cvs2cl bugs/requests

      The patch adds a '--rcs' option which tells it to parse the
      full_path_name based on the RCS file name rather than the Working file
      name.  The reason for this, is that when run on the server it uses
      rlog as it is operating on the repository directly, rather than in a
      working directory.  The Working file name is always listed without a
      path, however the RCS filename has the path.

      This was fixed by the patch for bug#36, which is really a duplicate of
      this bug.
    

      We have a great deal of CVS files with multi-line Commit-Comments akin
      to,

        + new feature BLAH added
        + source stripped from FOO
        + user FREAK accounted for

      etc.

      some start with '+', some with '-' others with '*' and yet more with
      nothing but new sentences.

      Here's my question, if I run cvs2cl.pl on what's noted above I get ONE
      paragraph with all the '+' jumbled inside something akin to

        * AUTHORS, TODO, dict_admin.cgi, dict_create.pl, dict_header.pl,
        dict_lib.pl, dict_main.cgi: + Added reconfirm password change (for
        admin) + Added hidden/setting attributes + Extracted all the embed
        files to _header + Added ability to create db.tables by a different
        user (if need be) + Corrected dict_main typos in

      Not very readable -- it would be best if it was something like this
      instead,

        * AUTHORS, TODO, dict_admin.cgi, dict_create.pl, dict_header.pl,
          dict_lib.pl, dict_main.cgi:
          + Added reconfirm password change (for admin)
          + Added hidden/setting attributes
          + Extracted all the embed files to _header
          + Added ability to create db.tables by a different user (if
            need be)
          + Corrected dict_main typos in

      Note continuation of line starts under first letter of line (ie. below
      'A' not below '*' or '+' to give a flowing feel to it).
    

      Every time I run cvs2cl entries are added to the ChangeLog file. There
      were no changes in the meantime and the added changes were already
      there. The added changes are all but the last log messages of today (4
      of the 5 logs). When running cvs2cl without the --utc everything works
      as expected.
    

      perl ~/cvs2cl.pl -t -W 0

      And was surprised to find

      "-W needs argument."
    

      The history is that I was looking to generate a log of changes since the
      instance of the tag, but the date system is unusable, since tags are not
      related to dates.  Using the -r option to cvs log helps, but not
      completely, since it includes the checkin that made up the tag.
    

      I have discovered a strange bug in cvs2cl which will only show up in a
      VERY specific situation. If you have a directory called '0' (The number
      0) anywhere in a directory tree where you are running cvs2cl, you will
      get into an infinite loop in the lines at about line 1208, 1209 (I may
      have screwed up the line numbers in my hacking).
    

      This patch adds a '--delta' flag to cvs2cl - this is a simplistic
      implementation that determines the last seen time for the two tags you
      give it and trims the ChangeLog output based on those dates.

      There's a possibility that this could yield too much or not enough log
      info if a subset of the tags have been moved. (Or, conceivably, if you
      hit the inherent race condition in a date-based algorithm, but thinking
      about that makes my head hurt. ;)

      One could write a dissertation on the problems with cvs' tagging scheme
      - files that were deleted at a specific revision just aren't tagged, so
      it's impossible to get complete information on the state of the
      repository at the time of any given release. For files that are missing
      one or both tags, you can only guess at the proper state based on the
      latest tagged date. This implementation uses that date as the criteria
      for *all* files, though.

      Nonetheless, it should work for 90/99% of cases...
    

      One problem is that, if you delete a subdirectory (delete all the
      files then cvs update -P), then cvs log stops looking at it.  So the
      changelog essentially rewrites history as if those files had never
      existed.
    

      List only the revisions occurring on a branch, not ancestors (or
      descendents?) to the branch.
    

      The patch adds the following command line option:

        -T, --tagdates             Show tags in output on their first occurance


      This allows for easy tracking of what changes occured before / after a
      certain tag was made. A lot more understandable than to have the tag
      listed on each file separately..


