D7net Mini Sh3LL v1
Current File : //proc/369106/../30/../79/../559/../12207/../12210/../../bin/splain |
#!/usr/bin/perl
eval 'exec /usr/bin/perl -S $0 ${1+"$@"}'
if $running_under_some_shell;
BEGIN { pop @INC if $INC[-1] eq '.' }
=head1 NAME
diagnostics, splain - produce verbose warning diagnostics
=head1 SYNOPSIS
Using the C<diagnostics> pragma:
use diagnostics;
use diagnostics -verbose;
enable diagnostics;
disable diagnostics;
Using the C<splain> standalone filter program:
perl program 2>diag.out
splain [-v] [-p] diag.out
Using diagnostics to get stack traces from a misbehaving script:
perl -Mdiagnostics=-traceonly my_script.pl
=head1 DESCRIPTION
=head2 The C<diagnostics> Pragma
This module extends the terse diagnostics normally emitted by both the
perl compiler and the perl interpreter (from running perl with a -w
switch or C<use warnings>), augmenting them with the more
explicative and endearing descriptions found in L<perldiag>. Like the
other pragmata, it affects the compilation phase of your program rather
than merely the execution phase.
To use in your program as a pragma, merely invoke
use diagnostics;
at the start (or near the start) of your program. (Note
that this I<does> enable perl's B<-w> flag.) Your whole
compilation will then be subject(ed :-) to the enhanced diagnostics.
These still go out B<STDERR>.
Due to the interaction between runtime and compiletime issues,
and because it's probably not a very good idea anyway,
you may not use C<no diagnostics> to turn them off at compiletime.
However, you may control their behaviour at runtime using the
disable() and enable() methods to turn them off and on respectively.
The B<-verbose> flag first prints out the L<perldiag> introduction before
any other diagnostics. The $diagnostics::PRETTY variable can generate nicer
escape sequences for pagers.
Warnings dispatched from perl itself (or more accurately, those that match
descriptions found in L<perldiag>) are only displayed once (no duplicate
descriptions). User code generated warnings a la warn() are unaffected,
allowing duplicate user messages to be displayed.
This module also adds a stack trace to the error message when perl dies.
This is useful for pinpointing what
caused the death. The B<-traceonly> (or
just B<-t>) flag turns off the explanations of warning messages leaving just
the stack traces. So if your script is dieing, run it again with
perl -Mdiagnostics=-traceonly my_bad_script
to see the call stack at the time of death. By supplying the B<-warntrace>
(or just B<-w>) flag, any warnings emitted will also come with a stack
trace.
=head2 The I<splain> Program
While apparently a whole nuther program, I<splain> is actually nothing
more than a link to the (executable) F<diagnostics.pm> module, as well as
a link to the F<diagnostics.pod> documentation. The B<-v> flag is like
the C<use diagnostics -verbose> directive.
The B<-p> flag is like the
$diagnostics::PRETTY variable. Since you're post-processing with
I<splain>, there's no sense in being able to enable() or disable() processing.
Output from I<splain> is directed to B<STDOUT>, unlike the pragma.
=head1 EXAMPLES
The following file is certain to trigger a few errors at both
runtime and compiletime:
use diagnostics;
print NOWHERE "nothing\n";
print STDERR "\n\tThis message should be unadorned.\n";
warn "\tThis is a user warning";
print "\nDIAGNOSTIC TESTER: Please enter a <CR> here: ";
my $a, $b = scalar <STDIN>;
print "\n";
print $x/$y;
If you prefer to run your program first and look at its problem
afterwards, do this:
perl -w test.pl 2>test.out
./splain < test.out
Note that this is not in general possible in shells of more dubious heritage,
as the theoretical
(perl -w test.pl >/dev/tty) >& test.out
./splain < test.out
Because you just moved the existing B<stdout> to somewhere else.
If you don't want to modify your source code, but still have on-the-fly
warnings, do this:
exec 3>&1; perl -w test.pl 2>&1 1>&3 3>&- | splain 1>&2 3>&-
Nifty, eh?
If you want to control warnings on the fly, do something like this.
