Debugging is twice as hard as writing the code in the first place. Therefore, if you write the code as cleverly as possible, you are, by definition, not smart enough to debug it. --Brian Kernighan |
The Bash shell contains no built-in debugger, and only bare-bones debugging-specific commands and constructs. Syntax errors or outright typos in the script generate cryptic error messages that are often of no help in debugging a non-functional script.
Example 29-1. A buggy script
#!/bin/bash # ex74.sh # This is a buggy script. # Where, oh where is the error? a=37 if [$a -gt 27 ] then echo $a fi exit 0 |
Output from script:
./ex74.sh: [37: command not found |
Example 29-2. Missing keyword
#!/bin/bash # missing-keyword.sh: What error message will this generate? for a in 1 2 3 do echo "$a" # done # Required keyword 'done' commented out in line 7. exit 0 |
Output from script:
missing-keyword.sh: line 10: syntax error: unexpected end of file |
Error messages may disregard comment lines in a script when reporting the line number of a syntax error.
What if the script executes, but does not work as expected? This is the all too familiar logic error.
Example 29-3. test24: another buggy script
#!/bin/bash # This script is supposed to delete all filenames in current directory #+ containing embedded spaces. # It doesn't work. # Why not? badname=`ls | grep ' '` # Try this: # echo "$badname" rm "$badname" exit 0 |
Try to find out what's wrong with Example 29-3 by uncommenting the echo "$badname" line. Echo statements are useful for seeing whether what you expect is actually what you get.
In this particular case, rm "$badname" will not give the desired results because $badname should not be quoted. Placing it in quotes ensures that rm has only one argument (it will match only one filename). A partial fix is to remove to quotes from $badname and to reset $IFS to contain only a newline, IFS=$'\n'. However, there are simpler ways of going about it.
# Correct methods of deleting filenames containing spaces. rm *\ * rm *" "* rm *' '* # Thank you. S.C. |
Summarizing the symptoms of a buggy script,
It bombs with a "syntax error" message, or
It runs, but does not work as expected (logic error).
It runs, works as expected, but has nasty side effects (logic bomb).
Tools for debugging non-working scripts include
Inserting echo statements at critical points in the script to trace the variables, and otherwise give a snapshot of what is going on.
Even better is an echo that echoes only when debug is on.
|
Using the tee filter to check processes or data flows at critical points.
Setting option flags -n -v -x
sh -n scriptname checks for syntax errors without actually running the script. This is the equivalent of inserting set -n or set -o noexec into the script. Note that certain types of syntax errors can slip past this check.
sh -v scriptname echoes each command before executing it. This is the equivalent of inserting set -v or set -o verbose in the script.
The -n and -v flags work well together. sh -nv scriptname gives a verbose syntax check.
sh -x scriptname echoes the result each command, but in an abbreviated manner. This is the equivalent of inserting set -x or set -o xtrace in the script.
Inserting set -u or set -o nounset in the script runs it, but gives an unbound variable error message at each attempt to use an undeclared variable.
Using an "assert" function to test a variable or condition at critical points in a script. (This is an idea borrowed from C.)
Example 29-4. Testing a condition with an assert
#!/bin/bash # assert.sh ####################################################################### assert () # If condition false, { #+ exit from script #+ with appropriate error message. E_PARAM_ERR=98 E_ASSERT_FAILED=99 if [ -z "$2" ] # Not enough parameters passed then #+ to assert() function. return $E_PARAM_ERR # No damage done. fi lineno=$2 if [ ! $1 ] then echo "Assertion failed: \"$1\"" echo "File \"$0\", line $lineno" # Give name of file and line number. exit $E_ASSERT_FAILED # else # return # and continue executing the script. fi } # Insert a similar assert() function into a script you need to debug. ####################################################################### a=5 b=4 condition="$a -lt $b" # Error message and exit from script. # Try setting "condition" to something else #+ and see what happens. assert "$condition" $LINENO # The remainder of the script executes only if the "assert" does not fail. # Some commands. # Some more commands . . . echo "This statement echoes only if the \"assert\" does not fail." # . . . # More commands . . . exit $? |
The exit command in a script triggers a signal 0, terminating the process, that is, the script itself. [1] It is often useful to trap the exit, forcing a "printout" of variables, for example. The trap must be the first command in the script.