      Example:


      2002-04-25 16:31  flavio

              * templates/html/pptpconn.html: fixed reload of parent window on
              pptpconn.html

      2002-03-27 22:15  tag V2_4

      2002-03-27 22:15  henrik

              * templates/firewall: Fixup local firewalling for multilink
    

      Context :
      FileA present in branch and have Tag A, B and C
      FileB earlier present but now deleted (between B and C)

      cvs2cl.pl -l -rA:: is OK

      but cvs2cl.pl -l -C:: say me "cvs log: warning: no revision `B in
      CVSROOT" and the log appear in changeLog.
    

      Example: I have the following ChangeLog

      2002-05-24 19:21  hartge

              * lib/keenspot: updated wendy release cycle

      2002-05-24 19:11  hartge

              * lib/keenspot: update to wendy and cutewendy

      2002-05-24 18:13  elliotglaysher

              * ChangeLog, debian/changelog, lib/comicspage, lib/helen: Sizer's
              patches

      2002-05-24 17:57  hartge

              * MANIFEST: added bf

      2002-05-24 17:55  hartge

              * Makefile.PL, lib/bf: Makefile.PL, lib/bf: added module for German
              Userfriendly (Benutzerfreundlich)

      2002-04-25 23:26  hochstrb

              * ChangeLog, lib/animation_express: added support for mini-animated
              gifs at animiation express

      2002-04-15 14:27  hochstrb

              * Netcomics/Util.pm, Netcomics/HTML/Page.pm,
              Netcomics/HTML/Theme.pm, Netcomics/HTML/Themes/CleanRound.pm,
              doc/Comic_Module-HOWTO.html: support for embedded objects


      I would like it better to have an output like the follwing:

      2002-05-24  hartge

              * lib/keenspot: updated wendy release cycle
              * lib/keenspot: update to wendy and cutewendy
              * MANIFEST: added bf
              * Makefile.PL, lib/bf: Makefile.PL, lib/bf: added module for German
              Userfriendly (Benutzerfreundlich)

      2002-05-24  elliotglaysher

              * ChangeLog, debian/changelog, lib/comicspage, lib/helen: Sizer's
              patches

      2002-04-25  hochstrb

              * ChangeLog, lib/animation_express: added support for mini-animated
              gifs at animiation express
              * Netcomics/Util.pm, Netcomics/HTML/Page.pm,
              Netcomics/HTML/Theme.pm, Netcomics/HTML/Themes/CleanRound.pm,
              doc/Comic_Module-HOWTO.html: support for embedded objects

      As you see, for every day, there is a summary of every users checkins.
    

      I have problems with 'cat filelist | xargs cvs2cl.pl --accum' when we
      have a realy long file list!  cvs2cl will run 'cvs log
      -D>newesdateinchangelog ...' cmd and filter out needed changelogs! ck

      I prevent this bei using --stdin option, but i changed a view lines in
      the script which seemed to be wrong or not general:
    

      using: cvs2cl.pl -t --no-wrap -S  -r -b

      causes the branch label to overwrite the revision number in ChangeLog as
      follows:

      2002-08-12 13:40  jons

               * Mem_access.cpp (tags: JSRS3043) (JSRS3005B1.11):

      whereas using: cvs2cl.pl -t --no-wrap -S -r gets:

      2002-08-12 13:40  jons

               * Mem_access.cpp (1.11.2.11, JSRS3043):
    

      I tried cvs rlog foo | cvs2cl --stdin --stdout
      and found out there is a slight difference between
      log and rlog:

      RCS file: /usr/local/cvs/cvsroot/packages/web/help/platoon/support/Makefile,v
      Working file: packages/web/help/platoon/support/Makefile

      in the rlog case becomes:

      RCS file: /dsk/lv02/cvs/cvsroot/packages/web/help/platoon/support/Makefile,v

      without the "Working file" which causes major problems
      with the line
      if ((! (defined $file_full_path)) and /^Working file: (.*)/)

      RCS file: ... perhaps does not look nice but works.