Make sure you do the C<use> first, or you won't be able to get
at the enable() or disable() methods.
use diagnostics; # checks entire compilation phase
print "\ntime for 1st bogus diags: SQUAWKINGS\n";
print BOGUS1 'nada';
print "done with 1st bogus\n";
disable diagnostics; # only turns off runtime warnings
print "\ntime for 2nd bogus: (squelched)\n";
print BOGUS2 'nada';
print "done with 2nd bogus\n";
enable diagnostics; # turns back on runtime warnings
print "\ntime for 3rd bogus: SQUAWKINGS\n";
print BOGUS3 'nada';
print "done with 3rd bogus\n";
disable diagnostics;
print "\ntime for 4th bogus: (squelched)\n";
print BOGUS4 'nada';
print "done with 4th bogus\n";
=head1 INTERNALS
Diagnostic messages derive from the F<perldiag.pod> file when available at
runtime. Otherwise, they may be embedded in the file itself when the
splain package is built. See the F<Makefile> for details.
If an extant $SIG{__WARN__} handler is discovered, it will continue
to be honored, but only after the diagnostics::splainthis() function
(the module's $SIG{__WARN__} interceptor) has had its way with your
warnings.
There is a $diagnostics::DEBUG variable you may set if you're desperately
curious what sorts of things are being intercepted.
BEGIN { $diagnostics::DEBUG = 1 }
=head1 BUGS
Not being able to say "no diagnostics" is annoying, but may not be
insurmountable.
The C<-pretty> directive is called too late to affect matters.
You have to do this instead, and I<before> you load the module.
BEGIN { $diagnostics::PRETTY = 1 }
I could start up faster by delaying compilation until it should be
needed, but this gets a "panic: top_level" when using the pragma form
in Perl 5.001e.
While it's true that this documentation is somewhat subserious, if you use
a program named I<splain>, you should expect a bit of whimsy.
=head1 AUTHOR
Tom Christiansen <F<tchrist@mox.perl.com>>, 25 June 1995.
=cut
use strict;
use 5.009001;
use Carp;
$Carp::Internal{__PACKAGE__.""}++;
our $VERSION = '1.36';
our $DEBUG;
our $VERBOSE;
our $PRETTY;
our $TRACEONLY = 0;
our $WARNTRACE = 0;
use Config;
use Text::Tabs 'expand';
my $privlib = $Config{privlibexp};
if ($^O eq 'VMS') {
require VMS::Filespec;
$privlib = VMS::Filespec::unixify($privlib);
}
my @trypod = (
"$privlib/pod/perldiag.pod",
"$privlib/pods/perldiag.pod",
);
# handy for development testing of new warnings etc
unshift @trypod, "./pod/perldiag.pod" if -e "pod/perldiag.pod";
(my $PODFILE) = ((grep { -e } @trypod), $trypod[$#trypod])[0];
$DEBUG ||= 0;
local $| = 1;
local $_;
local $.;
my $standalone;
my(%HTML_2_Troff, %HTML_2_Latin_1, %HTML_2_ASCII_7);
CONFIG: {
our $opt_p = our $opt_d = our $opt_v = our $opt_f = '';
unless (caller) {
$standalone++;
require Getopt::Std;
Getopt::Std::getopts('pdvf:')
or die "Usage: $0 [-v] [-p] [-f splainpod]";
$PODFILE = $opt_f if $opt_f;
$DEBUG = 2 if $opt_d;
$VERBOSE = $opt_v;
$PRETTY = $opt_p;
}
if (open(POD_DIAG, '<', $PODFILE)) {
warn "Happy happy podfile from real $PODFILE\n" if $DEBUG;
last CONFIG;
}
if (caller) {
INCPATH: {
for my $file ( (map { "$_/".__PACKAGE__.".pm" } @INC), $0) {
warn "Checking $file\n" if $DEBUG;
if (open(POD_DIAG, '<', $file)) {
while (<POD_DIAG>) {
next unless
/^__END__\s*# wish diag dbase were more accessible/;
print STDERR "podfile is $file\n" if $DEBUG;
last INCPATH;
}
}
}
}
} else {
print STDERR "podfile is <DATA>\n" if $DEBUG;
*POD_DIAG = *main::DATA;
}
}
if (eof(POD_DIAG)) {
die "couldn't find diagnostic data in $PODFILE @INC $0";
}
%HTML_2_Troff = (
'amp' => '&', # ampersand
'lt' => '<', # left chevron, less-than
'gt' => '>', # right chevron, greater-than
'quot' => '"', # double quote
"Aacute" => "A\\*'", # capital A, acute accent
# etc
);
%HTML_2_Latin_1 = (
'amp' => '&', # ampersand
'lt' => '<', # left chevron, less-than
'gt' => '>', # right chevron, greater-than
'quot' => '"', # double quote
"Aacute" => "\xC1" # capital A, acute accent
# etc
);
%HTML_2_ASCII_7 = (
'amp' => '&', # ampersand
'lt' => '<', # left chevron, less-than
'gt' => '>', # right chevron, greater-than
'quot' => '"', # double quote
"Aacute" => "A" # capital A, acute accent
# etc
);
our %HTML_Escapes;
*HTML_Escapes = do {
if ($standalone) {
$PRETTY ? \%HTML_2_Latin_1 : \%HTML_2_ASCII_7;
} else {
\%HTML_2_Latin_1;
}
};
*THITHER = $standalone ? *STDOUT : *STDERR;
my %transfmt = ();
my $transmo = <<EOFUNC;
sub transmo {
#local \$^W = 0; # recursive warnings we do NOT need!
