go command.

- [!] stderr [-count=N] pattern
  Apply the grep command (see above) to the standard error
  from the most recent exec, go, or wait command.

- [!] stdout [-count=N] pattern
  Apply the grep command (see above) to the standard output
  from the most recent exec, go, wait, or env command.

- stop [message]
  Stop the test early (marking it as passing), including the message if given.

- symlink file -> target
  Create file as a symlink to target. The -> (like in ls -l output) is required.

- wait
  Wait for all 'exec' and 'go' commands started in the background (with the '&'
  token) to exit, and display success or failure status for them.
  After a call to wait, the 'stderr' and 'stdout' commands will apply to the
  concatenation of the corresponding streams of the background commands,
  in the order in which those commands were started.

When TestScript runs a script and the script fails, by default TestScript shows
the execution of the most recent phase of the script (since the last # comment)
and only shows the # comments for earlier phases. For example, here is a
multi-phase script with a bug in it:

	# GOPATH with p1 in d2, p2 in d2
	env GOPATH=$WORK/d1${:}$WORK/d2

	# build & install p1
	env
	go install -i p1
	! stale p1
	! stale p2

	# modify p2 - p1 should appear stale
	cp $WORK/p2x.go $WORK/d2/src/p2/p2.go
	stale p1 p2

	# build & install p1 again
	go install -i p11
	! stale p1
	! stale p2

	-- $WORK/d1/src/p1/p1.go --
	package p1
	import "p2"
	func F() { p2.F() }
	-- $WORK/d2/src/p2/p2.go --
	package p2
	func F() {}
	-- $WORK/p2x.go --
	package p2
	func F() {}
	func G() {}

The bug is that the final phase installs p11 instead of p1. The test failure looks like:

	$ go test -run=Script
	--- FAIL: TestScript (3.75s)
	    --- FAIL: TestScript/install_rebuild_gopath (0.16s)
	        script_test.go:223:
	            # GOPATH with p1 in d2, p2 in d2 (0.000s)
	            # build & install p1 (0.087s)
	            # modify p2 - p1 should appear stale (0.029s)
	            # build & install p1 again (0.022s)
	            > go install -i p11
	            [stderr]
	            can't load package: package p11: cannot find package "p11" in any of:
	            	/Users/rsc/go/src/p11 (from $GOROOT)
	            	$WORK/d1/src/p11 (from $GOPATH)
	            	$WORK/d2/src/p11
	            [exit status 1]
	            FAIL: unexpected go command failure

	        script_test.go:73: failed at testdata/script/install_rebuild_gopath.txt:15 in $WORK/gopath/src

	FAIL
	exit status 1
	FAIL	cmd/go	4.875s
	$

Note that the commands in earlier phases have been hidden, so that the relevant
commands are more easily found, and the elapsed time for a completed phase
is shown next to the phase heading. To see the entire execution, use "go test -v",
which also adds an initial environment dump to the beginning of the log.

Note also that in reported output, the actual name of the per-script temporary directory
has been consistently replaced with the literal string $WORK.

The cmd/go test flag -testwork (which must appear on the "go test" command line after
standard test flags) causes each test to log the name of its $WORK directory and other
environment variable settings and also to leave that directory behind when it exits,
for manual debugging of failing tests:

	$ go test -run=Script -work
	--- FAIL: TestScript (3.75s)
	    --- FAIL: TestScript/install_rebuild_gopath (0.16s)
	        script_test.go:223:
	            WORK=/tmp/cmd-go-test-745953508/script-install_rebuild_gopath
	            GOARCH=
	            GOCACHE=/Users/rsc/Library/Caches/go-build
	            GOOS=
	            GOPATH=$WORK/gopath
	            GOROOT=/Users/rsc/go
	            HOME=/no-home
	            TMPDIR=$WORK/tmp
	            exe=

	            # GOPATH with p1 in d2, p2 in d2 (0.000s)
	            # build & install p1 (0.085s)
	            # modify p2 - p1 should appear stale (0.030s)
	            # build & install p1 again (0.019s)
	            > go install -i p11
	            [stderr]
	            can't load package: package p11: cannot find package "p11" in any of:
	            	/Users/rsc/go/src/p11 (from $GOROOT)
	            	$WORK/d1/src/p11 (from $GOPATH)
	            	$WORK/d2/src/p11
	            [exit status 1]
	            FAIL: unexpected go command failure

	        script_test.go:73: failed at testdata/script/install_rebuild_gopath.txt:15 in $WORK/gopath/src

	FAIL
	exit status 1
	FAIL	cmd/go	4.875s
	$

	$ WORK=/tmp/cmd-go-test-745953508/script-install_rebuild_gopath
	$ cd $WORK/d1/src/p1
	$ cat p1.go
	package p1
	import "p2"
	func F() { p2.F() }
	$