      It would be nice with support for the rlog-format.
    

      I have made a modification to specify the langage in xml header
      <?xml version="1.0" encoding="latin1"?>
      Now, cvs2cl accepts the -e (or --encoding) command parameter following
      by the langage specification.
      cvs2cl -e latin1 --xml ...
    

      This issue occurs only under perl5.8.0 on some platforms (specifically
      RedHat 8)
    

      "binmode LOG_SOURCE" is deadly on Windows platforms running newer
      versions of ActiveState perl.

      Reported against 2.44
    

      And we are interested in finding files in their latest revisions only
      (not all the revisions) that have got a token in the message log when
      commited (using regexp option);
    

      get the ouput sorted by file (not by date)
    

      Old versions of cvs2cl (2.38) would print the revision numbers of a
      single file as x.y.z.a, but of multiple files on the same branch as
      x.y.z.[a,b.c].  The new version prints even the singleton revision as
      x.y.z.[a].  It would be good if there was at least an option to prevent
      this, as it makes lots of spurious diffs appear in ChangeLogs.
    

      older cvs2cl's did the clever `use /bin/sh then exec perl trick' which
      meant the perl path didn't matter, the new one hard wires in
      /usr/bin/perl (which is not always right on, say, older versions of
      Solaris).
    

      Emacs C-x 4 a doesn't generate the HH:MM in ChangeLogs:

      2003-01-15 19:51  Foo Bar <foo@bar.org>

      so I added a --no-times switch to cvs2cl to make it behave the same
      way.
    

      This patch allows cvs2cl to write its output chronologically, rather
      than in reverse-chronological order as is the default.
    

      We recently needed a --show-tag option which works contrary to the
      --ignore-tag option, i.e. which displays only revisions with a certain
      tag. We need this because of our product release policy where we tag a
      release and during the testing phase move tags on files which had bug
      fixes. So we want to see only revisions which have been changed after a
      certain date and which have a certain tag. With "cvs log" options alone
      this is not possible because new files (which do not have the tag at
      all) still get displayed.
    

      Quick'n'dirty patch that replaces entries such as:

              * reference/: libidn-docs.sgml, libidn-docs.sgml:

              Reorder.

      with

             * reference/: libidn-docs.sgml:

              Reorder.
    

      It would be nice to be able to disable the common_dir stuff, the
      colon in the line confuses Emacs' font-lock which makes it highlight
      badly.
   

      Perhaps the defaults should be set up to generate ChangeLog's according
      to the FSF Coding Standard.  I.e., make --no-times --no-common-dir the
      default.
   

      Hm, now that I see the entry, I see that it doesn't handle reverting
      back to "reference/libidn-docs.sgml" when there is only one filename:

             * reference/libidn-docs.sgml:

              Reorder.

      Argh.  But when there are more than one file, or if it is in the top
      directory, it works.  Maybe the second version isn't as bad.
    

      In the following output from

        bperl bin/cvs2cl -r --stdin --stdout --utc < data/1.log

      the version numbers for Persistent.java should be collected together.

      2000-04-14 18:03  tdraier

              * Persistent.java (1.34), Persistent.java (1.6.2.31): debugged
              setNew()

      2000-04-14 17:36  tdraier

              * Persistent.java (1.33), Persistent.java (1.6.2.30): debugged
              setNew()

      2000-04-13 18:06  tdraier

              * Persistent.java (1.32), Persistent.java (1.6.2.29): debugged the
              setRestoring state

    

      The attached patch add automatic email address creation and GECOS
      information listing support.  New commandline options are fully
      documented as well.

