NEWS #7

=== Release 0.5, December 1997

This was an informal release to Perforce, so James could see what it
was like.


=== Release 1.0, January 30, 1998

- Contains several bug fixes you really want if you intend to use this
  for real.

- Verified the results of a complete conversion of the NetApp Data ONTAP
  source base, by converting then doing file-by file comparisons of
  the results of checking trees out from both CVS and Perforce, across
  several branches.

- Added minimal user documentation in the README file.

- Added a minimal automated test script.

- Placed under the Artistic license.


=== Release 1.0.1, February 3, 1998

- Incorporate James Strickland's README edits

- make bin/* kxtext

- bin/* script now find perl via PATH (it must be perl5 or they
  will fail).

- MANIFEST doesn't try to check checksums on FreeBSD systems.

- Added $CO to the test/config template file for specifying where to
  find the RCS "co" command

- CHECKPOINT_INTERVAL defaults to "10" for running the test config,
  but "0" otherwise in the test/config template.


=== Release 1.0.2, February 9, 1998

- Incorporate README typos fixes from James Strickland

- tests/* executables find Perl via $PATH

- Remove dup entry for test/changes.good from MANIFEST

- Remove "." from $check_submits in bin/dochanges (the "." was there for
  debugging, and, in effect, disables all output error checks on submits!

=== Release 1.1, March 10, 1998

- Change conversion client name from "dochanges" to "cvs2p4"

- The CVS trunk is now created as "main" in the Perforce depot.

- Converts "$Header: //guest/richard_geiger/utils/cvs2p4_meta/NEWS#7 $" to "$Id: //guest/richard_geiger/utils/cvs2p4_meta/NEWS#7 $" for text files.

- Added support for "vendor" branches (RCS branch 1.1.1); becomes branch
  "import" in Perforce.

- Added SYM_SPECIAL support for selectively importing top level subdirs
  of a CVS module via symlinks.

- Optimizations for genchanges, which help *lots( with huge checkins
  of vendor-supplied branches (I.e., where there are many changes with
  large numbers of files).

- Fixed a bug in genchanges such that, for purposes of change grouping,
  all revisions were considered to have happened in the midnight hour
  (plus the actual number of minutes after the hour at which they actually
  occurred). Taken with the other grouping hueristics, this was pretty
  benign, I think, but it's fixed now.

- Fixed a bug wherein checkppoints were always trying to checkpoint the
  server at port 1666 on the host running the server.

- Also, *only* do checkpoints at NetApp, 'cause they depend p4d_admin!

- Take a checkpoint at the very end

- Now detects binary files in additional ways:
    perl -B operator
    known list of binary file extensions (hardcoded into dochanges for now)
  (This helps where binaries were checked in to CVS but nobody told CVS
  by selecting a binary keyword expansion mode).

- The diffs in the test stuff in 1.0.1 and 1.0.2 was inadvertantly disabled,
  so you could get false PASSes. Oops. Fixed.
  
=== Release 1.1.1, March 10, 1998

- Damn. The fix for multi-binary detection was not complete. It is now.
  Really. At least it had better be, I'm about to convert 25K p4 changes
  worth of our real source with this sucker...

=== Release 1.2

- Implement "dochanges -s <n>" to make the converter restartable. Yeah.
  
- Added a couple of test cases for branch tags sharing tip & trunk revs.

- Reimplement revmap to do per-revision stuff, to allow for merging.

- Fix ALLTHEWAYBACK to look for the earliest non-trunk revision actually
  selected by any branch tag - not just the earliest one where a real
  branch was found.

* This was the version used during our actual main CVS -> Perforce
  conversion on 3/20-3/21 1998.

=== Release 1.2.1

- Fix bug reported by Ed Rapoport, wherein:

    "There was a problem, however, in the case of cvs files where
    there were at least two branches with the second branch branching
    off the first, but not yet having a delta created in the cvs file
    (an "incipient" branch) The incipient branch name (br_orca_autocal
    in the example) got applied to its parent branch (br_orca_intermed
    in the example) and the parent branch name got applied to the
    child incipient branch in perforce."

=== Release 1.2.2

- Fix spurious error reports from the final diff (p4_describe vs
  p4_describe.good) when running the test cases with r98.2.  (This was
  caused by an apparent change, in which the diffs produced by the
  r97.3 "p4 describe" include Id keyword expansions, but the r98.2
  "p4 describe" does not).

=== Release 1.2.3

- Only attempt checkpointing if $CHECKPOINT_INTERVAL > 0. (So I can really
  turn it off in the config file!)

