CHANGES for vcp - Version CoPy - Win32 fixes to - gentrevml - VCP::TestUtils (these break the Unix test suite in a really minor way; the t/90p4.t and t/90cvs.t test scripts exit with a non-zero result code; need to fix that) - VCP::*::cvs, VCP::*::p4. - VCP::Source::cvs: ignore waiting for and obtained lock messages. Suggested by Thomas Quinot - Added the VCPNODELETE option - VCP::Patch now closes it's patch file (this matters on Win32). - Added hexdump.pl, quite usefule for looking in to problems involving lineends, embedded ^Z, and embedded NULL (\000) chars. - Added lib/VCP/Newlines.pod - vcp will no longer die if it can't delete a working file. - VCP::Plugin now has an extra call to cwd() to get p4 to act right; this needs to be cleaned up a bit by directly setting the environment (I think) - the test suite now skips all cvs-related tests if it can't find cvs. 0.26 Tue Dec 18 05:21:09 EST 2001 - switched to use Text::Diff, no more Unix diff dependency. Added VCP::DiffFormat. - Added VCP:Patch. No more unix patch dependency 0.25 - Switched to an all-perl Diff::Unified instead of exec()ing the systems' diff. This is for NT as well as to avoid tripping over old, featureless diffs (which hasn't yet occurred, but...) - Built framework for VCP::Patch, including test suite. - VCP::Dest::*::handle_footer() now handle things ok when no revisions are transferred. - VCP::Dest::p4 now backfills the base rev when receiving an incremental update from a source other than VCP::Source::revml. VCP::Source::revml reaches out and does the backfill itself; other sources don't, so VCP::Dest::p4 now does a sync (like a CVS checkout) and compares it to the version the source provides. TODO: Do the same for V::D::cvs. Reported by Jerry Harmon . - VCP::Dest::p4 now inits itself properly when backfilling - Fix a few Perl nits ("$foo\_" => "${foo}_..."), patch submitted by Peter Prymmer . - VCP::Source::p4 now uses p4's -x option when getting all of the files in each label; no more huge command lines for that purpose. - VCP::Source::cvs tries to include revisions that -rTAG misses. If a file has been added (or re-added) since the tag, "cvs log" will report all of it's revisions. VCP::Source::cvs throws away any outside of the date range of the files that were tagged. This is not perfect, but it had a bug and is now a little less imperfect. - Totally rewrite the sorting code, add --sort-by option for the revml destination so that the test suites can get vcp to generate .revml files in a sorted order matching that now generated by gentrevml (see below). This may seem trivial, but it's the only way to assure consistent output in test suites, which is a prerequisite for distributing this package. Could add it to other repositories, but not sure if that's needed, leaving it out for now to keep complexity and documentation simple and to reduce the number of features that future version might need to support for backcompat. Can add it in as soon as a real world need shows up. - Added sort keys for (perldoc VCP::Dest for details): - change_id (happens to do a segment-wise compare just in case...) - time - comment - rev_id (using a segment-wise compare) - name (using a segment-wise relative path compare) - The rev_id and change_id sort keys both have several synonyms to make them mnemonic to people who happen to think of them differently. The defaul sort keys are change_id, time, comment. comment is useful for breaking apart mass commits in to changes. - Debug and optimize VCP::Dest::cvs. `cvs commit` is very slow if files have been added or altered, so VCP::Dest::cvs now waits as long as possible before doing one. It still needs to commit before (and after) deletes, among other things. The result is a big speedup even on the test suite, which has a more abusive (more adds and deletes than normal) sequence of operations. - VCP::Dest::cvs now does not prematurely delete working files. This hints that VCP::Rev needs to not delete any file that is pointed to by another VCP::Rev instance, since VCP::Rev objects may not be DESTROYed quickly enough and may be deleting a newer version of a file they point to. - Add File::Temp to PREREQ_PM in Makefile.PL. Older Perls don't come with it. - VCP::Source::cvs: beginnings of support for --cd to bypass the automatic workspace and use a user-supplied workspace. - Converted t/50revml.t to use VCP::TestUtils - Add revision number to the default sort comparison in VCP::Dest, remove name - gentrevml now produces the test .revml files in name,rev_id order to make it easier for test scripts to compare them with the revml output generated during testing. - added t/01sort.t to test sorting - revamped t/90* and tweaked VCP::TestUtils to have slightly simpler p4 and cvs mgmt apis - added VCP::TestUtils::get_vcp_output to encapsulate the task of extracting revml files in gentrevml order. - Improved VCP::TestUtils::launch_p4d so it will hunt for a random numbered free port up to three times so that multiple test suites can be run at once. - Added VCP::Debug::explicitly_debugging() predicate to allow for low-priority but noisy debugging statements. This predicate returns true only if the current package (or keywords passed in) exactly equal one of the debugging specs. The more general predicage debugging() lets debugging specs be regexps, this one doesn't, meaning that you can sometimes get more detail by spelling out debugging specs instead of using that old standby ".*". TODO: note on the list of possible specs that VCP::Debug can generate which words this is enabled for. - minor code consistency cleanups, include rename a few $r vars to $rev. 0.222 Tue Jul 31 08:49:00 EDT 2001 - VCP::Dest::cvs now properly handles $r->comment being undef. This can occur when there is no comment in the source repo (as opposed to a comment of ''). VCP::Source::p4 leaves $r->comment as undef when it can't parse some non-space chars out of the depot, which is a bit obnoxious, but VCP::Dest::cvs should handle it anyway. 0.221 Tue Jul 31 00:24:31 EDT 2001 - The test suite now sets uid, euid, and the LOGNAME env. var. when run logged in as (or "su -"ed to) root to avoid the dreaded cvs error "cannot commit files as 'root'". This imposes an unfortunate order on the non-cvs test suites in that cvs must be initted before any other repositories, since it changes uid and euid. If the other repos were initted before setting the uid and euid, then they might not have permission to access the files in question. NOTE: the only reason we have to set the uid is to keep perl from whining about -I when running setuid. I think cvs would be happy if we just tweaked euid, since that's all it checks. - The test suite now uses temporary dirs starting with "vcp$$\_" (eg "vcp2132_cvs_cvswork") to hold the repositories and workspaces needed. This makes it more obvious that they were created by vcp, in the event that something bombs out and they don't get cleaned up. - vcp now cleans up after itself a little better: the /tmp/vcp234 directories (which look like but are not related to the ones mentioned above used by the test suite) are now deleted when vcp exits normally. 0.22 - Sun Jul 22 23:52:46 EDT 2001 - VCP::Source::p4 now uses VCP::Utils::p4::parse_p4_repo_spec(). No functionality changes. - Created VCP::Utils::cvs to enable... - ...VCP::Source::cvs now checks out the appropriate module so the user does not need to have a checked-out version. - VCP::Dest::cvs no longer offers the --rev-root option, or any options at all for that matter. - The scheme and filespec are now stored in VCP::Plugin by VCP::Plugin::parse_repo_spec(), for ease of access and consistency. - Got rid of the overly funky VCP::Plugin::AUTOLOAD(). This used to emulate a class method that would run an external program of the same name as the method after that name was registered with the VCP::Plugin instance. Too baroque, now the VCP::Utils::foo classes provide however many invocation methods are needed to both VCP::Source::foo and VCP::Dest::foo. 0.21 - Fri Jul 20 23:08:43 EDT 2001 - Fixed VCP::Dest::p4; it was adding too many directories to a file, so files ended up with the first directory or two repeated. reported by david d zuhn - Cleaned up undefined var use in VCP::Utils::p4, reported by david d zuhn - Fixed VCP::Dest::revml's decision making so it will choose base64 encoding when faced with lots of NULs on input. Reported by david d zuhn . - Added a couple of compression thoughts for revml output to TODO. - Reworked VCP::Dest::revml's base64 encoding logic to be more efficient in time and in generated output. - Changed the test suite to use base64 for the "binary" file, and to use character codes from 0x00 to 0x07 in rotation to make sure binary data goes through ok. - refactored 90p4.t and 90cvs.t into VCP::TestUtils to make test maint. easier. Now you don't have to be clinically insane to work on the test suite. At least not all the time. - Added p4->revml, cvs->revml, cvs->p4->revml tests for the re-rooting feature. 