      Sample output (cvs2cl.pl -G -D orbis-terrarum.net):
      2003-02-22 20:23  Robin Johnson  <robbat2@orbis-terrarum.net>

              * LICENSE: License docs

      2003-02-22 20:23  Robin Johnson  <robbat2@orbis-terrarum.net>

              * Make.rules, Makefile, Makefile.dep, Makefile.inc: Make scripts
    

      In this case, I found that cvs2pl is even better than I had known from
      earlier use.  I was browsing the buglist when I saw the messages about
      --delta.  Cool!  Very handy.  I simply observe to you that the option
      isn't listed in the usage message.
    

      Occasionally in the changelog for the "wine" package you see an entry
      like this:

              * include/*.h, dlls/comctl32/*.c:
              Eric Kohl <ekohl@abo.rhein-zeitung.de>
              Added xxx_Unregister() functions to all common controls.

      I like the include/*.h, dlls/comctl32/*.c notation when several files in
      a directory have changed.
    

      Simple but debian-specific enhancement - upstream might try and be more
      portable by making the /etc/mailname lookup something that an argument
      can set, but it is actually right for debian in the simple case (and in
      the complex case, well, you've already got -U.)  Adds this as --passwd
      option.  Could also be extended to do NIS (via ypmatch) or LDAP
      (ldapsearch), though using nss instead would be better; could at least
      use some CPAN package for passwd access, even the getpwnam/getpwent
      builtin would do.  Mostly this is a 3 minute hack that was useful for
      me, but that's more expressive than just asking :-)
    

      Hi, this patch adds an --rcs option as suggested in the BUGS file, which
      allows cvs2cl to be used like

      cvs -d /foo rlog bar | cvs2cl --stdin --rcs /foo

      to generate ChangeLogs without needing a working copy.
    

      I modified the script to record the cvs state of each file in
      $qunk{'cvsstate'}. This is then used in xml output. A new --dead (or -D)
      switch reports when a commit has sent a file to the attic for normal
      ChangeLog output. Not sure if you'll like the output format, but that's
      easy enough to change. Modified script is attached.
    

      Extend cvs2cl.pl to include the ability to follow changes from the
      branch node forward when used in conjunction with the -F option.
    

      The first is an "--update" option, which is similar to, and based upon,
      "--accum", except that it lists only those files that have been changed
      since the last time it was run.  There is a quirk, however.  The last
      checkin listed is always a repeat.  I have had no problem with my
      userbase once I explained the quirk to them.  I have thought of a way to
      deal with this - if you think that "--update" should operate without
      the repeats, please let me know.
    

      The second is a "--summary" option, intended for reviewing checkins.
      This sorts the files in a checkin group into added, deleted, and changed
      files, and prints the change details for the changed files.  Note that
      Summary mode does not do text wrapping - this is deliberate, as human
      readbility is severely impaired by text wrapping in this output style -
      generally, a very wide window is needed.
    

      I like the --rcs option, but, under Windows, and I guess depending on
      how one defines/specifies the CVSROOT value, the "RCS file:" record may
      contain forward and backward slashes
    

      Also, I noticed the option '--chrono' is not displayed in the subroutine
      usage().
    

      The message from cvs2cl.pl --help says that --mailname is 'only used
      with --passwd'.  But why should this be?  Doesn't it make sense to
      specify a domain name for email addresses without looking up real names
      in the passwd file?

      OTOH, if you want to keep the current rule that it makes no sense to
      give --mailname and not --passwd, there should be an error message when
      this happens.
    

      Here's another patch: it removes some trailing whitespace on lines in
      the ChangeLog, which is quite annoying if you use an emacs mode which
      highlights trailing whitespace.  Hopefully it will apply even though the
      line numbers are fuzzy.
    

      I wasn't able to pinpoint any change that caused the following to
      happen, before:

      2003-02-19  Simon Josefsson <simon@josefsson.org>

              * profiles.c: Don't normalize or map ASCII space for XMPP nodeprep.
              Tiny patch from Travis Shirk <TShirk@jabber.com>.

      now:

      2003-02-19  Simon Josefsson <simon@josefsson.org>

              * profiles.c: Don't normalize or map ASCII space for XMPP nodeprep.
               Tiny patch from Travis Shirk <TShirk@jabber.com>.

