Devel::Peek - A data debugging tool for the XS programmer

        use Devel::Peek;
        Dump( $a );
        Dump( $a, 5 );
        DumpArray( 5, $a, $b, ... );
        mstat "Point 5";

        use Devel::Peek ':opd=st';

Devel::Peek contains functions which allows raw Perl datatypes to be manipulated from a Perl script. This is used by those who do XS programming to check that the data they are sending from C to Perl looks as they think it should look. The trick, then, is to know what the raw datatype is supposed to look like when it gets to Perl. This document offers some tips and hints to describe good and bad raw data.

It is very possible that this document will fall far short of being useful to the casual reader. The reader is expected to understand the material in the first few sections of perlguts.

Devel::Peek supplies a CWDump() function which can dump a raw Perl datatype, and CWmstat("marker") function to report on memory usage (if perl is compiled with corresponding option). The function DeadCode() provides statistics on the data frozen into inactive CWCV. Devel::Peek also supplies CWSvREFCNT(), CWSvREFCNT_inc(), and CWSvREFCNT_dec() which can query, increment, and decrement reference counts on SVs. This document will take a passive, and safe, approach to data debugging and for that it will describe only the CWDump() function.

Function CWDumpArray() allows dumping of multiple values (useful when you need to analyze returns of functions).

The global variable CW$Devel::Peek::pv_limit can be set to limit the number of character printed in various string values. Setting it to 0 means no limit.

If CWuse Devel::Peek directive has a CW:opd=FLAGS argument, this switches on debugging of opcode dispatch. CWFLAGS should be a combination of CWs, CWt, and CWP (see -D flags in perlrun). CW:opd is a shortcut for CW:opd=st.

CWCvGV($cv) return one of the globs associated to a subroutine reference CW$cv.

debug_flags() returns a string representation of CW$^D (similar to what is allowed for -D flag). When called with a numeric argument, sets $^D to the corresponding value. When called with an argument of the form CW"flags-flags", set on/off bits of CW$^D corresponding to letters before/after CW-. (The returned value is for CW$^D before the modification.)

runops_debug() returns true if the current opcode dispatcher is the debugging one. When called with an argument, switches to debugging or non-debugging dispatcher depending on the argument (active for newly-entered subs/etc only). (The returned value is for the dispatcher before the modification.)

When perl is compiled with support for memory footprint debugging (default with Perl's malloc()), Devel::Peek provides an access to this API.

Use mstat() function to emit a memory state statistic to the terminal. For more information on the format of output of mstat() see "Using CW$ENV{PERL_DEBUG_MSTATS}" in perldebguts.

Three additional functions allow access to this statistic from Perl. First, use CWmstats_fillhash(%hash) to get the information contained in the output of mstat() into CW%hash. The field of this hash are

  minbucket nbuckets sbrk_good sbrk_slack sbrked_remains sbrks start_slack
  topbucket topbucket_ev topbucket_odd total total_chain total_sbrk totfree

Two additional fields CWfree, CWused contain array references which provide per-bucket count of free and used chunks. Two other fields CWmem_size, CWavailable_size contain array references which provide the information about the allocated size and usable size of chunks in each bucket. Again, see "Using CW$ENV{PERL_DEBUG_MSTATS}" in perldebguts for details.

Keep in mind that only the first several odd-numbered buckets are used, so the information on size of the odd-numbered buckets which are not used is probably meaningless.

The information in

 mem_size available_size minbucket nbuckets

is the property of a particular build of perl, and does not depend on the current process. If you do not provide the optional argument to the functions mstats_fillhash(), fill_mstats(), mstats2hash(), then the information in fields CWmem_size, CWavailable_size is not updated.

CWfill_mstats($buf) is a much cheaper call (both speedwise and memory-wise) which collects the statistic into CW$buf in machine-readable form. At a later moment you may need to call CWmstats2hash($buf, %hash) to use this information to fill CW%hash.

All three APIs CWfill_mstats($buf), CWmstats_fillhash(%hash), and CWmstats2hash($buf, %hash) are designed to allocate no memory if used the second time on the same CW$buf and/or CW%hash.

So, if you want to collect memory info in a cycle, you may call

  $#buf = 999;
  fill_mstats($_) for @buf;
  mstats_fillhash(%report, 1);          # Static info too

  foreach (@buf) {
    # Do something...
    fill_mstats $_;                     # Collect statistic
  }
  foreach (@buf) {
    mstats2hash($_, %report);           # Preserve static info
    # Do something with %report
  }

The following examples don't attempt to show everything as that would be a monumental task, and, frankly, we don't want this manpage to be an internals document for Perl. The examples do demonstrate some basics of the raw Perl datatypes, and should suffice to get most determined people on their way. There are no guidewires or safety nets, nor blazed trails, so be prepared to travel alone from this point and on and, if at all possible, don't fall into the quicksand (it's bad for business).