- Fix bug reported by Todd LaWall <toddl@SiTera.com>, wherein:

    a branched file, that has not been changed since the branch was
    made, is dropped from the branch, if $BRANCH_FLASH is not defined
    in the cvs2p4 config file. (bug in genmetadata).

=== Release 1.2.4

- In some circumstances, "p4 files" can return a different error message
  for "p4 files foo", when "foo" doesn't already exist in the depot.
  This change allows "dochanges" to work properly when this happens.
  (The precise circumstances in which the alternate message is returned
  are not yet well characterized).
   
=== Release 1.2.5 January 3, 2000.

- This adds adjustments for making cvs2p4 cope correctly with post-1999
  revisions in the RCS archives being converted, as supplied by David Simon
  of Goldman Sachs. (Thanks).


=== Release 1.2.6, January 18, 2000.

- The conversion scripts should now be able to cope with shell metacharacters
  (notable, ' ' and '$') in file names. This has not been thouroughly tested,
  so if you do have such files, please pay particualr scrutiny to them.


=== Release 1.2.7, April 5, 2000.

- A fix for a bug that caused genmetadata to croak when it
  encountered an RCS file with the default branch set, a la
  "rcs -b1 file,v"

=== Release 1.2.8, May 11, 2000.

- Handles cr/lf line endings in RCS archives.

=== Release 1.2.9, May 16, 2000.

- Allow "." in "id" symbols in RCS files. (Author names like "ralph.spoilsport")

=== Release 1.2.10, August 24, 2000.

- Tolerate "empty" RCS ,v files that can (reportedly) be created due to
  certain failing cvs operations. (These files will be ignored;
  previously, they would cause a fatal error in genmetadata).

=== Release 1.2.11, August 30, 2000.

- Skip (and note) ,v files with nonprintable characters in the fileame.

=== Release 1.2.12, September 11, 2000.

- Includes a bugfix, supplied by Thomas Quinot <quinot@inf.enst.fr>,
  to correctly handle the fact that RCS timestamps in repository files
  are in UTC; this was causing the timestamps to be off (by the
  difference between UTC and the Perforce server's local time) in
  converted depots, and, in some cases, cause outright
  failures. (I.e., when the UTC time had never existed in local time,
  due to daylight savings transitions.)

=== Release 1.2.13, September 12, 2000.

- Fixes a bug (resulting in bad date translations) introduced in the
  bugfix in 1.2.12. (Fix supplied by Thomas Quinot).

=== Release 1.2.14, November 20, 2000.

- Now performs metadata sort using a sort routine coded directly in perl,
  rather than by using the host system's "sort" command. (Differences
  in "sort" behavior from one host to another had been observed to
  cause irregularities).

=== Release 1.2.15, December 12, 2000.

- Now rejects filenames with illegal Perforce characters.
  (Fix supplied by Doug Quist).

=== Release 1.2.16, August 27, 2001

- Reflect my recent job (and email address) change

- Changes some of the default values (and names) of "config" file
  variables, to better match the environment I'm most like to work
  in now.

- Change all "p4" invokations to explicitly specify P4PORT, P4CLIENT and P4USER
  on the command line, so as not to be confused by and P4CONFIG files that
  might otherwise come into play.

=== Release 1.2.17, November 5, 2001

- Add an assertion for duplicate filenames d/f,v & d/Attic/f,v
  (This is not valid in a CVS repository!)

- Fix a bug where one-letter identifiers would cause an internal error in
  genmetadata. (Bug reported by Rich Ireland <rireland@enterasys.com>)

=== Release 1.3, December 26, 2001

- Add
    Change [0-9]+ renamed change [0-9]+ and submitted.\$
  to
    $check_submits in bin/dochanges

  and recognize the /^Change \d+ renamed change (\d+) and submitted\.$/
  form of the success message.

  Allows messages about change renumbering, so cvs2p4 can be used to
  import to an existing (nonvirginal!) server. Note that I *don't*
  claim this works completely now, but from my experience at Orca a
  year ago, this was the only bugaboo that bit. Note also that it's
  still the user's obligation to insure that there are no pathname
  overlaps with files that already exist. *That* is guaranteed not to
  work!

- First whack at support for creating Perforce labels from CVS tags.
  CVS *branch* tags are treated as branching information, (same as
  ever), but plain old tagged revisions get built into Perforce labels.
  The Perforce labels are created with the default (all-depot) view
  mapping.

  The mods involved changes to genmetadata (to remember the label
  info), dochanges (to make the RCS-to-Perforce revisions map (the
  existing revmap went the other way, so now we have both), and a new
  stage "dolabels" to create and populate the labels.

  Also sundry files for test cases & machinery.

  This implementation has not yet been tested a great deal with real
  road mileage, but, alas, this time I'm adding the feature for other,
  rather than as part of my own big conversion effort. So, we'll see.