EOFUNC
my %msg;
my $over_level = 0; # We look only at =item lines at the first =over level
{
print STDERR "FINISHING COMPILATION for $_\n" if $DEBUG;
local $/ = '';
local $_;
my $header;
my @headers;
my $for_item;
my $seen_body;
while (<POD_DIAG>) {
sub _split_pod_link {
$_[0] =~ m'(?:([^|]*)\|)?([^/]*)(?:/("?)(.*)\3)?'s;
($1,$2,$4);
}
unescape();
if ($PRETTY) {
sub noop { return $_[0] } # spensive for a noop
sub bold { my $str =$_[0]; $str =~ s/(.)/$1\b$1/g; return $str; }
sub italic { my $str = $_[0]; $str =~ s/(.)/_\b$1/g; return $str; }
s/C<<< (.*?) >>>|C<< (.*?) >>|[BC]<(.*?)>/bold($+)/ges;
s/[IF]<(.*?)>/italic($1)/ges;
s/L<(.*?)>/
my($text,$page,$sect) = _split_pod_link($1);
defined $text
? $text
: defined $sect
? italic($sect) . ' in ' . italic($page)
: italic($page)
/ges;
s/S<(.*?)>/
$1
/ges;
} else {
s/C<<< (.*?) >>>|C<< (.*?) >>|[BC]<(.*?)>/$+/gs;
s/[IF]<(.*?)>/$1/gs;
s/L<(.*?)>/
my($text,$page,$sect) = _split_pod_link($1);
defined $text
? $text
: defined $sect
? qq '"$sect" in $page'
: $page
/ges;
s/S<(.*?)>/
$1
/ges;
}
unless (/^=/) {
if (defined $header) {
if ( $header eq 'DESCRIPTION' &&
( /Optional warnings are enabled/
|| /Some of these messages are generic./
) )
{
next;
}
$_ = expand $_;
s/^/ /gm;
$msg{$header} .= $_;
for my $h(@headers) { $msg{$h} .= $_ }
++$seen_body;
undef $for_item;
}
next;
