Adding capr.pl
This commit is contained in:
419
src/cap5/capr.pl
Normal file
419
src/cap5/capr.pl
Normal file
@@ -0,0 +1,419 @@
|
||||
#!/usr/bin/perl -w
|
||||
|
||||
#######################################################################
|
||||
#
|
||||
# capr: A program that attempts to do a "dbpr" command via channel
|
||||
# access.
|
||||
#
|
||||
#######################################################################
|
||||
|
||||
use strict;
|
||||
use Getopt::Std;
|
||||
use lib "/dls_sw/epics/R3.14.11/base/lib/perl";
|
||||
use CA;
|
||||
|
||||
######### Globals ##########
|
||||
|
||||
# Non APS users will want to modify theDbdFile default settings
|
||||
|
||||
my $theDbdFile = "/net/helios/iocapps/R3.13.9/ioc/linac/2/dbd/linac.dbd";
|
||||
|
||||
my $hostArch;
|
||||
if ( defined $ENV{"EPICS_HOST_ARCH"} ) {
|
||||
$hostArch = $ENV{"EPICS_HOST_ARCH"} ;
|
||||
} else {
|
||||
$hostArch="solaris-sparc";
|
||||
}
|
||||
|
||||
our( $opt_d, $opt_f, $opt_r);
|
||||
my $usage = "Usage:\n $0 [-d dbd_file] ( [-f record_type] | [-r] | [record_name] ) [interest]\n";
|
||||
my %record = (); # Empty hash to put dbd data in
|
||||
my $iIdx = 0; # Array indexes for interest, data type and base
|
||||
my $tIdx = 1;
|
||||
my $bIdx = 2;
|
||||
my %device = (); # Empty hash to record which rec types have device support
|
||||
my $DEBUG=0; # DEBUG
|
||||
|
||||
# EPICS field types referenced to their equivalent EZCA types
|
||||
my %fieldType = (
|
||||
DBF_STRING => "ezcaString",
|
||||
DBF_BYTE => "ezcaByte",
|
||||
DBF_CHAR => "ezcaByte",
|
||||
DBF_UCHAR => "ezcaChar",
|
||||
DBF_SHORT => "ezcaShort",
|
||||
DBF_USHORT => "ezcaLong",
|
||||
DBF_LONG => "ezcaLong",
|
||||
DBF_ULONG => "ezcaDouble",
|
||||
DBF_FLOAT => "ezcaFloat",
|
||||
DBF_DOUBLE => "ezcaDouble",
|
||||
DBF_ENUM => "ezcaString",
|
||||
DBF_MENU => "ezcaString",
|
||||
DBF_DEVICE => "ezcaString",
|
||||
DBF_INLINK => "ezcaString",
|
||||
DBF_OUTLINK => "ezcaString",
|
||||
DBF_FWDLINK => "ezcaString",
|
||||
DBF_NOACCESS => "ezcaNoAccess"
|
||||
);
|
||||
|
||||
|
||||
######### Main program ############
|
||||
|
||||
getopts('d:f:r') or die "$usage";
|
||||
|
||||
# Select the dbd file to use
|
||||
if($opt_d) { # command line has highest priority
|
||||
$theDbdFile = $opt_d;
|
||||
}
|
||||
elsif (exists $ENV{CAPR_DBD_FILE}) { # Use the env var if it exists
|
||||
$theDbdFile = $ENV{CAPR_DBD_FILE};
|
||||
} # Otherwise use the default set above
|
||||
|
||||
parseDbd($theDbdFile);
|
||||
print "Using $theDbdFile\n\n";
|
||||
|
||||
# Print a list of record types
|
||||
if($opt_r) {
|
||||
print ("Record types defined in $theDbdFile\n");
|
||||
printList(0);
|
||||
exit;
|
||||
}
|
||||
|
||||
# Print the fields defined for given record
|
||||
if($opt_f) {
|
||||
printRecordList($opt_f);
|
||||
exit;
|
||||
}
|
||||
#my @data;
|
||||
#@data = EZCA::Get("S:BM:TMDispAI.VAL", "ezcaDouble",1);
|
||||
#if( $data[0] ) { die "ca get error on field --" };
|
||||
#print(" $data[1] ");
|
||||
#die "end";
|
||||
|
||||
# Do the business
|
||||
# Allow commas between arguments as in vxWorks dbpr
|
||||
die "$usage" unless defined $ARGV[0];