- revmap output for the Perforce -> RCS direction is now cleaner.

- Finally, in putting in the support for labels, I was forced to look
  a lot closer at a lot more of the code than I have in some time.
  Needless to say, there's much I would do differently a second time
  around (and considering the tool to be more than a one-use thing to
  begin with!). Also, I'm now armed with enough knowledge, I think, to
  do a fast "direct to metadata" conversion...

=== Release 1.3.1, January 4, 2002

- Make dolabels ignore errors from "- label in sync.", which happens when
  the revision of a file in the label is a delete.

- Add the $IMPORTTAGSPOOF option (see test/config comments for details)

=== Release 1.3.2, February 5, 2002

- Reduce the memory footprint of bin/genmetadata. Previously, it was
  holding and sorting a complete copy of the metadata file "in-core"
  (as well as a copy of all of the RCS revision tags data!). This adds
  up quick, and some users saw genmetadata gobbling memory voraciously
  (and in some cases being running out, causing thrashing and/or process
  termination by the OS).

  genmetadata now keeps the metadata in a temp file, (sorting it in
  primary-key-sized chunks), and the revision tag information in a
  db-backed hash.

- Fix the label handling so that _all_ perforce revisions based on the
  labeled cvs revision are included in the generated
  labels. Previously, one of the N "correct" Perforce revisions were
  being tagged (effectively, at random). This stems from the fact that
  lazy copying and branching are explicit in Perforce, but implicit in
  CVS. I.e., the "#1" revision in a new Perforce branch _appears_ to
  be a separate entity (identical to the revision from which it was
  branched. This means that to use the converted labels, it will be up
  to _you_ to remember what labels go with what branches: but that's
  the way it is in CVS, too.
  
- A minor change in revmap to have a meaningful usage message, and
  properly handle the new rrevmap format.

- dochanges correctly deletes revmap database files for either
  *.db or *.pag/*.dat style databases.

=== Release 1.3.3, February 9, 2002

- Yet another fix for label handling, when files which had been
  moved to the Attic/ in CVS have been readded.

=== Release 2.0b1, April 15, 2002

- This release contains a major overhaul for performance. The
  "dochanges" and "dolabels" stages now directly generate
  checkpoint-format Perforce metadata, and then replay it with "p4d
  -jr" in order to create the Perforce metadata; also, the generated
  metadata causes the RCS archives in the CVS repository being
  converted to be used directly. Thus, the time-consuming "replay"
  of Perforce operations is avoided.

  "genmetadata" and "genchanges" remain unaltered in this release.

- The modification times on individual file revisions are now
  correctly preserved in the converted Perforce database (in previous
  versions, the changelist times correctly reflected the times of the
  original CVS submits, but individual file revisions times were set
  to the time of conversion).

=== Release 2.0b2, April 16, 2002

- Add $CHECKBIN config option to control whether RCS revisions are
  checked out to apply the perl -B operator as part of determining
  what revisions should be treated as binary in Perforce.

- Make the list of file extentions to be treated as binary a config
  file item ($EXT).

- Add $COPYIMPORT (and command line -c) to dochanges, to request that
  the CVS tree be copied into (rather than linked to from)
  $P4ROOT/depot/IMPORT.

=== Release 2.0b3, April 17, 2002

- dochanges now writes a record for the change counter! (Previously,
  it was left uninitialized, so the converted server would try to
  allocate change #1 again.)

=== Release 2.0b4, April 29, 2002

- dochanges now writes a record for the upgrade counter counter.
  it also checks the p4d version to verify we have 2002.1 or
  better, and does a p4d -xu if later than 2002.1.

=== Release 2.0b5, April 29, 2002

- handle binaries correctly. Since we're using imported RCS ,v
  files in the archive, we need to make binaries "binary +D"
  (or "xbinary +D"). Add a .gif to the tests to test this.

=== Release 2.0b6, May 1, 2002

- handle Attic files correctly - this required a change from the old
  version (and required touching genmetadata), since we now use the
  old RCS archive tree in place (or a copy thereof).

- dolabels now closes the labels journal file explicitly. Previously,
  the subsequent "p4d -jr" could see premature eof to due buffered
  data, and the last bunch of labels in the dblbls file would not
  be converted.

- fix a bug that broke conversion of files that had been deleted and
  then re-added in CVS; upon re-adding the file, it would be given
  revision #1 agian, instead of the next unused revision number.

- added a couple of test cases for these.

Many thanks to Fan Zhang of Numeritech for helping to wring out
these problems!


--------------------------------------
$Id: //guest/richard_geiger/utils/cvs2p4_meta/NEWS#7 $