1exec(n) Tcl Built-In Commands exec(n)
2______________________________________________________________________________
6
8 exec - Invoke subprocesses
9
11 exec ?switches? arg ?arg ...? ?&?
12______________________________________________________________________________
13
15 This command treats its arguments as the specification of one or more
16 subprocesses to execute. The arguments take the form of a standard
17 shell pipeline where each arg becomes one word of a command, and each
18 distinct command becomes a subprocess.
19 If the initial arguments to exec start with - then they are treated as
21 command-line switches and are not part of the pipeline specification.
22 The following switches are currently supported:
23 -ignorestderr
25 Stops the exec command from treating the output of mes‐
26 sages to the pipeline's standard error channel as an error
27 case.
28 -keepnewline Retains a trailing newline in the pipeline's output. Nor‐
30 mally a trailing newline will be deleted.
31 -- Marks the end of switches. The argument following this
33 one will be treated as the first arg even if it starts
34 with a -.
35 If an arg (or pair of args) has one of the forms described below then
37 it is used by exec to control the flow of input and output among the
38 subprocess(es). Such arguments will not be passed to the subpro‐
39 cess(es). In forms such as “< fileName”, fileName may either be in a
40 separate argument from “<” or in the same argument with no intervening
41 space (i.e. “<fileName”).
42 | Separates distinct commands in the pipeline. The stan‐
44 dard output of the preceding command will be piped into
45 the standard input of the next command.
46 |& Separates distinct commands in the pipeline. Both stan‐
48 dard output and standard error of the preceding command
49 will be piped into the standard input of the next com‐
50 mand. This form of redirection overrides forms such as
51 2> and >&.
52 < fileName The file named by fileName is opened and used as the
54 standard input for the first command in the pipeline.
55 <@ fileId FileId must be the identifier for an open file, such as
57 the return value from a previous call to open. It is
58 used as the standard input for the first command in the
59 pipeline. FileId must have been opened for reading.
60 << value Value is passed to the first command as its standard
62 input.
63 > fileName Standard output from the last command is redirected to
65 the file named fileName, overwriting its previous con‐
66 tents.
67 2> fileName Standard error from all commands in the pipeline is
69 redirected to the file named fileName, overwriting its
70 previous contents.
71 >& fileName Both standard output from the last command and standard
73 error from all commands are redirected to the file named
74 fileName, overwriting its previous contents.
75 >> fileName Standard output from the last command is redirected to
77 the file named fileName, appending to it rather than
78 overwriting it.
79 2>> fileName Standard error from all commands in the pipeline is
81 redirected to the file named fileName, appending to it
82 rather than overwriting it.
83 >>& fileName Both standard output from the last command and standard
85 error from all commands are redirected to the file named
86 fileName, appending to it rather than overwriting it.
87 >@ fileId FileId must be the identifier for an open file, such as
89 the return value from a previous call to open. Standard
90 output from the last command is redirected to fileId's
91 file, which must have been opened for writing.
92 2>@ fileId FileId must be the identifier for an open file, such as
94 the return value from a previous call to open. Standard
95 error from all commands in the pipeline is redirected to
96 fileId's file. The file must have been opened for writ‐
97 ing.
98 2>@1 Standard error from all commands in the pipeline is
100 redirected to the command result. This operator is only
101 valid at the end of the command pipeline.
102 >&@ fileId FileId must be the identifier for an open file, such as
104 the return value from a previous call to open. Both
105 standard output from the last command and standard error
106 from all commands are redirected to fileId's file. The
107 file must have been opened for writing.
108 If standard output has not been redirected then the exec command
110 returns the standard output from the last command in the pipeline,
111 unless “2>@1” was specified, in which case standard error is included
112 as well. If any of the commands in the pipeline exit abnormally or are
113 killed or suspended, then exec will return an error and the error mes‐
114 sage will include the pipeline's output followed by error messages
115 describing the abnormal terminations; the -errorcode return option will
116 contain additional information about the last abnormal termination
117 encountered. If any of the commands writes to its standard error file
118 and that standard error is not redirected and -ignorestderr is not
119 specified, then exec will return an error; the error message will
120 include the pipeline's standard output, followed by messages about
121 abnormal terminations (if any), followed by the standard error output.
122 If the last character of the result or error message is a newline then
124 that character is normally deleted from the result or error message.
125 This is consistent with other Tcl return values, which do not normally
126 end with newlines. However, if -keepnewline is specified then the
127 trailing newline is retained.
128 If standard input is not redirected with “<”, “<<” or “<@” then the
130 standard input for the first command in the pipeline is taken from the
131 application's current standard input.
