VCP::Maintenance - VCP code maintenance tips & tricks
VCPNODELETE=yes
bin/hexdump.pl
vcp revml: --dtd <dtd file name> --save_dtd <how>
As untarred, the distribution hierarchy looks like (some files not show for brevity):
VCP-x.yy/ +--- Makefile.PL ## Generates the Makefile +--- MANIFEST ## what files to ship, built with `make manifest` +--- MANIFEST.SKIP ## what filenames `make manifest` should ignore +--- CHANGES ## A detailed summart of all edits +--- TODO ## Things we can't get to quite yet +--- LICENSE +--- README +--- revml.dtd ## Defines legal RevML +--+ bin/ ## Executable files | +---- vcp ## The command line interface | +---- gentrevml ## builds RevML files for the test suite | +---- hexdump.pl ## dumps files in hex, a debugging aid | +--+ lib/ ## All modules that comprize VCP itself | +--- VCP.pm ## Drives the VCP process | +--- *.pod ## Supplemental documentation | +--- Plugin.pm ## Base class for all sources & destinations | | | +--- Source.pm ## Base class for all sources | +--+ Source/ | | +--- foo.pm ## A backend to read from repo. type "foo" | | | +--- Dest.pm ## Base class for all destinations | +--+ Dest/ | | +--- foo.pm ## A backend to write to repo. type "foo" | | | +--+ Utils/ | | +--- foo.pm ## Routines shared by Source/ and Dest/ foo.pm's | | | +--- Rev.pm ## VCP's concept of a revision | +--- Revs.pm ## A collection of VCP::Rev instances | | | +--- RevML/ ## Files defining RevML | +--+ t/ ## The test suite +--- *.t ## Test scripts +--- test-*.revml ## Fodder for test scripts
In addition, the following files are created:
| +--- vcp_html ## By running the `vcp html` command +--- pod2htm* ## Gunk left over from `vcp html` +--- blib/ ## By running `make`
perl -Ilib -cw %
Useful to map an editor key to this. Use whatever path for -I works given your cwd usage habit, and replace the "%" with whatever macro your editor replaces with the path to the current file.
make test TEST_FILES=t/90foo.t TEST_VERBOSE=1
perl -Ilib bin/vcp ...
vcp
manually without setting PERL5LIB or installing it.
export PERL5LIB=lib
vcp
and gentrevml
to be run from command line
make test
t/test-*.revml
if need be.
Some environment variables that are useful in debugging:
VCPNODELETE
yes
to tell vcp
not to delete it's working
directories. This allows you to take a look at them and see what
the source and dest command lines saw.
IPCRUNDEBUG
vcp
(via IPC::Run)
is treating it's subprocesses.
VCPDEBUG
VCP::*
emit debugging messages.
See also VCP::Debug for more details.
In addition, both CVS and Perforce backends pay attention to the relevant environment variables.
hexdump.pl
bin/hexdump.pl
is useful for debugging issues involving line endings,
embedded control code (esp. ^Z on Win32 and NULL on every platform, since most
C programs die when encountering NULL in a text file).
The t/*.t files in the source distribution contain a large number of
tests. The tests begin with a two digit number that orders them so that
more primitive tests run first, followed by higher level "end-to-end"
tests in t/90*.t
.
The more primitive tests are pretty standard fare for Perl test suites.
Where things get interesting is the t/90*.t
tests. These tests use the
command-line vcp
interface to move reasonably sized chunks of RevML
(up to about 65k currently; pretty small in comparison to "real"
exports) in to, out of, and between various repositories.
The RevML extracted from the repositories is compared to the RevML that was fed in to them, after sanitizing both of data that is known to not make it through ok.
The module VCP::TestUtils contains support routines for initializing repositories and shutting them down (when a server is needed). Existing repositories are not used in the test suite; they'd be useless and in danger if we did.
The file bin/gentrevml
is used to generate the RevML files
(t/test*.revml
) from scratch using hand written code instead of
VCP::Dest::revml in order to reduce the chances of a bug in
VCP::Dest::revml causing false positives in the test results.
It's highly likely that new back ends will require new test RevML files.
bin/gentrevml
should be hacked up to generate these and new rules
should be added to the Makefile
by editing Makefile.PL
.
The t/*.t scripts generally all follow the same structure.
First, they import all necessary modules and init themselves.
Then they build a
my @tests = ( sub { ... }, );
array that contain the actual tests. Each
test is a single anonymous sub {...}
, every once in a while a single
sub {...}
will contain two tests; it is preceded by an empty (or nearly
empty) sub {}
in these cases.
At the bottom, some final checks are run to see if all tests should be skipped, then
plan tests => scalar @tests;
emits how many tests are to be run, and
$_->() for @tests;
runs the tests.
This structure is used for several reasons:
1. No counting tests
scalar @tests
does this. Otherwise you end up having to remember to manually fudge
this every time a test is adding or removed.
2. Clear demarcation
my @tests =
line and are
pretty obvious.
3. Commenting out tests
); @tests = (
before the first test sub
you want to run, and a line like
sub { last },
after the last one. This is especially important
4. Test line numbers
TEST_VERBOSE=1
as shown above, will
report their line numbers. This is the fastest way to find out what's
failing. The list-of-subs approach is used even when @tests
could
actually contain just test vectors and the for @tests
loop could
check the vectors because we want line numbers to tell us where to look.
If we used test vectors, then the call to ok()
would report some line
number in the for @tests
loop, not the line number of the test.
5.
vcp
or IPC::Run debugging (see
VCP::Debug, too), you can put a
local $ENV{VCPDEBUG} = '.*';
or
local $ENV{IPCRUNDEBUG} = 1; ## 1..10
in the appropriate sub {...}
and it won't cause all the other tests
to spew a lot of noise.
In general, backends are created in pairs so that a t/90foo.t
can be
written to import the RevML files in to the backend repository, extract
new RevML from it, and compare the two.
A backend usually consists of 3 Perl classes: VCP::Source::foo,
VCP::Dest::foo, and VCP::Utils::foo. The reason for the lowercase
final name is so that the vcp
command can translate from the scheme
name (cvs:
) to a source or destination module name trivially.
The VCP::Source::foo generally contains a parser to extract metadata
from repository log reports and a routine to drive the copying process.
The VCP module (/lib/VCP.pm
)
There is one command used for maintainance: save_dtd
. This should not be
necessary in the normal course of events; it's only needed if the RevML DTD has
changed. Here's how it looks:
This outputs the DTD to stdout if <how>
is -
, or to a file named
like v0_28.pm
if <how>
looks like a version number (0.28 in this
case), or in a module named <how>
if <how>
contains '::'.
The generated file is placed in ./lib/RevML/Doctype/
or ./RevML/Doctype/
or ./
, whichever is found first. No directories will be created.
This does not update the VCP::Source::revml or
VCP::Dest::revml drivers to match the DTD, nor does
it affect bin/gentrevml
, which is used by the test suite to build the
RevML files used for testing.
Barrie Slaymaker <barries@slaysys.com>
Copyright (c) 2000, 2001, Perforce Software, Inc. All rights reserved.
See VCP::License (vcp help license
) for the terms of use.