|
||||
$ARGV[0] =~ s/,/ /; # Get rid of pesky comma if it's there
|
||||
if($ARGV[0] =~ m/\s+\d/) { # If we replace comma with a space,
|
||||
($ARGV[0], $ARGV[1]) = split(/ /, $ARGV[0]); #split it
|
||||
}
|
||||
$ARGV[0] =~ s/\s+//; # Remove any spaces
|
||||
$ARGV[0] =~ s/\..*//; # Get rid of field name if it's there
|
||||
$ARGV[1] =~ s/\D//g; # Make sure we only use digits
|
||||
$ARGV[1] = $ARGV[1] || 0; # interest defaults to 0
|
||||
printRecord($ARGV[0], $ARGV[1]); # Do the do
|
||||
|
||||
|
||||
########## End of main ###########
|
||||
|
||||
|
||||
|
||||
# parseDbd
|
||||
# takes given dbd file and parses it to produce a hash table of record types
|
||||
# giving their fields, and for each field its interest level and data type
|
||||
# usage: void parseDbd("fileName");
|
||||
# Output is in the hash %record. This is a hash of (references to) another.
|
||||
# hash containing the fields of this record, as keys. The value keyed by
|
||||
# the field names are (references to) arrays. Each of these arrays contains
|
||||
# the interest level, data type and base of the field
|
||||
sub parseDbd {
|
||||
my $dbdFile = $_[0];
|
||||
my @dbd;
|
||||
my $length;
|
||||
my $level = 0;
|
||||
my $i = 0;
|
||||
my $isArecord = 0;
|
||||
my $isAfield;
|
||||
my $thisRecord;
|
||||
my $thisField;
|
||||
my $thisType;
|
||||
my %field = ();
|
||||
my @params = ();
|
||||
my $interest = 0;
|
||||
my $thisBase = "DECIMAL";
|
||||
my $item;
|
||||
my $newDevice;
|
||||
|
||||
open(DBD, "< $dbdFile") || die "Can't open dbd file $dbdFile --";
|
||||
@dbd = <DBD>;
|
||||
$length = @dbd;
|
||||
close(DBD) || die "Can't close $dbdFile --";
|
||||
|
||||
while ($i < $length) {
|
||||
$_ = $dbd[$i];
|
||||
chomp;
|
||||
print("line $i - level $level\n") if ($DEBUG);
|
||||
#$line = $dbd[$i] || die "Unexpected end of file: $dbdFile, line $.";
|
||||
if( m/recordtype/ ) {
|
||||
($level == 0) || die "dbd file format error in or before line $i --";
|
||||
m/\((.*)\)/; #get record type
|
||||
#@records = (@records, $1);
|
||||
$isArecord = 1;
|
||||
$thisRecord = $1;
|
||||
}
|
||||
if( m/field/ ) {
|
||||
($level == 1 && $isArecord) || die "dbd file format error in or before line $i --";
|
||||
m/\((.*),/; # get field name
|
||||
$thisField = $1;
|
||||
m/,(.*)\)/; # get field type
|
||||
$thisType = $1;
|
||||
$isAfield = 1;
|
||||
#print("$1 , line $i ");
|
||||
}
|
||||
if( m/interest/ ) {
|
||||
($level == 2 && $isAfield) || die "dbd file format error in or before line $i --";
|
||||
m/\((.*)\)/ ; # get interest level, default = 0
|
||||
$interest = $1;
|
||||
}
|
||||
if( m/base/ ) {
|
||||
($level == 2 && $isAfield) || die "dbd file format error in or before line $i --";
|
||||
m/\((.*)\)/ ; # get base, default = DECIMAL
|
||||
$thisBase = $1;
|
||||
}
|
||||
if( m/\{/ ) { $level++ };
|
||||
if( m/\}/ ) {
|
||||
if( $level == 2 && $isAfield) {
|
||||
$isAfield = 0;
|
||||
$params[$iIdx] = $interest;
|
||||
$params[$tIdx] = $thisType;
|
||||
$params[$bIdx] = $thisBase;
|
||||
$field{$thisField} = [@params];
|
||||
#print("interest $interest\n");
|
||||
$interest = 0; # set up default for next time
|
||||
$thisBase = "DECIMAL"; # set up default for next time
|
||||
}
|
||||
if( $level == 1 && $isArecord) {
|
||||
$isArecord = 0;
|
||||
$record{$thisRecord} = { %field };
|
||||
#print("record type $thisRecord ");
|
||||
#foreach $key (keys(%field)) {
|
||||
# print("Field $key - interest $field{$key}\n");
|
||||
#}
|
||||
%field = (); # set up for next time
|
||||
}
|
||||
$level--;
|
||||
}
|
||||
# Parse for record types with device support
|
||||
if( m/device/ ) {
|
||||
m/\((.*?),/;
|
||||
if(!exists($device{$1})) {
|
||||
# Use a hash to make a list of record types with device support
|
||||
$device{$1} = 1;
|
||||
}
|
||||
}
|
||||
$i++;