}
# If we have not come across the body of the description yet, then
# the previous header needs to share the same description.
if ($seen_body) {
@headers = ();
}
else {
push @headers, $header if defined $header;
}
if ( ! s/=item (.*?)\s*\z//s || $over_level != 1) {
if ( s/=head1\sDESCRIPTION//) {
$msg{$header = 'DESCRIPTION'} = '';
undef $for_item;
}
elsif( s/^=for\s+diagnostics\s*\n(.*?)\s*\z// ) {
$for_item = $1;
}
elsif( /^=over\b/ ) {
$over_level++;
}
elsif( /^=back\b/ ) { # Stop processing body here
$over_level--;
if ($over_level == 0) {
undef $header;
undef $for_item;
$seen_body = 0;
next;
}
}
next;
}
if( $for_item ) { $header = $for_item; undef $for_item }
else {
$header = $1;
$header =~ s/\n/ /gs; # Allow multi-line headers
}
# strip formatting directives from =item line
$header =~ s/[A-Z]<(.*?)>/$1/g;
# Since we strip "(\.\s*)\n" when we search a warning, strip it here as well
$header =~ s/(\.\s*)?$//;
my @toks = split( /(%l?[dxX]|%[ucp]|%(?:\.\d+)?[fs])/, $header );
if (@toks > 1) {
my $conlen = 0;
for my $i (0..$#toks){
if( $i % 2 ){
if( $toks[$i] eq '%c' ){
$toks[$i] = '.';
} elsif( $toks[$i] =~ /^%(?:d|u)$/ ){
$toks[$i] = '\d+';
} elsif( $toks[$i] =~ '^%(?:s|.*f)$' ){
$toks[$i] = $i == $#toks ? '.*' : '.*?';
} elsif( $toks[$i] =~ '%.(\d+)s' ){
$toks[$i] = ".{$1}";
} elsif( $toks[$i] =~ '^%l*([pxX])$' ){
$toks[$i] = $1 eq 'X' ? '[\dA-F]+' : '[\da-f]+';
}
} elsif( length( $toks[$i] ) ){
$toks[$i] = quotemeta $toks[$i];
$conlen += length( $toks[$i] );
}
}
my $lhs = join( '', @toks );
$lhs =~ s/(\\\s)+/\\s+/g; # Replace lit space with multi-space match
$transfmt{$header}{pat} =
" s^\\s*$lhs\\s*\Q$header\Es\n\t&& return 1;\n";
$transfmt{$header}{len} = $conlen;
} else {
my $lhs = "\Q$header\E";
$lhs =~ s/(\\\s)+/\\s+/g; # Replace lit space with multi-space match
$transfmt{$header}{pat} =
" s^\\s*$lhs\\s*\Q$header\E\n\t && return 1;\n";
$transfmt{$header}{len} = length( $header );
}
print STDERR __PACKAGE__.": Duplicate entry: \"$header\"\n"
if $msg{$header};
$msg{$header} = '';
$seen_body = 0;
}
close POD_DIAG unless *main::DATA eq *POD_DIAG;
die "No diagnostics?" unless %msg;
# Apply patterns in order of decreasing sum of lengths of fixed parts
# Seems the best way of hitting the right one.
for my $hdr ( sort { $transfmt{$b}{len} <=> $transfmt{$a}{len} }
keys %transfmt ){
$transmo .= $transfmt{$hdr}{pat};
}
$transmo .= " return 0;\n}\n";
print STDERR $transmo if $DEBUG;
eval $transmo;
die $@ if $@;
}
if ($standalone) {
if (!@ARGV and -t STDIN) { print STDERR "$0: Reading from STDIN\n" }
while (defined (my $error = <>)) {
splainthis($error) || print THITHER $error;
}
exit;
}
my $olddie;
my $oldwarn;
sub import {
shift;
$^W = 1; # yup, clobbered the global variable;
# tough, if you want diags, you want diags.
return if defined $SIG{__WARN__} && ($SIG{__WARN__} eq \&warn_trap);
for (@_) {
/^-d(ebug)?$/ && do {
$DEBUG++;
next;
};
/^-v(erbose)?$/ && do {
$VERBOSE++;
next;
};
/^-p(retty)?$/ && do {
print STDERR "$0: I'm afraid it's too late for prettiness.\n";
$PRETTY++;
next;
};
# matches trace and traceonly for legacy doc mixup reasons
/^-t(race(only)?)?$/ && do {
$TRACEONLY++;
next;
};
/^-w(arntrace)?$/ && do {
$WARNTRACE++;
next;
};
warn "Unknown flag: $_";
}
$oldwarn = $SIG{__WARN__};
$olddie = $SIG{__DIE__};
$SIG{__WARN__} = \&warn_trap;
$SIG{__DIE__} = \&death_trap;
}
sub enable { &import }
sub disable {
shift;
return unless $SIG{__WARN__} eq \&warn_trap;
$SIG{__WARN__} = $oldwarn || '';
$SIG{__DIE__} = $olddie || '';
}
sub warn_trap {
my $warning = $_[0];
if (caller eq __PACKAGE__ or !splainthis($warning)) {
if ($WARNTRACE) {
print STDERR Carp::longmess($warning);
} else {
print STDERR $warning;
}
}
goto &$oldwarn if defined $oldwarn and $oldwarn and $oldwarn ne \&warn_trap;
};
sub death_trap {
my $exception = $_[0];