Oh, one final bit of advice: take perlguts with you. When you return we expect to see it well-thumbed.

Let's begin by looking a simple scalar which is holding a string.

        use Devel::Peek;
        $a = "hello";
        Dump $a;

The output:

        SV = PVIV(0)
          REFCNT = 1
          FLAGS = (POK,pPOK)
          IV = 0
          PV = 0xb2048 "hello"\0
          CUR = 5
          LEN = 6

This says CW$a is an SV, a scalar. The scalar is a PVIV, a string. Its reference count is 1. It has the CWPOK flag set, meaning its current PV field is valid. Because POK is set we look at the PV item to see what is in the scalar. The \0 at the end indicate that this PV is properly NUL-terminated. If the FLAGS had been IOK we would look at the IV item. CUR indicates the number of characters in the PV. LEN indicates the number of bytes requested for the PV (one more than CUR, in this case, because LEN includes an extra byte for the end-of-string marker).

If the scalar contains a number the raw SV will be leaner.

        use Devel::Peek;
        $a = 42;
        Dump $a;

The output:

        SV = IV(0)
          REFCNT = 1
          FLAGS = (IOK,pIOK)
          IV = 42

This says CW$a is an SV, a scalar. The scalar is an IV, a number. Its reference count is 1. It has the CWIOK flag set, meaning it is currently being evaluated as a number. Because IOK is set we look at the IV item to see what is in the scalar.

If the scalar from the previous example had an extra reference:

        use Devel::Peek;
        $a = 42;
        $b = \$a;
        Dump $a;

The output:

        SV = IV(0)
          REFCNT = 2
          FLAGS = (IOK,pIOK)
          IV = 42

Notice that this example differs from the previous example only in its reference count. Compare this to the next example, where we dump CW$b instead of CW$a.

This shows what a reference looks like when it references a simple scalar.

        use Devel::Peek;
        $a = 42;
        $b = \$a;
        Dump $b;

The output:

        SV = RV(0)
          REFCNT = 1
          FLAGS = (ROK)
          RV = 0xbab08
        SV = IV(0)
          REFCNT = 2
          FLAGS = (IOK,pIOK)
          IV = 42

Starting from the top, this says CW$b is an SV. The scalar is an RV, a reference. It has the CWROK flag set, meaning it is a reference. Because ROK is set we have an RV item rather than an IV or PV. Notice that Dump follows the reference and shows us what CW$b was referencing. We see the same CW$a that we found in the previous example.

Note that the value of CWRV coincides with the numbers we see when we stringify CW$b. The addresses inside RV() and IV() are addresses of CWX*** structure which holds the current state of an CWSV. This address may change during lifetime of an SV.

This shows what a reference to an array looks like.

        use Devel::Peek;
        $a = [42];
        Dump $a;

The output:

        SV = RV(0)
          REFCNT = 1
          FLAGS = (ROK)
          RV = 0xb2850
        SV = PVAV(0)
          REFCNT = 1
          FLAGS = ()
          IV = 0
          NV = 0
          ARRAY = 0xb2048
          ALLOC = 0xb2048
          FILL = 0
          MAX = 0
          ARYLEN = 0x0
          FLAGS = (REAL)
        Elt No. 0 0xb5658
        SV = IV(0)
          REFCNT = 1
          FLAGS = (IOK,pIOK)
          IV = 42

This says CW$a is an SV and that it is an RV. That RV points to another SV which is a PVAV, an array. The array has one element, element zero, which is another SV. The field CWFILL above indicates the last element in the array, similar to CW$#$a.

If CW$a pointed to an array of two elements then we would see the following.

        use Devel::Peek 'Dump';
        $a = [42,24];
        Dump $a;

The output:

        SV = RV(0)
          REFCNT = 1
          FLAGS = (ROK)
          RV = 0xb2850
        SV = PVAV(0)
          REFCNT = 1
          FLAGS = ()
          IV = 0
          NV = 0
          ARRAY = 0xb2048
          ALLOC = 0xb2048
          FILL = 0
          MAX = 0
          ARYLEN = 0x0
          FLAGS = (REAL)
        Elt No. 0  0xb5658
        SV = IV(0)
          REFCNT = 1
          FLAGS = (IOK,pIOK)
          IV = 42
        Elt No. 1  0xb5680
        SV = IV(0)
          REFCNT = 1
          FLAGS = (IOK,pIOK)
          IV = 24

Note that CWDump will not report all the elements in the array, only several first (depending on how deep it already went into the report tree).

