compat zope-2.12
[GroupUserFolder.git] / cvs2cl.pl
1 #!/bin/sh
2 exec perl -w -x $0 ${1+"$@"} # -*- mode: perl; perl-indent-level: 2; -*-
3 #!perl -w
4
5
6 ##############################################################
7 ### ###
8 ### cvs2cl.pl: produce ChangeLog(s) from `cvs log` output. ###
9 ### ###
10 ##############################################################
11
12 ## $Revision: 1.2 $
13 ## $Date: 2005-08-19 23:51:07 +0200 (ven, 19 aoĆ» 2005) $
14 ## $Author: dreamcatcher $
15 ##
16 ## (C) 2001,2002,2003 Martyn J. Pearce <fluffy@cpan.org>, under the GNU GPL.
17 ## (C) 1999 Karl Fogel <kfogel@red-bean.com>, under the GNU GPL.
18 ##
19 ## (Extensively hacked on by Melissa O'Neill <oneill@cs.sfu.ca>.)
20 ##
21 ## cvs2cl.pl is free software; you can redistribute it and/or modify
22 ## it under the terms of the GNU General Public License as published by
23 ## the Free Software Foundation; either version 2, or (at your option)
24 ## any later version.
25 ##
26 ## cvs2cl.pl is distributed in the hope that it will be useful,
27 ## but WITHOUT ANY WARRANTY; without even the implied warranty of
28 ## MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
29 ## GNU General Public License for more details.
30 ##
31 ## You may have received a copy of the GNU General Public License
32 ## along with cvs2cl.pl; see the file COPYING. If not, write to the
33 ## Free Software Foundation, Inc., 59 Temple Place - Suite 330,
34 ## Boston, MA 02111-1307, USA.
35
36 \f
37 use strict;
38 use Text::Wrap;
39 use Time::Local;
40 use File::Basename;
41
42 \f
43 # The Plan:
44 #
45 # Read in the logs for multiple files, spit out a nice ChangeLog that
46 # mirrors the information entered during `cvs commit'.
47 #
48 # The problem presents some challenges. In an ideal world, we could
49 # detect files with the same author, log message, and checkin time --
50 # each <filelist, author, time, logmessage> would be a changelog entry.
51 # We'd sort them; and spit them out. Unfortunately, CVS is *not atomic*
52 # so checkins can span a range of times. Also, the directory structure
53 # could be hierarchical.
54 #
55 # Another question is whether we really want to have the ChangeLog
56 # exactly reflect commits. An author could issue two related commits,
57 # with different log entries, reflecting a single logical change to the
58 # source. GNU style ChangeLogs group these under a single author/date.
59 # We try to do the same.
60 #
61 # So, we parse the output of `cvs log', storing log messages in a
62 # multilevel hash that stores the mapping:
63 # directory => author => time => message => filelist
64 # As we go, we notice "nearby" commit times and store them together
65 # (i.e., under the same timestamp), so they appear in the same log
66 # entry.
67 #
68 # When we've read all the logs, we twist this mapping into
69 # a time => author => message => filelist mapping for each directory.
70 #
71 # If we're not using the `--distributed' flag, the directory is always
72 # considered to be `./', even as descend into subdirectories.
73
74 \f
75 ############### Globals ################
76
77 # What we run to generate it:
78 my $Log_Source_Command = "cvs log";
79
80 # In case we have to print it out:
81 my $VERSION = '$Revision: 1.2 $';
82 $VERSION =~ s/\S+\s+(\S+)\s+\S+/$1/;
83
84 ## Vars set by options:
85
86 # Print debugging messages?
87 my $Debug = 0;
88
89 # Just show version and exit?
90 my $Print_Version = 0;
91
92 # Just print usage message and exit?
93 my $Print_Usage = 0;
94
95 # Single top-level ChangeLog, or one per subdirectory?
96 my $Distributed = 0;
97
98 # What file should we generate (defaults to "ChangeLog")?