      I.e., there is a spurious SPC after the \n\t.  The CVS log entry is
      "...prep.  Tiny...".
    

      when you have alot of files in one commit and get that list wrapped you
      end up with stg like this:


      ---snip---

      2003-05-12 20:07  marko

              * ExpEdit/EditLampe.php, ExpEdit/InputExperiment.php,
              ExpEdit/InputKat.php, ExpView/Experiment.php,
              ExpView/Kategorie.php, ExpView/Lampe.php:

              vergessene relative Pfade ergaenzt

      ---snip---

      Would it look nice to indent the wrapped lines according to the indent of
      the first line like:

      ---snip---

              * ExpEdit/EditLampe.php, ExpEdit/InputExperiment.php,
                ExpEdit/InputKat.php, ExpView/Experiment.php,
                ExpView/Kategorie.php, ExpView/Lampe.php:

      ---snip---
    

      We ran across an issue which, from what I could tell, wasn't addressed
      by any of the existing commandline switches on cvs2cl.pl.  I was
      wondering if you could provide feedback and/or accept a patch to correct
      it.

      The problem that we saw is that files that get added on a branch get a
      dead 1.1 revision on the trunk.  When you use cvs2cl to follow just the
      trunk, these files show as dead.  For various reasons we need to show
      dead files in the output.  The problem is that the files in question are
      not nor have ever been part of the trunk and thus in my opinion
      shouldn't be seen in the change log at all.

      To this end, I hacked in a new command line switch to hide these
      files. I've attached my modified 2.48 cvs2cl.pl.  All of my changes are
      marked with my initials: KML.  My perl skills aren't the best, so if
      there is a better way to do this, feel free.  Just let me know if you'll
      accept this change as it would be nice not to have to run a hacked
      version.
    

      The scenario is I need the file name, with a revision with a tag. Then I
      need any new files and revisions for the next tag.

      Example:

      Counter  Filename  Old Release(tag) /Revsion  New Release(tag)/Revision
      1        foo.c     Release1_1_8 revision 1.4  Release2_0_0 revision 1.5

      I'm trying to track specific revisions from one release to the next and
      print it out in the above format.
    

      This patch lets you use "[user@]host:/file" syntax to refer to remote
      user files in -U. I needed this to support "-U `cat
      CVS/Root`/CVSROOT/users", and think it might be slightly useful to
      others.
    

      It seems that the new --no-ancestors option works backward.  Running
       cvs2cl.pl -F branch
      causes only the changes on that branch (no ancestors) to be included in
      the change log

       cvs2cl.pl -F branch --no-ancestors
      causes the branch and all ancestors to be included in the log.

      The fix looks to be trivial; the one place where $No_Ancestors is used
      need to be inverted:

                 elsif ((length ($branch_number)) > (length ($revision))
                        and
                        !$No_Ancestors)
    

      The XML outputted by "--xml" has some bugs. For example, see this log
      entry:

      <entry>
      <date>2003-08-05</date>
      <weekday>Tuesday</weekday>
      <time>12:39</time>
      <author>Author</author>
      <file>
      <name>filename</name>
      <cvsstate>Exp</cvsstate>
      <revision>1.1</revision>
      </file>
      <msg>Commit message here.
      </msg>
      </entry>

      Note the additional newline at the end of <msg>...</msg>. A
      XML file is like a database, and all characters between tags are
      evaluated. A standard XML parser doesn't ignore newlines or blank spaces
      (except if configured to do so), because it can damage the data inside
      tags.
    

      A second problem is in the <changelog> tag. It is printed as:

      <changelog xmlns="http://www.red-bean.com/xmlns/cvs2cl/">

      But there is no namespace on http://www.red-bean.com/xmlns/cvs2cl/.
      Also, a root tag like this can confuse xsltproc (when using XSLT to
      transform the XML output), so it's best to change this line to

      <changelog>

      Or add a "--noxmlns" option to disable this if nedeed.