Specifies an action on receipt of a signal; also useful for debugging.
A signal is a message sent to a process, either by the kernel or another process, telling it to take some specified action (usually to terminate). For example, hitting a Control-C sends a user interrupt, an INT signal, to a running program. |
A simple instance:
trap '' 2 # Ignore interrupt 2 (Control-C), with no action specified. trap 'echo "Control-C disabled."' 2 # Message when Control-C pressed. |
Example 29-5. Trapping at exit
#!/bin/bash # Hunting variables with a trap. trap 'echo Variable Listing --- a = $a b = $b' EXIT # EXIT is the name of the signal generated upon exit from a script. # # The command specified by the "trap" doesn't execute until #+ the appropriate signal is sent. echo "This prints before the \"trap\" --" echo "even though the script sees the \"trap\" first." echo a=39 b=36 exit 0 # Note that commenting out the 'exit' command makes no difference, #+ since the script exits in any case after running out of commands. |
Example 29-6. Cleaning up after Control-C
#!/bin/bash # logon.sh: A quick 'n dirty script to check whether you are on-line yet. umask 177 # Make sure temp files are not world readable. TRUE=1 LOGFILE=/var/log/messages # Note that $LOGFILE must be readable #+ (as root, chmod 644 /var/log/messages). TEMPFILE=temp.$$ # Create a "unique" temp file name, using process id of the script. # Using 'mktemp' is an alternative. # For example: # TEMPFILE=`mktemp temp.XXXXXX` KEYWORD=address # At logon, the line "remote IP address xxx.xxx.xxx.xxx" # appended to /var/log/messages. ONLINE=22 USER_INTERRUPT=13 CHECK_LINES=100 # How many lines in log file to check. trap 'rm -f $TEMPFILE; exit $USER_INTERRUPT' TERM INT # Cleans up the temp file if script interrupted by control-c. echo while [ $TRUE ] #Endless loop. do tail -n $CHECK_LINES $LOGFILE> $TEMPFILE # Saves last 100 lines of system log file as temp file. # Necessary, since newer kernels generate many log messages at log on. search=`grep $KEYWORD $TEMPFILE` # Checks for presence of the "IP address" phrase, #+ indicating a successful logon. if [ ! -z "$search" ] # Quotes necessary because of possible spaces. then echo "On-line" rm -f $TEMPFILE # Clean up temp file. exit $ONLINE else echo -n "." # The -n option to echo suppresses newline, #+ so you get continuous rows of dots. fi sleep 1 done # Note: if you change the KEYWORD variable to "Exit", #+ this script can be used while on-line #+ to check for an unexpected logoff. # Exercise: Change the script, per the above note, # and prettify it. exit 0 # Nick Drage suggests an alternate method: while true do ifconfig ppp0 | grep UP 1> /dev/null && echo "connected" && exit 0 echo -n "." # Prints dots (.....) until connected. sleep 2 done # Problem: Hitting Control-C to terminate this process may be insufficient. #+ (Dots may keep on echoing.) # Exercise: Fix this. # Stephane Chazelas has yet another alternative: CHECK_INTERVAL=1 while ! tail -n 1 "$LOGFILE" | grep -q "$KEYWORD" do echo -n . sleep $CHECK_INTERVAL done echo "On-line" # Exercise: Discuss the relative strengths and weaknesses # of each of these various approaches. |
Of course, the trap command has other uses aside from debugging, such as disabling certain keystrokes within a script (see Example A-43).
