diff options
Diffstat (limited to 'test/wrap.tcl')
-rw-r--r-- | test/wrap.tcl | 99 |
1 files changed, 99 insertions, 0 deletions
diff --git a/test/wrap.tcl b/test/wrap.tcl new file mode 100644 index 0000000..1d341c7 --- /dev/null +++ b/test/wrap.tcl @@ -0,0 +1,99 @@ +# See the file LICENSE for redistribution information. +# +# Copyright (c) 2000-2009 Oracle. All rights reserved. +# +# $Id$ +# +# Sentinel file wrapper for multi-process tests. This is designed to avoid a +# set of nasty bugs, primarily on Windows, where pid reuse causes watch_procs +# to sit around waiting for some random process that's not DB's and is not +# exiting. + +source ./include.tcl +source $test_path/testutils.tcl + +# Arguments: +if { $argc < 2 } { + puts "FAIL: wrap.tcl: Usage: wrap.tcl script log [scriptargs]" + exit +} + +set script [lindex $argv 0] +set logfile [lindex $argv 1] +if { $argc >= 2 } { + set skip [lindex $argv 2] + set args [lrange $argv 3 end] +} else { + set skip "" + set args "" +} +# +# Account in args for SKIP command, or not. +# +if { $skip != "SKIP" && $argc >= 2 } { + set args [lrange $argv 2 end] +} + +# Create a sentinel file to mark our creation and signal that watch_procs +# should look for us. +set parentpid [pid] +set parentsentinel $testdir/begin.$parentpid +set f [open $parentsentinel w] +close $f + +# Create a Tcl subprocess that will actually run the test. +set t [open "|$tclsh_path >& $logfile" w] + +# Create a sentinel for the subprocess. +set childpid [pid $t] +puts "Script watcher process $parentpid launching $script process $childpid." +set childsentinel $testdir/begin.$childpid +set f [open $childsentinel w] +close $f + +# +# For the upgrade tests where a current release tclsh is starting up +# a tclsh in an older release, we cannot tell it to source the current +# test.tcl because new things may not exist in the old release. So, +# we need to skip that and the script we're running in the old +# release will have to take care of itself. +# +if { $skip != "SKIP" } { + puts $t "source $test_path/test.tcl" +} +puts $t "set script $script" + +# Set up argv for the subprocess, since the args aren't passed in as true +# arguments thanks to the pipe structure. +puts $t "set argc [llength $args]" +puts $t "set argv [list $args]" + +set has_path [file dirname $script] +if { $has_path != "." } { + set scr $script +} else { + set scr $test_path/$script +} +#puts "Script $script: path $has_path, scr $scr" +puts $t "set scr $scr" +puts $t {set ret [catch { source $scr } result]} +puts $t {if { [string length $result] > 0 } { puts $result }} +puts $t {error_check_good "$scr run: $result: pid [pid]" $ret 0} + +# Close the pipe. This will flush the above commands and actually run the +# test, and will also return an error a la exec if anything bad happens +# to the subprocess. The magic here is that closing a pipe blocks +# and waits for the exit of processes in the pipeline, at least according +# to Ousterhout (p. 115). + +set ret [catch {close $t} res] + +# Write ending sentinel files--we're done. +set f [open $testdir/end.$childpid w] +close $f +set f [open $testdir/end.$parentpid w] +close $f + +error_check_good "Pipe close ($childpid: $script $argv: logfile $logfile)"\ + $ret 0 +exit $ret |