    

      However, I'd like to get daily CVS commit output sorted by person.
      Currently, when a person commits multiple changes in a day, they appear
      one by one, like this:

      * 2003-07-30 21:11 - Jeroen Coumans
      Evil update to all files - disable IE smarttags and the stupid image toolbar

      * 2003-07-30 18:50 - Jeroen Coumans
      Add the LFS RSS file. Until Rob whips up a perl script, this will have
      to be maintained manually from the lfs/news page.

      * 2003-07-30 18:29 - Jeroen Coumans
      Fix some validation issues and HTML abuse in the FAQ.

      * 2003-07-30 18:05 - Jeroen Coumans
      Updated the FAQ links to the mail archives.

      Instead, I'd like them all grouped together like this:

      * 2003-07-30 21:11 - Jeroen Coumans
      - Evil update to all files - disable IE smarttags and the stupid image
      toolbar
      - Add the LFS RSS file. Until Rob whips up a perl script, this will have
      to be maintained manually from the lfs/news page.
      - Fix some validation issues and HTML abuse in the FAQ.
      - Updated the FAQ links to the mail archives.
    

      Currently filters may be applied to the log output only via -l to get
      cvs log to do the work.  However, this relies on the semantics of cvs
      log, which is not always perfectly flexible or intuitive.  Therefore
      some filters could be added directly; e.g., filter by date, and/or
      author, or filename regex, or...
    

      Following GNU Chanlog standards, is there a mean to append the
      function/class name after the file name with cvs2cl :

      * sort.el (sort-subr): Return nil.

      If I my commit message is

      (sort-subr): Return nil.

      I get

      * sort.el: (sort-subr): Return nil.
    

      When I use --gecos option as "cvs2cl.pl --gecos" I get lots of
      messages as follows.

      Use of uninitialized value in substitution (s///) at ./cvs2cl.pl.org line 610,
      +<LOG_SOURCE> line 201.
    

      When using --rcs for piping a raw rcs log from cvs rlog, there's no
      'Working file:' line for each file in the input log. I noticed you've
      already got this case covered. However, if a file is in the attic, the
      changelog output from cvs2cl has .../Attic/... in the filename. All I
      did was add the following line to zap Attic in filenames

        $new_full_path =~ s/Attic\///;

      This way, you get the same output as when cvs2cl runs without --rcs.
    

      I don't use RCS either, just CVS. It's easy enough to create a raw rcs
      log:

         cvs -d/your/repository rlog < module > rawlog

      then shove it into cvs2cl

         cvs2cl.pl --stdin --rcs /your/repository/module < rawlog

      I do it like this so that I can run cvs2cl several times to create
      changelogs for individual cvs branches with --follow, while only
      generating the log data once.
    

      I've found another tiny bug. This time flagged up by spaces in file
      names (who puts file names with spaces in CVS?). Here's the culprit:

          unless ($Hide_Branch_Additions and $msg_txt =~ /file \S+ was
      initially added on branch \S+./) {

      It doesn't (appear to) account for a filename with a space. I must
      confess I don't actually know what \S matches. Anyway, replacing '\S+'
      with '.*' seems to fix things for me. I'm sure you can come up with a
      more robust regexp.
    

      For any kind of XML post-processing, an ISO-formatted date is fairly
      useful (and also clarifies any ambiguity over local time or UTC). This
      patch adds one in, as an 'isoDate' element.
    

      I would find it very useful to give sample valid date formats in the
      usage documentation, since it is really a pain to find out the correct
      one to launch cvs2cl.pl for a given date range (due to the internal
      escaping of parameters probably)
    

      I am interested in the feature of displaying in the Changelog the tag
      names used on a repository, chronologically.  The idea is to be able to
      use tha tags extracted from 'cvs status -v' to mark the point in time
      where a tag was applied, so a Changelog will implicitly better separate
      changes between different tags (which could be used to mark official
      releases).
    