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
# Given a record name attempts to find the record and its type.
|
||||
# Usage: getRecType(recordName) - returns ($error, $recordType)
|
||||
sub getRecType {
|
||||
my $name = $_[0] . ".RTYP";
|
||||
my $type;
|
||||
my $data;
|
||||
|
||||
$data = caget($name);
|
||||
if ($data =~ m/Invalid channel name/) { die "Record \"$_[0]\" not found\n"; }
|
||||
chomp $data;
|
||||
$data =~ s/\s+//;
|
||||
#print("$name is a \"$data\"type\n");
|
||||
return($data);
|
||||
}
|
||||
|
||||
# Given the record type and the field returns the interest level, data type
|
||||
# and base for the field
|
||||
# Usage: ($dataType, $interest, $base) getFieldParams( $recType, $field)
|
||||
sub getFieldParams {
|
||||
my $recType = $_[0];
|
||||
my $field = $_[1];
|
||||
my ($fType, $fInterest, $fBase);
|
||||
|
||||
exists($fieldType{$record{$recType}{$field}[$tIdx]}) ||
|
||||
die "Field data type $field for $recType not found in dbd file --";
|
||||
exists($record{$recType}{$field}[$iIdx]) ||
|
||||
die "Interest level for $field in $recType not found in dbd file --";
|
||||
|
||||
$fType = $fieldType{$record{$recType}{$field}[$tIdx]};
|
||||
$fInterest = $record{$recType}{$field}[$iIdx];
|
||||
$fBase = $record{$recType}{$field}[$bIdx];
|
||||
#print("getFieldParams: $recType, $field, $fType, $fInterest\n");
|
||||
return($fType, $fInterest, $fBase);
|
||||
}
|
||||
|
||||
# Prints field name and data for given field. Formats output so
|
||||
# that fields align in to 4 columns. Tries to imitate dbpf format
|
||||
# Usage: printField( $fieldName, $data, $dataType, $base, $firstColumnPosn)
|
||||
sub printField {
|
||||
my $fieldName = $_[0];
|
||||
my $fieldData = $_[1];
|
||||
my $dataType = $_[2];
|
||||
my $base = $_[3]; # base to display numeric data in
|
||||
my $col = $_[4]; # first column to print in
|
||||
|
||||
my $screenWidth = 80;
|
||||
my ($outStr, $len, $wide, $pad, $field);
|
||||
|
||||
$field = $fieldName . ":";
|
||||
|
||||
if( $dataType eq "ezcaString" ) {
|
||||
$outStr = sprintf("%-5s %s", $field, $fieldData);
|
||||
} elsif ( $base eq "HEX" ) {
|
||||
$outStr = sprintf("%-5s %x", $field, $fieldData);
|
||||
} elsif ( $dataType eq "ezcaDouble" || $dataType eq "ezcaFloat" ) {
|
||||
$outStr = sprintf("%-5s %.8f", $field, $fieldData);
|
||||
} elsif ( $dataType eq "ezcaShort" ) {
|
||||
$outStr = sprintf("%-5s %d", $field, $fieldData);
|
||||
} elsif ( $dataType eq "ezcaChar" ) {
|
||||
$outStr = sprintf("%-5s %d", $field, ord($fieldData));
|
||||
}else {
|
||||
$outStr = sprintf("%-5s %d", $field, $fieldData);
|
||||
}
|
||||
|
||||
$len = length($outStr);
|
||||
if($len <= 20) { $wide = 20; }
|
||||
elsif( $len <= 40 ) { $wide = 40; }
|
||||
elsif( $len <= 60 ) { $wide = 60; }
|
||||
else { $wide = 80;}
|
||||
|
||||
$pad = $wide - $len;
|
||||
#print "outStr=$outStr .. pad=$pad\n";
|
||||
|
||||
if( $col + $wide > $screenWidth ) {
|
||||
print("\n");
|
||||
$col = 0;
|
||||
}
|
||||
|
||||
print sprintf("$outStr%*s",$pad," ");
|
||||
$col = $col + $wide;
|
||||
|
||||
return($col);
