mod_perl_traps - common/known mod_perl traps

In the CGI environment, the server starts a single external process (Perl interpreter) per HTTP request which runs single script in that process space. When the request is over, the process goes away everything is cleaned up and a fresh script is started for the next request. mod_perl brings Perl inside of the HTTP server not only for speedup of CGI scripts, but also for access to server functionality that CGI scripts do not and/or cannot have. Now that we're inside the server, each process will likely handle more than one Perl script and keep it compiled in memory for longer than a single HTTP request. This new location and longer lifetime of Perl execution brings with it some common traps. This document is here to tell you what they are and how to prevent them. The descriptions here are short, please consult the mod_perl FAQ for more detail. If you trip over something not documented here, please send a message to the mod_perl list.

•

Be sure to have read cgi_to_mod_perl

•

Scripts under Apache::Registry are not run in package main, they are run in a unique namespace based on the requested uri.

•

Apache::Registry scripts cannot contain __END__ or __DATA__ tokens

•

Output of CWsystem, CWexec and CWopen PIPE, "|program" calls will not be sent to the browser unless you Perl was configured with sfio.

•

Perl's exit() built-in function cannot be used in mod_perl scripts. The Apache::exit() function should be used instead. Apache::exit() automatically overrides the built-in exit() for Apache::Registry and Apache::PerlRun scripts.

•

Your script *will not* run from the command line if your script makes any direct calls to Apache->methods. See Apache::FakeRequest.

undefined subroutine &Apache::Registry::handler

Interaction with certain modules causes the shortcut configuration to break, if you see this message change your configuration from this:

 <Location /perl>
 PerlHandler Apache::Registry
 ...
 </Location>

To this:

 PerlModule Apache::Registry
 <Location /perl>
 PerlHandler Apache::Registry::handler
 ...
 </Location>

•

CGI.pm users must have version 2.39 of the package or higher, earlier versions will not work under mod_perl.

•

If you use the CWSendHeaders() function, be sure to call CW$req_obj->cgi->done when you are done with a request, just as you would under CGI::MiniSrv.

•

Files pulled in via CWuse or CWrequire statements are not automatically reloaded when changed on disk. See the Apache::StatINC or the Apache::Reload module to add this functionality.

Undefined subroutines

A common trap with required files may result in an error message similar to this in the error_log:

 [Thu Sep 11 11:03:06 1997] Undefined subroutine
 &Apache::ROOT::perl::test_2epl::some_function called at
 /opt/www/apache/perl/test.pl line 79.

As the above items explains, a file pulled in via CWrequire will only happen once per-process (unless CW%INC is modified). If the file does not contain a CWpackage declaration, the file's subroutines and variables will be created in the current package. Under CGI, this is commonly package CWmain. However, Apache::Registry scripts are compiled into a unique package name (base on the uri). So, if multiple scripts in the same process try to require the same file, which does not declare a package, only one script will actually be able to see the subroutines. The solution is to read perlmodlib, perlmod and related perl documentation and re-work your required file into a module which exports functions or defines a method interface. Or something more simple, along these lines:

 #required_file.pl
 package Test;

 sub some_function {...}

 ...

 __END__

Now, have your scripts say:

 require "required_file.pl";

 Test::some_function();

Undefined subroutine &Foo::Bar::handler called at PerlHandler subroutine `Foo::Bar' line 1.

You mistyped the module name in the 'package' line in your module. Because of eval context, you may see warnings with useless filename/line, example:

 Use of uninitialized value at (eval 80) line 12.
 Use of uninitialized value at (eval 80) line 43.
 Use of uninitialized value at (eval 80) line 44.

To track down where this eval is really happening, try using a __WARN__ handler to give you a stack trace:

 use Carp ();
 local $SIG{__WARN__} = \&Carp::cluck;

If something goes really wrong with your code, Perl may die with an Out of memory! message and or Callback called exit. A common cause of this are never-ending loops, deep recursion or calling an undefined subroutine. Here's one way to catch the problem: See Perl's INSTALL document for this item:

-DPERL_EMERGENCY_SBRK

If PERL_EMERGENCY_SBRK is defined, running out of memory need not be a fatal error: a memory pool can allocated by assigning to the special variable $^M. See perlvar(1) for more details. If you compile with that option and add 'use Apache::Debug level => 4;' to your PerlScript, it will allocate the $^M emergency pool and the CW$SIG{__DIE__} handler will call Carp::confess, giving you a stack trace which should reveal where the problem is. See the Apache::Resource module for prevention of spinning httpds.

•

If you wish to use a module that is normally linked static with your Perl, it must be listed in static_ext in Perl's Config.pm to be linked with httpd during the mod_perl build.

Can't load '$Config{sitearchexp}/auto/Foo/Foo.so' for module Foo...

When starting httpd some people have reported seeing an error along the lines of:

 [Thu Jul  9 17:33:42 1998] [error] Can't load
 '/usr/local/ap/lib/perl5/site_perl/sun4-solaris/auto/DBI/DBI.so' for
 module DBI: ld.so.1: src/httpd: fatal: relocation error: file
 /usr/local/ap/lib/perl5/site_perl/sun4-solaris/auto/DBI/DBI.so: symbol
 Perl_sv_undef: referenced symbol not found at
 /usr/local/ap/lib/perl5/sun4-solaris/5.00404/DynaLoader.pm line 166.