132 If the last arg is “&” then the pipeline will be executed in back‐
134 ground. In this case the exec command will return a list whose ele‐
135 ments are the process identifiers for all of the subprocesses in the
136 pipeline. The standard output from the last command in the pipeline
137 will go to the application's standard output if it has not been redi‐
138 rected, and error output from all of the commands in the pipeline will
139 go to the application's standard error file unless redirected.
140 The first word in each command is taken as the command name; tilde-sub‐
142 stitution is performed on it, and if the result contains no slashes
143 then the directories in the PATH environment variable are searched for
144 an executable by the given name. If the name contains a slash then it
145 must refer to an executable reachable from the current directory. No
146 “glob” expansion or other shell-like substitutions are performed on the
147 arguments to commands.
148
150 Windows (all versions)
151 Reading from or writing to a socket, using the “@ fileId” nota‐
152 tion, does not work. When reading from a socket, a 16-bit DOS
153 application will hang and a 32-bit application will return imme‐
154 diately with end-of-file. When either type of application
155 writes to a socket, the information is instead sent to the con‐
156 sole, if one is present, or is discarded.
157 Note that the current escape resp. quoting of arguments for win‐
159 dows works only with executables using CommandLineToArgv, CRT-
160 library or similar, as well as with the windows batch files
161 (excepting the newline, see below). Although it is the common
162 escape algorithm, but, in fact, the way how the executable
163 parses the command-line (resp. splits it into single arguments)
164 is decisive.
165 Unfortunately, there is currently no way to supply newline char‐
167 acter within an argument to the batch files (.cmd or .bat) or to
168 the command processor (cmd.exe /c), because this causes trunca‐
169 tion of command-line (also the argument chain) on the first new‐
170 line character. But it works properly with an executable (using
171 CommandLineToArgv, etc).
172 The Tk console text widget does not provide real standard IO
174 capabilities. Under Tk, when redirecting from standard input,
175 all applications will see an immediate end-of-file; information
176 redirected to standard output or standard error will be dis‐
177 carded.
178 Either forward or backward slashes are accepted as path separa‐
180 tors for arguments to Tcl commands. When executing an applica‐
181 tion, the path name specified for the application may also con‐
182 tain forward or backward slashes as path separators. Bear in
183 mind, however, that most Windows applications accept arguments
184 with forward slashes only as option delimiters and backslashes
185 only in paths. Any arguments to an application that specify a
186 path name with forward slashes will not automatically be con‐
187 verted to use the backslash character. If an argument contains
188 forward slashes as the path separator, it may or may not be rec‐
189 ognized as a path name, depending on the program.
190 Additionally, when calling a 16-bit DOS or Windows 3.X applica‐
192 tion, all path names must use the short, cryptic, path format
193 (e.g., using “applba~1.def” instead of “applbakery.default”),
194 which can be obtained with the “file attributes fileName -short‐
195 name” command.
196 Two or more forward or backward slashes in a row in a path refer
198 to a network path. For example, a simple concatenation of the
199 root directory c:/ with a subdirectory /windows/system will
200 yield c://windows/system (two slashes together), which refers to
201 the mount point called system on the machine called windows (and
202 the c:/ is ignored), and is not equivalent to c:/windows/system,
203 which describes a directory on the current computer. The file
204 join command should be used to concatenate path components.
205 Note that there are two general types of Win32 console applica‐
207 tions:
208 [1] CLI — CommandLine Interface, simple stdio
210 exchange. netstat.exe for example.
211 [2] TUI — Textmode User Interface, any application
213 that accesses the console API for doing such
214 things as cursor movement, setting text color,
215 detecting key presses and mouse movement, etc. An
216 example would be telnet.exe from Windows 2000.
217 These types of applications are not common in a
218 windows environment, but do exist.
219 exec will not work well with TUI applications when a console is
221 not present, as is done when launching applications under wish.
222 It is desirable to have console applications hidden and
223 detached. This is a designed-in limitation as exec wants to
224 communicate over pipes. The Expect extension addresses this
225 issue when communicating with a TUI application.
226 When attempting to execute an application, exec first searches
228 for the name as it was specified. Then, in order, .com, .exe,
229 .bat and .cmd are appended to the end of the specified name and
230 it searches for the longer name. If a directory name was not
231 specified as part of the application name, the following direc‐
232 tories are automatically searched in order when attempting to
233 locate the application:
234 · The directory from which the Tcl executable was loaded.
236 · The current directory.
238 · The Windows NT 32-bit system directory.