99 my $Log_File_Name = "ChangeLog";
100
101 # Grab most recent entry date from existing ChangeLog file, just add
102 # to that ChangeLog.
103 my $Cumulative = 0;
104
105 # Expand usernames to email addresses based on a map file?
106 my $User_Map_File = "";
107
108 # Output to a file or to stdout?
109 my $Output_To_Stdout = 0;
110
111 # Eliminate empty log messages?
112 my $Prune_Empty_Msgs = 0;
113
114 # Tags of which not to output
115 my @ignore_tags;
116
117 # Don't call Text::Wrap on the body of the message
118 my $No_Wrap = 0;
119
120 # Separates header from log message. Code assumes it is either " " or
121 # "\n\n", so if there's ever an option to set it to something else,
122 # make sure to go through all conditionals that use this var.
123 my $After_Header = " ";
124
125 # XML Encoding
126 my $XML_Encoding = '';
127
128 # Format more for programs than for humans.
129 my $XML_Output = 0;
130
131 # Do some special tweaks for log data that was written in FSF
132 # ChangeLog style.
133 my $FSF_Style = 0;
134
135 # Show times in UTC instead of local time
136 my $UTC_Times = 0;
137
138 # Show times in output?
139 my $Show_Times = 1;
140
141 # Show day of week in output?
142 my $Show_Day_Of_Week = 0;
143
144 # Show revision numbers in output?
145 my $Show_Revisions = 0;
146
147 # Show tags (symbolic names) in output?
148 my $Show_Tags = 0;
149
150 # Show tags separately in output?
151 my $Show_Tag_Dates = 0;
152
153 # Show branches by symbolic name in output?
154 my $Show_Branches = 0;
155
156 # Show only revisions on these branches or their ancestors.
157 my @Follow_Branches;
158
159 # Don't bother with files matching this regexp.
160 my @Ignore_Files;
161
162 # How exactly we match entries. We definitely want "o",
163 # and user might add "i" by using --case-insensitive option.
164 my $Case_Insensitive = 0;
165
166 # Maybe only show log messages matching a certain regular expression.
167 my $Regexp_Gate = "";
168
169 # Pass this global option string along to cvs, to the left of `log':
170 my $Global_Opts = "";
171
172 # Pass this option string along to the cvs log subcommand:
173 my $Command_Opts = "";
174
175 # Read log output from stdin instead of invoking cvs log?
176 my $Input_From_Stdin = 0;
177
178 # Don't show filenames in output.
179 my $Hide_Filenames = 0;
180
181 # Max checkin duration. CVS checkin is not atomic, so we may have checkin
182 # times that span a range of time. We assume that checkins will last no
183 # longer than $Max_Checkin_Duration seconds, and that similarly, no
184 # checkins will happen from the same users with the same message less
185 # than $Max_Checkin_Duration seconds apart.
186 my $Max_Checkin_Duration = 180;
187
188 # What to put at the front of [each] ChangeLog.
189 my $ChangeLog_Header = "";
190
191 # Whether to enable 'delta' mode, and for what start/end tags.
192 my $Delta_Mode = 0;
193 my $Delta_From = "";
194 my $Delta_To = "";
195
196 ## end vars set by options.
197
198 # latest observed times for the start/end tags in delta mode
199 my $Delta_StartTime = 0;
200 my $Delta_EndTime = 0;
201
202 # In 'cvs log' output, one long unbroken line of equal signs separates
203 # files:
204 my $file_separator = "======================================="
205 . "======================================";
206
207 # In 'cvs log' output, a shorter line of dashes separates log messages
208 # within a file:
209 my $logmsg_separator = "----------------------------";
210
211 ############### End globals ############
212
213 \f
214 &parse_options ();
215 &derive_change_log ();
216
217 \f
218 ### Everything below is subroutine definitions. ###
219
220 # If accumulating, grab the boundary date from pre-existing ChangeLog.
