	MOTIVATION (A): output from the configuration/build steps
	=========================================================

CpanTesters framework provides a wonderful system for finding out how your
module fares in close-to-realworld environments.  Unfortunately, several errors
in design make this framework less helpful in the most important case: when
your test suite fails.  The output provided by CpanTesters is optimized for
compactness, and some critically important pieces of build info are omitted
in the report.

In more details: the build of the module is broken into 3 parts:
   (*) Configuration (e.g., perl Makefile.PL);
   (*) Build proper (e.g., make);
   (*) Testing.
For many modules, the first two steps are practically a no-OP (at least, they
run exactly the same on all architectures).  It is (probably) because of this
fact that the output from the first two steps is NOT send to the author of the
module when these steps are successful, but the test suite fails.

On the other hand, if these two steps ARE non-trivial, the diagnostic output
from these steps may be NECESSARY to understand the reasons for the failure
of the test suite.  However, short of rebuilding the module, there is no
way that the test suite may recover this diagnostic output.

	MOTIVATION (B): debugging the problems in XS code
	=================================================

Additionally, when the errors happen in your XS code, and you cannot
immediately guess "why", the only recourse you have is to contact the persons
who runs the machine on the CpanTester framework, and ASK them to help in
debugging.

I should not explain how probable it is that such "remote debugging" would
succeed before the patience of the people on CpanTester machine would run out!

	PARTIAL SOLUTION: auto-debug-module.pl
	======================================

With this code,
  (*) one can rerun the build of the distribution (as a conditional part of
      the test suite) and see its output.
  (*) the rebuild may be performed in C-debugging-mode (with cc-like compilers)
  (*) scripts may be rerun under C debugger, so one would get a detailed report
      of what XS code was doing when it crashed.

(With the last part, one gets only a partial solution to the "remote debugging
problem" - you get information only in the case of crashes, and not in the case
of "wrong answers".  However, by forcing your XS code to (conditionally) crash
at a key point of the execution, one might get at least SOME leeway to remote
debugging.)

	DETAILS of auto-debug-module.pl (on example of Audio::FindChunks)
	=================================================================

(A) Running auto-debug-module.pl without arguments prints the usage info.

(B) For an example of the usage (via `@ARGV=...; do $debugger')
    see ../t/zzz_debug-crash.t.  With this usage, there is no extraneous output
    unless something fishy is going on.

(C) zzz_debug-crash.t should run AFTER Audio-FindChunks.t is run;
    this relies on the .t tests running in alphabetical order.
    However, this breaks on arm-freebsd; see
         http://www.cpantesters.org/cpan/report/ef2ee424-1c8e-11e6-b928-8293027c4940
            (bug in the shell's *.t ???)

(D) Audio-FindChunks.t uses the trick (near the start, before loading modules)
	# A poor-man indicator of a crash
	my $SELF;
	BEGIN { ($SELF = __FILE__) =~ s(.*[/\\])();
	   open CR, ">tst-run-$SELF" and close CR}	# touch a file
	END {unlink "tst-run-$SELF"}			# remove it - unless we crash
    to communicate the fact that it crashed to zzz_debug-crash.t, which checks
    for the presence of this file before starting its debugging session.

(E) As a result of (D),
     (*) when Audio-FindChunks.t crashes when running under automated testing,
	 auto-debug-module.pl is called with arguments
         	-q Audio::FindChunks t/Audio-FindChunks.t
         , otherwise only with arguments
         	-q Audio::FindChunks
         so that the rebuild would be enabled only if the module cannot be loaded
	 (as with a crash in XS_INIT) or if a crash was detected.

     (*) the output from the debugging rebuild would be seen on the
         CPANTester's report.

     (*) if crashes are detected, the state of the program during the crash
         is reported in as much detail as possible (with gdb - if present,
         otherwise with dbx or lldb).  Examples of output are in this directory.

(F) A more advanced example (with multiple failures possible) is in the test
    files of the Math::Pari distribution.

(G) Add the debug build directories to the 'clean' setting of the call to
    WriteMakefile(), as in
	clean => { FILES => 'dbg-bld dbg-bldO MORE_PATTERNS' }

  TO REMOTELY DEBUG YOUR DISTRIBUTION via CPAN-Testers
  ====================================================

(*) Copy the trick BEGIN/END chunks (from above) to the beginning of one of
    your .t test files;
(*) add zzz_debug-crash.t to your test suite (modifying the name of the module,
    and the filename of the test file);
(*) add utils/auto-debug-module.pl to your distribution;
(*) add
      tst-run-* had-tst-run-* dbg-bld dbg-bldO
    to the 'clean' key of WriteMakefile() of your Makefile.PL.