0.2 - Thu Jul 19 00:18:34 EDT 2001 - *** VCP::Dest::p4 now uses a depot spec (//depot/...) instead of a file spec on the local filesystem. This differs from VCP::Source::p4 until I can upgrade it. The p4 where command is no longer used. - *** VCP::Dest::p4 now uses it's own client view pointing in to the temp directory. This should solve the link() failures when going from cvs to p4, since VCP::Source::cvs and VCP::Dest::p4 will both use the same temp dir. This should also make normal usage a bit easier, since the user need not create a client spec, etc. - Altered labelsync to use p4's "-x -", now the huge lists of labelled files that david d zuhn 's CVS repository has are passed to p4 on stdin instead of blowing up the command line. - VCP::Dest::p4 now passes absolute paths to the working files to p4 labelsync instead of paths relative to rev_root. Reported by david d zuhn . - Remove -f (force_missing) option from VCP::Source::cvs. It is now the default behavior. If need be, we can add an option to not force_missing in case extraneous files get brought along, but we'll wait for user demand. - Tweak deduce_rev_root algorithm to take the entire path given if no wildcards are present. - Added debug logging of command-line options, which is helpful when vcp is run by a script, such as our test scripts. Use the VCPDEBUG environment variable to enable this, set it to something like ".*" or "main". - Fixed documentation for --debug (the spec is required and it's not a shellish re, it's a perl5 re). - Fixed docs and comments for VCP::Source::cvs's -d options (was -D) - Deprecated the wacky VCP::Plugin::AUTOLOAD which was used to make things like $self->p4( [ options ], redirects ) work. So far, only VCP::Dest::p4 uses the new VCP::Plugin::run_safely() API, but the others will be upgraded. - Added a VCP::Utils::p4 to contain code that is used by VCP::Dest::p4 and which will soon be used by VCP::Source::p4. Expect a VCP::Utils::cvs to come along, too. 0.1 Wed Jul 4 00:27:35 EDT 2001 - Fix VCP::Dest::p4 to take the filespec from the p4: spec and use it as the rev_root. No --rev-root option at this time, not sure if it's needed. Reported by david d zuhn . 0.091 Thu Jun 7 08:32:15 EDT 2001 - Fix bug in CVS log file parsing that was causing it to think it was seeing all revisions as revisions of the first file it saw ("====... lines were being ignored) as the first file. Reported by Matthew Attaway. 0.09 Thu May 24 17:15:38 EDT 2001 - Debugged and beefed up the CVS log file parsing code. It can still be fooled by revision comments (the "message" field) things that mimic the output of `cvs log`, like a line beginning with the right number of "-" characters, but it's more robust. - Underscorify CVS tags. Reported by Matthew Attaway. - Only warn about undeleted files when in debugging mode, also reported by Matthew Attaway. The warning's there because we don't want to use too much disk space, but it gets in the way when an exception is thrown, since the END {...} handlers kick in before some (often many) the VCP::Rev objects are DESTROYed. 0.08 Wed May 23 10:18:28 EDT 2001 - Stop using p4 print -s, it puts a newline in every 4097 characters. Reported by Nick Ing-Simmons. - Require p4d >= 99.2 because we need filelog's -m option. Reported by Nick Ing-Simmons. Updated t/90cvs.t, t/90p4.t, and lib/VCP/TestUtils.pm to improve detection of stale p4ds. - Fix a "use strict" problem in lib/VCP/TestUtils.pm, reported by Nick Ing-Simmons. 0.07 Tue May 22 16:07:48 EDT 2001 - Minnor documentation changes. 