The following shows the raw form of a reference to a hash.

        use Devel::Peek;
        $a = {hello=>42};
        Dump $a;

The output:

        SV = RV(0) at 0x816a618
          REFCNT = 1
          FLAGS = (ROK)
          RV = 0x814fc10
          SV = PVHV(0) at 0x814fc10
            REFCNT = 1
            FLAGS = (SHAREKEYS)
            IV = 1
            NV = 0
            ARRAY = 0x816c5b8  (0:7, 1:1)
            hash quality = 100.0%
            KEYS = 1
            FILL = 1
            MAX = 7
            RITER = -1
            EITER = 0x0
            Elt "hello" HASH = 0xc8fd181b
            SV = IV(0) at 0x814fcf4
              REFCNT = 1
              FLAGS = (IOK,pIOK)
              IV = 42

This shows CW$a is a reference pointing to an SV. That SV is a PVHV, a hash. Fields RITER and EITER are used by CWeach.

The quality of a hash is defined as the total number of comparisons needed to access every element once, relative to the expected number needed for a random hash. The value can go over 100%.

The total number of comparisons is equal to the sum of the squares of the number of entries in each bucket. For a random hash of CW<n> keys into CW<k> buckets, the expected value is:

                n + n(n-1)/2k

The CWDump() function, by default, dumps up to 4 elements from a toplevel array or hash. This number can be increased by supplying a second argument to the function.

        use Devel::Peek;
        $a = [10,11,12,13,14];
        Dump $a;

Notice that CWDump() prints only elements 10 through 13 in the above code. The following code will print all of the elements.

        use Devel::Peek 'Dump';
        $a = [10,11,12,13,14];
        Dump $a, 5;

This is what you really need to know as an XS programmer, of course. When an XSUB returns a pointer to a C structure that pointer is stored in an SV and a reference to that SV is placed on the XSUB stack. So the output from an XSUB which uses something like the T_PTROBJ map might look something like this:

        SV = RV(0)
          REFCNT = 1
          FLAGS = (ROK)
          RV = 0xb8ad8
        SV = PVMG(0)
          REFCNT = 1
          FLAGS = (OBJECT,IOK,pIOK)
          IV = 729160
          NV = 0
          PV = 0
          STASH = 0xc1d10       "CookBookB::Opaque"

This shows that we have an SV which is an RV. That RV points at another SV. In this case that second SV is a PVMG, a blessed scalar. Because it is blessed it has the CWOBJECT flag set. Note that an SV which holds a C pointer also has the CWIOK flag set. The CWSTASH is set to the package name which this SV was blessed into.

The output from an XSUB which uses something like the T_PTRREF map, which doesn't bless the object, might look something like this:

        SV = RV(0)
          REFCNT = 1
          FLAGS = (ROK)
          RV = 0xb8ad8
        SV = PVMG(0)
          REFCNT = 1
          FLAGS = (IOK,pIOK)
          IV = 729160
          NV = 0
          PV = 0

Looks like this:

        SV = RV(0)
          REFCNT = 1
          FLAGS = (TEMP,ROK)
          RV = 0x1d453c
        SV = PVCV(0)
          REFCNT = 2
          FLAGS = ()
          IV = 0
          NV = 0
          COMP_STASH = 0x31068  "main"
          START = 0xb20e0
          ROOT = 0xbece0
          XSUB = 0x0
          XSUBANY = 0
          GVGV::GV = 0x1d44e8   "MY" :: "top_targets"
          FILE = "(eval 5)"
          DEPTH = 0
          PADLIST = 0x1c9338

This shows that

•

the subroutine is not an XSUB (since CWSTART and CWROOT are non-zero, and CWXSUB is zero);

•

that it was compiled in the package CWmain;

•

under the name CWMY::top_targets;

•

inside a 5th eval in the program;

•

it is not currently executed (see CWDEPTH);

•

it has no prototype (CWPROTOTYPE field is missing).

CWDump, CWmstat, CWDeadCode, CWDumpArray, CWDumpWithOP and CWDumpProg, CWfill_mstats, CWmstats_fillhash, CWmstats2hash by default. Additionally available CWSvREFCNT, CWSvREFCNT_inc and CWSvREFCNT_dec.

Readers have been known to skip important parts of perlguts, causing much frustration for all.

Ilya Zakharevich ilya@math.ohio-state.edu

Copyright (c) 1995-98 Ilya Zakharevich. All rights reserved. This program is free software; you can redistribute it and/or modify it under the same terms as Perl itself.

Author of this software makes no claim whatsoever about suitability, reliability, edability, editability or usability of this product, and should not be kept liable for any damage resulting from the use of it. If you can use it, you are in luck, if not, I should not be kept responsible. Keep a handy copy of your backup tape at hand.

perlguts, and perlguts, again.