Example 29-8. Running multiple processes (on an SMP box)
#!/bin/bash # parent.sh # Running multiple processes on an SMP box. # Author: Tedman Eng # This is the first of two scripts, #+ both of which must be present in the current working directory. LIMIT=$1 # Total number of process to start NUMPROC=4 # Number of concurrent threads (forks?) PROCID=1 # Starting Process ID echo "My PID is $$" function start_thread() { if [ $PROCID -le $LIMIT ] ; then ./child.sh $PROCID& let "PROCID++" else echo "Limit reached." wait exit fi } while [ "$NUMPROC" -gt 0 ]; do start_thread; let "NUMPROC--" done while true do trap "start_thread" SIGRTMIN done exit 0 # ======== Second script follows ======== #!/bin/bash # child.sh # Running multiple processes on an SMP box. # This script is called by parent.sh. # Author: Tedman Eng temp=$RANDOM index=$1 shift let "temp %= 5" let "temp += 4" echo "Starting $index Time:$temp" "$@" sleep ${temp} echo "Ending $index" kill -s SIGRTMIN $PPID exit 0 # ======================= SCRIPT AUTHOR'S NOTES ======================= # # It's not completely bug free. # I ran it with limit = 500 and after the first few hundred iterations, #+ one of the concurrent threads disappeared! # Not sure if this is collisions from trap signals or something else. # Once the trap is received, there's a brief moment while executing the #+ trap handler but before the next trap is set. During this time, it may #+ be possible to miss a trap signal, thus miss spawning a child process. # No doubt someone may spot the bug and will be writing #+ . . . in the future. # ===================================================================== # # ----------------------------------------------------------------------# ################################################################# # The following is the original script written by Vernia Damiano. # Unfortunately, it doesn't work properly. ################################################################# #!/bin/bash # Must call script with at least one integer parameter #+ (number of concurrent processes). # All other parameters are passed through to the processes started. INDICE=8 # Total number of process to start TEMPO=5 # Maximum sleep time per process E_BADARGS=65 # No arg(s) passed to script. if [ $# -eq 0 ] # Check for at least one argument passed to script. then echo "Usage: `basename $0` number_of_processes [passed params]" exit $E_BADARGS fi NUMPROC=$1 # Number of concurrent process shift PARAMETRI=( "$@" ) # Parameters of each process function avvia() { local temp local index temp=$RANDOM index=$1 shift let "temp %= $TEMPO" let "temp += 1" echo "Starting $index Time:$temp" "$@" sleep ${temp} echo "Ending $index" kill -s SIGRTMIN $$ } function parti() { if [ $INDICE -gt 0 ] ; then avvia $INDICE "${PARAMETRI[@]}" & let "INDICE--" else trap : SIGRTMIN fi } trap parti SIGRTMIN while [ "$NUMPROC" -gt 0 ]; do parti; let "NUMPROC--" done wait trap - SIGRTMIN exit $? : <<SCRIPT_AUTHOR_COMMENTS I had the need to run a program, with specified options, on a number of different files, using a SMP machine. So I thought [I'd] keep running a specified number of processes and start a new one each time . . . one of these terminates. The "wait" instruction does not help, since it waits for a given process or *all* process started in background. So I wrote [this] bash script that can do the job, using the "trap" instruction. --Vernia Damiano SCRIPT_AUTHOR_COMMENTS |
trap '' SIGNAL (two adjacent apostrophes) disables SIGNAL for the remainder of the script. trap SIGNAL restores the functioning of SIGNAL once more. This is useful to protect a critical portion of a script from an undesirable interrupt. |
trap '' 2 # Signal 2 is Control-C, now disabled. command command command trap 2 # Reenables Control-C |
Version 3 of Bash adds the following internal variables for use by the debugger.
|
[1] | By convention, signal 0 is assigned to exit. |
Friday, December 27, 2024 @ 8:28:12 PM
z.ServerAdmin@lam1.us