      I regularly use cvs2cl.pl to produce a ChangeLog for a project of mine
      and it's a great script, the only thing i'm missing is the possibility
      to have the date +listed only once, so that if I use "--no-times", all
      the items from that date be listed +under the same date.
    

      I wanted to use cvs2cl but only to collect cvs log messages without
      processing them by any other means. I found --summary, which does
      basically what I want except that it prints info on the files and
      indents the log messages by a tab.

      The first problem can be solved by allowing the use of --hide-filenames
      together with --summary. Well, it degrades --summary to almost nothing,
      but the user should know what he does...

      To solve my second problem I added a new command line option
      --no-indent, which, if used, disables prepending tab characters to the
      lines.
    

      cvs2cl has no manpage, and this really should change.  I took the
      liberty of converting the output of cvs2cl --help to a manual page.  You
      can find it at

      https://cvs.noreply.org/cgi-bin/viewcvs.cgi/cvs2cl/debian/cvs2cl.1?view=markup

      Maybe you find it useful and can include it in cvs2cl's repository.
    

      I have called the script with the following parameters:

      cvs2cl.pl --FSF --utc --no-indent --no-wrap -S -U user.db
      --hide-filenames --window 60 -F B_PHPTODO_R1_00_KAMA

      I got the following output:
      -----------------------------
      2004-01-08  Karl-Heinz Marbaise <karl-heinz@minkenberg-medien.de>

              * - ChangeLog hinzu gekommen
      - Release 0.61.26

      2004-01-08  Karl-Heinz Marbaise <karl-heinz@minkenberg-medien.de>

              * - Branch Wechsel

      2004-01-07  Karl-Heinz Marbaise <karl-heinz@minkenberg-medien.de>

              * - PHPDocumentor Eintr?ge ?berarbeitet
        - Einordnung in Lib / PTL etc.
      - Release 0.61.25
      ------------------------------
    

      Just make sure you have some that are long enough to need to be wrapped.

      Run this:

      cvs2cl.pl --summary --fsf

      It doesn't wrap the lines, which I feel it should. If you don't want
      wrapping, I figured you'd specify --no-wrap. If you did want wrapping,
      under the old code, I don't see any way to make that happen.
    

      $ cvs2cl -b -t --stdout
      Can't use an undefined value as an ARRAY reference at /usr/bin/cvs2cl.orig line 475.
    

      For any authors whose gecos entry lacks a comma, cvs2cl warns of

      Exiting subroutine via next at /usr/users/tst_s000/cvs2cl.pl line 2033, <LOG_SOURCE> line 303.

      and fails.  For auhors missing in gecos field, cvs2cl errors complaining
      of trying to call '->gecos' on an undef.
    

      with cvs2cl.pl 2.52 I get lots of the following errors
      when trying to use --follow BRANCH:

      Use of uninitialized value in substr at /home/js/src/other/cvs2cl/cvs2cl.pl line 1987,
      +<LOG_SOURCE> line 784198.
      Use of uninitialized value in length at /home/js/src/other/cvs2cl/cvs2cl.pl line 1987,
      +<LOG_SOURCE> line 784198.
    

      Current cvs2cl doesn't exits with error status if cvs fails:

      $ cvs2cl; echo $?
      cvs log: No CVSROOT specified!  Please use the `-d' option
      cvs [log aborted]: or set the CVSROOT environment variable.
      0
      $

      Thus, I can't detect reliably if it fails from my scripts. Note that cvs
      does exit with nonzero status on error:

      $ cvs log; echo $?
      cvs log: No CVSROOT specified!  Please use the `-d' option
      cvs [log aborted]: or set the CVSROOT environment variable.
      1
      $

      I browsed the source, but I don't know perl, so I don't know how to
      detect the exit status of a piped comand. Perhaps it could be done at
      line 649, before the "close (LOG_SOURCE);".
    