Or similar for the IO module or whatever dynamic module mod_perl tries to pull in first. The solution is to re-configure, re-build and re-install Perl and dynamic modules with the following flags when Configure asks for additional LD flags:

 -Xlinker --export-dynamic

or

 -Xlinker -E

This problem is only known to be caused by installing gnu ld under Solaris. Other known causes of this problem: OS distributions that ship with a (broken) binary Perl installation. The `perl' program and `libperl.a' library are somehow built with different binary compatiblity flags. The solution to these problems is to rebuild Perl and extension modules from a fresh source tree. Tip for running Perl's Configure script, use the `CW-des' flags to accepts defaults and `CW-D' flag to override certain attributes:

 % ./Configure -des -Dcc=gcc ... && make test && make install

Read Perl's INSTALL doc for more details.

mod_auth_dbm

If you are a user of mod_auth_dbm or mod_auth_db, you may need to edit Perl's CWConfig module. When Perl is configured it attempts to find libraries for ndbm, gdbm, db, etc., for the *DBM*_File modules. By default, these libraries are linked with Perl and remembered by the Config module. When mod_perl is configured with apache, the ExtUtils::Embed module returns these libraries to be linked with httpd so Perl extensions will work under mod_perl. However, the order in which these libraries are stored in Config.pm, may confuse CWmod_auth_db*. If CWmod_auth_db* does not work with mod_perl, take a look at this order with the following command:

 % perl -V:libs

If CW-lgdbm or CW-ldb is before CW-lndbm, example:

 libs='-lnet -lnsl_s -lgdbm -lndbm -ldb -ldld -lm -lc -lndir -lcrypt';

Edit Config.pm and move CW-lgdbm and CW-ldb to the end of the list. Here's how to find Config.pm:

 % perl -MConfig -e 'print "$Config{archlibexp}/Config.pm\n"'

Another solution for building Apache/mod_perl+mod_auth_dbm under Solaris is to remove the DBM and NDBM emulation from libgdbm.a. Seems Solaris already provides its own DBM and NDBM, and there's no reason to build GDBM with them (for us anyway). In our Makefile for GDBM, we changed

  OBJS = $(DBM_OF) $(NDBM_OF) $(GDBM_OF)

to

  OBJS = $(GDBM_OF)

Rebuild libgdbm, then Apache/mod_perl.

When using a regular expression that contains an interpolated Perl variable, if it is known that the variable (or variables) will not vary during the execution of the program, a standard optimization technique consists of adding the CWo modifier to the regexp pattern, to direct the compiler to build the internal table once, for the entire lifetime of the script, rather than every time the pattern is executed. Consider:

        my $pat = '^foo$'; # likely to be input from an HTML form field
        foreach( @list ) {
                print if /$pat/o;
        }

This is usually a big win in loops over lists, or when using CWgrep or CWmap.

In long-lived CWmod_perl scripts, however, this can pose a problem if the variable changes according to the invocation. The first invocation of a fresh httpd child will compile the table and perform the search correctly, however, all subsequent uses by the httpd child will continue to match the original pattern, regardless of the current contents of the Perl variables the pattern is dependent on. Your script will appear broken.

There are two solutions to this problem.

The first is to use CWeval q//, to force the code to be evaluated each time. Just make sure that the CWeval block covers the entire loop of processing, and not just the pattern match itself.

The above code fragment would be rewritten as:

        my $pat = '^foo$';
        eval q{
                foreach( @list ) {
                        print if /$pat/o;
                }
        }

Just saying

        eval q{ print if /$pat/o; };

is going to be a horribly expensive proposition.

You use this approach if you require more than one pattern match operator in a given section of code. If the section contains only one operator (be it an CWm// or CWs///), you can rely on the property of the null pattern, that reuses the last pattern seen. This leads to the second solution, which also eliminates the use of CWeval.

The above code fragment becomes:

        my $pat = '^foo$';
        "something" =~ /$pat/; # dummy match (MUST NOT FAIL!)
        foreach( @list ) {
                print if //;
        }

The only gotcha is that the dummy match that boots the regular expression engine must absolutely, positively succeed, otherwise the pattern will not be cached, and the // will match everything. If you can't count on fixed text to ensure the match succeeds, you have two possibilities.

If you can guaranteee that the pattern variable contains no meta-characters (things like CW*, CW+, CW^, CW$...), you can use the dummy match:

        "$pat" =~ /\Q$pat\E/; # guaranteed if no meta-characters present

If there is a possibility that the pattern can contain meta-characters, you should search for the pattern or the unsearchable CW\377 character as follows:

        "\377" =~ /$pat|^[\377]$/; # guarenteed if meta-characters present

        The Camel Book, 2nd edition, p. 538 (p. 356 in the 1st edition).

Doug MacEachern, with contributions from Jens Heunemann <heunemann2@janet.de>, David Landgren <david@landgren.net>, Mark Mills <mark@ntr.net>, Randal Schwartz <merlyn@stonehenge.com> and Ask Bjoern Hansen <ask@develooper.com>