0.066 Mon May 21 10:34:50 EDT 2001 - Remove PWD from the environment when calling the p4 command, so as to force it to use the cwd() - Improve logging of the directory the p4 command (and all commands) are being run in. - Many thanks to Matthew Attaway for help in finding and testing this one. 0.064 Tue May 8 11:26:11 EDT 2001 - Some versions of patch have problems handling embedded NUL (\0x00) characters. Fixed gentrevml and VCP::Dest::revml to not generate deltas for files containing them. If a file contains a NUL, it will sent in it's entirety, though it might or might not be base64 encoded. 0.063 Tue Apr 24 23:57:28 EDT 2001 - Fixed bug in parsing of repo specs (scheme:user:passwd@server:files) 0.062 Tue Apr 17 14:07:48 EDT 2001 - Added the requirement to install expat first to INSTALL - Commented out the call to "vcp help" to prevent perldoc from whining when run as root. Notified Brad Appleton, Pod::Usage's maintainer to see what might be done, but it's not a big deal for me unless an actual, real live user runs "vcp help" when logged in as root. - Converted shift to be CORE::shift in VCP::Revs, sinne perl 5.6.1 now complains that it had to resolve shift as CORE::shift due to the sub shift in there. 0.061 Sun Apr 15 15:56:32 EDT 2001 - Removed a couple of 5.6.0 dependancies - require perl 5.00503 in Makefile.PL, not sure how far back we could easily go. - Added logic to t/*.t to set -I more robustly by calling File::Spec->abs2rel on @INC and the path to bin/vcp and pass them on to child perls, so they could find their libraries. This fixed a nasty chicken-and-egg problem whereby you could only run the tests after installing. - t/90cvs.t and t/90p4.t use File::Spec->tmpdir for the repositories and workspaces they build. They also delete these temp dirs when done. - Made Makefile.PL require a minimum of XML::AutoWriter v.036 0.06 Wed Dec 20 23:19:15 EST 2000 - bin/vcp: Added --versions, which loads all modules and checks them for a $VERSION and print the results out. This should help with diagnosing out-of-sync modules. - Added $VERSION vars to a few modules :-). Forgot to increment any $VERSION strings. - VCP::Dest::cvs: The directory "deeply" was not being `cvs add`ed on paths like "a/deeply/nested/file", assuming "deeply" had no files in it. - VCP::Dest::revml: fixed a bug that was causing files with a lot of linefeeds to be emitted in base64 instead of deltaed. This means most text files. - Various minor cleanups of diagnostics and error messages, including exposing "Can't locate Foo.pm" when a VCP::Source or VCP::Dest module depends on a module that's not installed, as reported by Jeff Anton. 0.05 Mon Dec 18 07:27:53 EST 2000 - Use `p4 labels //...@label` command as per Rober Cowham's suggestion, with the '-s' flag recommended by Christopher Siewald and Amaury.FORGEOTDARC@atsm.fr. Though it's actually something like vcp: running /usr/bin/p4 -u safari -c safari -p localhost:5666 -s files //.../NtLkly //...@compiler_a3 //.../NtLkly //...@compiler_may3 and so //on //for 50 parameters to get the speed up. I use the //.../NtLkly "file" as //a separator between the lists of files in various //revisions. Hope nobody has any files named that :-). What I should do is choose a random label that doesn't occur in the labels list, I guess. - VCP::Source::revml and VCP::Dest::revml are now binary, control code, and "hibit ASCII" (I know, that's an oxymoron) clean. The , , and elements now escape anything other than tab, line feed, space, or printable chars (32 <= c <= ASCII 126) using a tag like ''. The test suite tests all this. Filenames should also be escaped this way, but I didn't get to that. - The decision whether to do deltas or encode the content in base64 is now based on how many characters would need to be escaped. - We now depend on the users' diff program to have a "-a" option to force it to diff even if the files look binary to it. I need to use Diff.pm and adapt it for use on binary data. - VCP::Dest::cvs now makes sure that no two consecutive revisions of the same file have the same mod_time. VCP::Source::p4 got so fast at pulling revisions from the repositories the test suite sets up that CVS was not noticing that files had changed. - VCP::Plugin now allows you to set a list of acceptable result codes, since we now use p4 in ways that make it return non-zero result codes. - VCP::Revs now croaks if you try to add two entries of the same VCP::Rev (ie matching filename and rev_id). - The tag is now limited to "text" or "binary", and is meant to pass that level of info between foreign repositories. - The on each file now carries the one line p4 description of the file so that p4->p4 transferes can pick out the more detailed info. VCP::Source::p4, VCP::Dest::p4 do this. - VCP::{Source,Dest}::{p4,cvs} now set binaryness on added files properly, I think. For p4->p4, the native p4 type is preserved. For CVS sources, seeing the keyword substitution flag 'o' or 'b' implies binaryness, for p4, seeing a filetype like qr/u?x?binary/ or qr/x?tempobj/ or "resource" implies binaryness (to non-p4 destinations). NOTE: Seeing a 'o' or 'b' in a CVS source only ends up setting the 'b' option on the destination. That should be ok for most uses, but we can make it smarter for cvs->cvs transfers if need be. 0.04 Tue Dec 12 00:15:57 EST 2000 - Reorg of VCP::Source::p4 - One large filelog command is run instead of many small ones. This takes advantage of the -m option to make sure enough changes are listed. Many extra revisions of most files are probably listed, but listing and ignoring them is quicker than spawning p4 over and over. Wish p4 filelog had a revision range... - it now doesn't suck the entire filelog output in to memory, it parses it line by line as it's emitted from the `p4 filelog` - `p4 print` is now used to print a bunch of files at once, using the header line to separate one file from the next, kind of like splitting a mime-encoded message. There's a very slight chance that it will misjudge the boundary between two files if a file happens to have a line that looks very much like the header line for the next file. This is pretty unlikely and I'll fix it if it crops up. I could batch them more, right now it never puts two revisions of the same filename in the same batch, for no really good reason. Another method might be to batch 25 or 50 revs each time. - it turns out there's a problem spawning multiple p4 commands at the same time against the same p4d (p4d is 99.2, FWIW). Or at least running large `p4 files ...` while there's a large `p4 filelog` still also running. - filelog lines beginning with "... ..." are now ignored. These are notifications of copy, branch, and integrate events that we don't yet do anything with. - deleted cur() and P4_CUR - deleted P4_IS_INCREMENTAL - Made an assertion in VCP::Dest::revml::handle_rev() a little clearer - Added some ok(1) calls to 90p4.t to make it easier to figure out which child process is whining or aborting - Made the message that's printed when a subcommand emits unexpected output say "stderr" instead of "stdout". - Cleaned up documentation for VC::Plugin::work_path(). 0.03 Sun Dec 10 13:14:07 EST 2000 - Cleaned up VCP::Source::p4 a bit. It doesn't whine as much now when it sees what it considers to be old news in the log file. - Added an easy way to monitor the commands being issued to a repository: simply add "::cmd" to the debug spec for that source: vcp -d Source::cvs::cvs vcp -d Dest::p4::p4 - The next step is to use the -m option to p4 filelog to speed things up. 0.02 Tue Dec 5 01:20:31 EST 2000 - VCP::Dest::p4 now does change number aggregation based on the comment field changing or whenever a new revision of a file with unsubmitted changes shows up on the input stream. Since revisions of files are normally sorted in time order, this should work in a number of cases. I'm sure we'll need to generalize it, perhaps with a time thresholding function. - t/90cvs.t now tests cvs->p4 replication. - VCP::Dest::p4 now doesn't try to `p4 submit` when no changes are pending. - VCP::Rev now prevents the same label from being applied twice to a revision. This was occuring because the "r_1"-style label that gets added to a target revision by VCP::Dest::p4 could duplicate a label "r_1" that happened to already be on a revision. - Added t/00rev.t, the beginnings of a test suite for VCP::Rev. - Tweaked bin/gentrevml to comment revisions with their change number instead of using a unique comment for every revision for non-p4 t/test-*-in-0.revml files. This was necessary to test cvs->p4 functionality. 0.01 Sat Dec 2 23:10:43 EST 2000 - First public release. Moderately functional alpha code. Supports p4, cvs, RevML. Expect APIs, UIs and enerthing else to change before things settle down.