|
||||
}
|
||||
|
||||
# Given record name and interest level prints data from record fields
|
||||
# that are at or below the interest level specified.
|
||||
# Useage: printRecord( $recordName, $interestLevel)
|
||||
sub printRecord {
|
||||
my $name = $_[0];
|
||||
my $interest = $_[1];
|
||||
my ($error, $recType, $field, $fType, $fInterest, $data);
|
||||
my ($fToGet, $col, $base);
|
||||
#print("checking record $name, interest $interest\n");
|
||||
$col = 0;
|
||||
|
||||
$recType = getRecType($name);
|
||||
print("$name is record type $recType\n");
|
||||
exists($record{$recType}) || die "Record type $recType not found in dbd file --";
|
||||
|
||||
foreach $field (sort keys %{$record{$recType}}) {
|
||||
# Skip DTYP field if this rec type doesn't have device support defined
|
||||
if($field eq "DTYP" && !(exists($device{$recType}))) { next; }
|
||||
|
||||
($fType, $fInterest, $base) = getFieldParams($recType, $field);
|
||||
unless( $fType eq "ezcaNoAccess" ) {
|
||||
if( $interest >= $fInterest ) {
|
||||
$fToGet = $name . "." . $field;
|
||||
$data = caget($fToGet);
|
||||
chomp $data;
|
||||
#$data =~ s/\s+//;
|
||||
#if( $data[0] ) { die "ca get error on field $field --" };
|
||||
$col = printField($field, $data, $fType, $base, $col);
|
||||
#print ("fType = $fType $data[1]\n");
|
||||
}
|
||||
}
|
||||
}
|
||||
print("\n"); # Final line feed
|
||||
}
|
||||
|
||||
# Prints list of record types found in dbd file. If level > 0
|
||||
# then the fields of that record type, their interest levels and types are
|
||||
# also printed.
|
||||
# Diagnostic routine, usage: void printList(level);
|
||||
sub printList {
|
||||
my $level = $_[0];
|
||||
my ($rkey, $fkey);
|
||||
|
||||
foreach $rkey (sort keys(%record)) {
|
||||
print("$rkey\n");
|
||||
if($level > 0) {
|
||||
foreach $fkey (keys %{$record{$rkey}}) {
|
||||
print("\tField $fkey - interest $record{$rkey}{$fkey}[$iIdx] ");
|
||||
print("- type $record{$rkey}{$fkey}[$tIdx] ");
|
||||
print("- base $record{$rkey}{$fkey}[$bIdx]\n");
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
# Prints list of fields with interest levels for given record type
|
||||
# Diagnostic routine, usage: void printRecordList("recordType");
|
||||
sub printRecordList {
|
||||
my ($rkey, $fkey);
|
||||
my $type = $_[0];
|
||||
|
||||
if( exists($record{$type}) ) {
|
||||
print("Record type - $type\n");
|
||||
foreach $fkey (sort keys %{$record{$type}}) {
|
||||
printf("%-4s", $fkey);
|
||||
printf(" interest = $record{$type}{$fkey}[$iIdx]");
|
||||
printf(" type = %-12s ",$record{$type}{$fkey}[$tIdx]);
|
||||
print (" base = $record{$type}{$fkey}[$bIdx]\n");
|
||||
}
|
||||
}
|
||||
else {
|
||||
print("Record type $type not defined in dbd file $theDbdFile\n");
|
||||
}
|
||||
}
|
||||
|
||||
# variable to communicate callback handler result
|
||||
my $callback_data;
|
||||
my $incomplete;
|
||||
my $cadebug = 0;
|
||||
|
||||
# returns a terse result of a caget operation
|
||||
# using perlCA
|
||||
sub caget {
|
||||
my $name = $_[0];
|
||||
print $name . "\n" if $cadebug;
|
||||
show_call_stack() if $cadebug;
|
||||
my $channel = CA->new($name);
|
||||
my $wait = 1;
|
||||
eval { CA->pend_io($wait); };
|
||||
if ($@) {
|
||||
if ($@ =~ m/^ECA_TIMEOUT/) {
|
||||
my $err = "'" . $channel . "'";
|
||||
print "Channel connect timed out: $err not found.\n";
|
||||
}
|
||||
die $@;
|
||||
}
|
||||
my $type = $channel->field_type;
|
||||
my $count = $channel->element_count;
|
||||
$channel->get_callback(\&callback_handler, $type);
|
||||
$incomplete = 1;
|
||||
CA->pend_event(0.1) while ($incomplete);
|
||||
return $callback_data;
|
||||
}
|
||||
|
||||
sub callback_handler
|
||||
{
|
||||
my ($chan, $status, $data) = @_;
|
||||
die $status if $status;
|
||||
#display($chan, $type, $data);
|
||||
$callback_data = $data;
|
||||
$incomplete=0;
|
||||
}
|
||||
|
||||
sub show_call_stack {
|
||||
my ( $path, $line, $subr );
|
||||
my $max_depth = 30;
|
||||
my $i = 1;
|
||||
print("--- Begin stack trace ---\n");
|
||||
my @call_details;
|
||||
@call_details = caller($i);
|
||||
while ( (@call_details) && ($i<$max_depth) ) {
|
||||
print("$call_details[1] line $call_details[2] in function $call_details[3]\n");
|
||||
$i = $i +1;
|
||||
@call_details = caller($i);
|
||||
}
|
||||
print("--- End stack trace ---\n");
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user