      --delta regex for tags should include [,]
    

      I've attached a simple patch which adds a new command line flag to
      cvs2cl, "--lines-modified". This instructs cvs2cl to emit information
      (in the XML output mode only) describing the # of lines added and
      removed for a given commit.
    

      Option required that acts like --follow but ignores sub-branches.
    

      Hi there,

      CVS 1.12.9 uses a new date format in its log output. I had to patch my
      copy of cvs2cl.pl (2.55) to get it working again. I have attached the
      patch in case you're interested.
    

      Here is the output of cvs2cv -d:
      rzyj@rzyj:~/projekty/speagram$ cvs2cl -d
      cvs log: Logging .
      cvs log: Logging src
      Undefined subroutine &CVS::Utils::ChangeLog::FileEntry::fileparse called at /usr/bin/cvs2cl line 1799, <LOG_SOURCE> line 19.                               rzyj@rzyj:~/projekty/speagram$ cvs [log aborted]: received broken pipe signal
      Write failed flushing stdout buffer.
      write stdout: Broken pipe
    

      I'm currently installing cvs2cl with a couple of other programs related
      to cvs in our research group here at the Met Office. I would like to use
      cvs2cl with it's --gecos option to have nicer ChangeLog with the real
      names of the programmers.  This did work nice in an older version of
      cvs2cl, but since I downloaded the recent version (2.55) from your web
      page, I get the following error message:

        Can't call method "gecos" without a package or object reference at
        /home/fr1210/frcm/bin/cvs2cl line 2212, <LOG_SOURCE> line 14.

      I believe this is due to an error within the
      read_date_author_and_state() function. I have attached a patch that
      seems to fix the problem for me (but I'm certainly not a born perl
      programmer, so there will be better ways to do that).
    

      Add --xml-header option akin to --header, but inserting it's contents
      within the XML header area in XML mode.
    

      Unfortunately Windows don't understand command arguments enclosed in
      single quotes as cvs2cl.pl v2.57 does. I offer following fix at line
      #2608 in parse_options():

       #push @log_source_command, map "'$_'", @ARGV;
       push @log_source_command, map "\"$_\"", @ARGV;

      I believe double-quotes enclosement should work for Unix also.
    

      The manual page is quite clear on this:

             -I REGEXP, --ignore REGEXP
                 Ignore files whose names match REGEXP.  This option may be
                 used multiple times.

      Although there is no REGEXP asked but the full filename path, according
      to this code snippet from cvs2cl in line 2128:

          if ( grep index($path, $_) > -1, @Ignore_Files ) {
      According to "perldoc -f grep" it works "grep EXPR,LIST", $path is the
      path including the filename within the repository and @Ignore_Files is
      filled with the (multiple) --ignore value(s).

      So, I guess this could either need a code change (I would truly
      appreciate it, because real REGEXP would be very very neat in here!) or
      at least a documentation change (then you might consider leaving it open
      as wishlist, pretty please).
    

      If you look in the subroutine _revision_is_wanted, and search for the
      comment "Trivial case", then just below that there is a conditional that
      seems wrong to me.

      # Trivial case: is this revision on the branch?  (Compare this way to
      # avoid regexps that screw up Emacs indentation, argh.)
      if ( substr($revision, 0, (length($branch_number) + 1))
           eq
           ($branch_number . ".") ) {
        if ( $followsub ) {
          return 1;
        } elsif (length($revision) == length($branch_number)+2 ) {
          return 1;    # THIS IS NOT COMPLETE FOR REVISION NUMBERS WITH
                       # FINAL DIGITS > 9
        }

      But what happens if the revision on the branch is two digits? E.g
      revision is 1.16.2.10 and the branch is 1.16.2 Then what was ok for
      1.16.2.9 no longer works.
    

      The cvs2cl --xml output doesn't contain date of tags though.  I think
      is useful, so I create the following patch.  What do you think?