240 · The Windows NT 16-bit system directory.
242 · The Windows NT home directory.
244 · The directories listed in the path.
246 In order to execute shell built-in commands like dir and copy,
248 the caller must prepend the desired command with “cmd.exe /c ”
249 because built-in commands are not implemented using executables.
250 Unix (including Mac OS X)
252 The exec command is fully functional and works as described.
253
255 Here are some examples of the use of the exec command on Unix. To exe‐
256 cute a simple program and get its result:
257 exec uname -a
259 WORKING WITH NON-ZERO RESULTS
261 To execute a program that can return a non-zero result, you should wrap
262 the call to exec in catch and check the contents of the -errorcode
263 return option if you have an error:
264 set status 0
266 if {[catch {exec grep foo bar.txt} results options]} {
267 set details [dict get $options -errorcode]
268 if {[lindex $details 0] eq "CHILDSTATUS"} {
269 set status [lindex $details 2]
270 } else {
271 # Some other error; regenerate it to let caller handle
272 return -options $options -level 0 $results
273 }
274 }
275 This is more easily written using the try command, as that makes it │
277 simpler to trap specific types of errors. This is done using code like │
278 this: │
279 try { │
281 set results [exec grep foo bar.txt] │
282 set status 0 │
283 } trap CHILDSTATUS {results options} { │
284 set status [lindex [dict get $options -errorcode] 2] │
285 } │
286 WORKING WITH QUOTED ARGUMENTS
288 When translating a command from a Unix shell invocation, care should be
289 taken over the fact that single quote characters have no special sig‐
290 nificance to Tcl. Thus:
291 awk '{sum += $1} END {print sum}' numbers.list
293 would be translated into something like:
295 exec awk {{sum += $1} END {print sum}} numbers.list
297 WORKING WITH GLOBBING
299 If you are converting invocations involving shell globbing, you should
300 remember that Tcl does not handle globbing or expand things into multi‐
301 ple arguments by default. Instead you should write things like this:
302 exec ls -l {*}[glob *.tcl]
304 WORKING WITH USER-SUPPLIED SHELL SCRIPT FRAGMENTS
306 One useful technique can be to expose to users of a script the ability
307 to specify a fragment of shell script to execute that will have some
308 data passed in on standard input that was produced by the Tcl program.
309 This is a common technique for using the lpr program for printing. By
310 far the simplest way of doing this is to pass the user's script to the
311 user's shell for processing, as this avoids a lot of complexity with
312 parsing other languages.
313 set lprScript [get from user...]
315 set postscriptData [generate somehow...]
316 exec $env(SHELL) -c $lprScript << $postscriptData
318
320 Here are some examples of the use of the exec command on Windows. To
321 start an instance of notepad editing a file without waiting for the
322 user to finish editing the file:
323 exec notepad myfile.txt &
325 To print a text file using notepad:
327 exec notepad /p myfile.txt
329 WORKING WITH CONSOLE PROGRAMS
331 If a program calls other programs, such as is common with compilers,
332 then you may need to resort to batch files to hide the console windows
333 that sometimes pop up:
334 exec cmp.bat somefile.c -o somefile
336 With the file cmp.bat looking something like:
338 @gcc %*
340 or like another variant using single parameters:
341 @gcc %1 %2 %3 %4 %5 %6 %7 %8 %9
342 WORKING WITH COMMAND BUILT-INS
344 Sometimes you need to be careful, as different programs may have the
345 same name and be in the path. It can then happen that typing a command
346 at the DOS prompt finds a different program than the same command run
347 via exec. This is because of the (documented) differences in behaviour
348 between exec and DOS batch files.
349 When in doubt, use the command auto_execok: it will return the complete
351 path to the program as seen by the exec command. This applies espe‐
352 cially when you want to run “internal” commands like dir from a Tcl
353 script (if you just want to list filenames, use the glob command.) To
354 do that, use this:
355 exec {*}[auto_execok dir] *.tcl
357 WORKING WITH NATIVE FILENAMES
359 Many programs on Windows require filename arguments to be passed in
360 with backslashes as pathname separators. This is done with the help of
361 the file nativename command. For example, to make a directory (on NTFS)
362 encrypted so that only the current user can access it requires use of
363 the CIPHER command, like this:
364 set secureDir "~/Desktop/Secure Directory"
366 file mkdir $secureDir
367 exec CIPHER /e /s:[file nativename $secureDir]
368
370 error(n), file(n), open(n)
371
373 execute, pipeline, redirection, subprocess
374Tcl 8.5 exec(n)