# See if we are coming from anywhere within an eval. If so we don't
# want to explain the exception because it's going to get caught.
my $in_eval = 0;
my $i = 0;
while (my $caller = (caller($i++))[3]) {
if ($caller eq '(eval)') {
$in_eval = 1;
last;
}
}
splainthis($exception) unless $in_eval;
if (caller eq __PACKAGE__) {
print STDERR "INTERNAL EXCEPTION: $exception";
}
&$olddie if defined $olddie and $olddie and $olddie ne \&death_trap;
return if $in_eval;
# We don't want to unset these if we're coming from an eval because
# then we've turned off diagnostics.
# Switch off our die/warn handlers so we don't wind up in our own
# traps.
$SIG{__DIE__} = $SIG{__WARN__} = '';
$exception =~ s/\n(?=.)/\n\t/gas;
die Carp::longmess("__diagnostics__")
=~ s/^__diagnostics__.*?line \d+\.?\n/
"Uncaught exception from user code:\n\t$exception"
/re;
# up we go; where we stop, nobody knows, but i think we die now
# but i'm deeply afraid of the &$olddie guy reraising and us getting
# into an indirect recursion loop
};
my %exact_duplicate;
my %old_diag;
my $count;
my $wantspace;
sub splainthis {
return 0 if $TRACEONLY;
for (my $tmp = shift) {
local $\;
local $!;
### &finish_compilation unless %msg;
s/(\.\s*)?\n+$//;
my $orig = $_;
# return unless defined;
# get rid of the where-are-we-in-input part
s/, <.*?> (?:line|chunk).*$//;
# Discard 1st " at <file> line <no>" and all text beyond
# but be aware of messages containing " at this-or-that"
my $real = 0;
my @secs = split( / at / );
return unless @secs;
$_ = $secs[0];
for my $i ( 1..$#secs ){
if( $secs[$i] =~ /.+? (?:line|chunk) \d+/ ){
$real = 1;
last;
} else {
$_ .= ' at ' . $secs[$i];
}
}
# remove parenthesis occurring at the end of some messages
s/^\((.*)\)$/$1/;
if ($exact_duplicate{$orig}++) {
return &transmo;
} else {
return 0 unless &transmo;
}
my $short = shorten($orig);
if ($old_diag{$_}) {
autodescribe();
print THITHER "$short (#$old_diag{$_})\n";
$wantspace = 1;
} elsif (!$msg{$_} && $orig =~ /\n./s) {
# A multiline message, like "Attempt to reload /
# Compilation failed"
my $found;
for (split /^/, $orig) {
splainthis($_) and $found = 1;
}
return $found;
} else {
autodescribe();
$old_diag{$_} = ++$count;
print THITHER "\n" if $wantspace;
$wantspace = 0;
print THITHER "$short (#$old_diag{$_})\n";
if ($msg{$_}) {
print THITHER $msg{$_};
} else {
if (0 and $standalone) {
print THITHER " **** Error #$old_diag{$_} ",
($real ? "is" : "appears to be"),
" an unknown diagnostic message.\n\n";
}
return 0;
}
}
return 1;
}
}
sub autodescribe {
if ($VERBOSE and not $count) {
print THITHER &{$PRETTY ? \&bold : \&noop}("DESCRIPTION OF DIAGNOSTICS"),
"\n$msg{DESCRIPTION}\n";
}
}
sub unescape {
s {
E<
( [A-Za-z]+ )
>
} {
do {
exists $HTML_Escapes{$1}
? do { $HTML_Escapes{$1} }
: do {
warn "Unknown escape: E<$1> in $_";
"E<$1>";
}
}
}egx;
}
sub shorten {
my $line = $_[0];
if (length($line) > 79 and index($line, "\n") == -1) {
my $space_place = rindex($line, ' ', 79);
if ($space_place != -1) {
substr($line, $space_place, 1) = "\n\t";
}
}
return $line;
}
1 unless $standalone; # or it'll complain about itself
__END__ # wish diag dbase were more accessible
AnonSec - 2021 | Recode By D7net