221 sub maybe_grab_accumulation_date ()
222 {
223 if (! $Cumulative) {
224 return "";
225 }
226
227 # else
228
229 open (LOG, "$Log_File_Name")
230 or die ("trouble opening $Log_File_Name for reading ($!)");
231
232 my $boundary_date;
233 while (<LOG>)
234 {
235 if (/^(\d\d\d\d-\d\d-\d\d\s+\d\d:\d\d)/)
236 {
237 $boundary_date = "$1";
238 last;
239 }
240 }
241
242 close (LOG);
243 return $boundary_date;
244 }
245
246 # Fills up a ChangeLog structure in the current directory.
247 sub derive_change_log ()
248 {
249 # See "The Plan" above for a full explanation.
250
251 my %grand_poobah;
252
253 my $file_full_path;
254 my $time;
255 my $revision;
256 my $author;
257 my $msg_txt;
258 my $detected_file_separator;
259
260 my %tag_date_printed;
261
262 # Might be adding to an existing ChangeLog
263 my $accumulation_date = &maybe_grab_accumulation_date ();
264 if ($accumulation_date) {
265 # Insert -d immediately after 'cvs log'
266 my $Log_Date_Command = "-d\'>${accumulation_date}\'";
267 $Log_Source_Command =~ s/(^.*log\S*)/$1 $Log_Date_Command/;
268 &debug ("(adding log msg starting from $accumulation_date)\n");
269 }
270
271 # We might be expanding usernames
272 my %usermap;
273
274 # In general, it's probably not very maintainable to use state
275 # variables like this to tell the loop what it's doing at any given
276 # moment, but this is only the first one, and if we never have more
277 # than a few of these, it's okay.
278 my $collecting_symbolic_names = 0;
279 my %symbolic_names; # Where tag names get stored.
280 my %branch_names; # We'll grab branch names while we're at it.
281 my %branch_numbers; # Save some revisions for @Follow_Branches
282 my @branch_roots; # For showing which files are branch ancestors.
283
284 # Bleargh. Compensate for a deficiency of custom wrapping.
285 if (($After_Header ne " ") and $FSF_Style)
286 {
287 $After_Header .= "\t";
288 }
289
290 if (! $Input_From_Stdin) {
291 &debug ("(run \"${Log_Source_Command}\")\n");
292 open (LOG_SOURCE, "$Log_Source_Command |")
293 or die "unable to run \"${Log_Source_Command}\"";
294 }
295 else {
296 open (LOG_SOURCE, "-") or die "unable to open stdin for reading";
297 }
298
299 binmode LOG_SOURCE;
300
301 %usermap = &maybe_read_user_map_file ();
302
303 while (<LOG_SOURCE>)
304 {
305 # Canonicalize line endings
306 s/\r$//;
307 # If on a new file and don't see filename, skip until we find it, and
308 # when we find it, grab it.
309 if ((! (defined $file_full_path)) and /^Working file: (.*)/)
310 {
311 $file_full_path = $1;
312 if (@Ignore_Files)
313 {
314 my $base;
315 ($base, undef, undef) = fileparse ($file_full_path);
316 # Ouch, I wish trailing operators in regexps could be
317 # evaluated on the fly!
318 if ($Case_Insensitive) {
319 if (grep ($file_full_path =~ m|$_|i, @Ignore_Files)) {
320 undef $file_full_path;
321 }
322 }
323 elsif (grep ($file_full_path =~ m|$_|, @Ignore_Files)) {
324 undef $file_full_path;
325 }
326 }
327 next;
328 }
329
330 # Just spin wheels if no file defined yet.
331 next if (! $file_full_path);
332
333 # Collect tag names in case we're asked to print them in the output.
334 if (/^symbolic names:$/) {
335 $collecting_symbolic_names = 1;
336 next; # There's no more info on this line, so skip to next
337 }
338 if ($collecting_symbolic_names)
339 {
340 # All tag names are listed with whitespace in front in cvs log
341 # output; so if see non-whitespace, then we're done collecting.
342 if (/^\S/) {
343 $collecting_symbolic_names = 0;
344 }
345 else # we're looking at a tag name, so parse & store it
346 {
347 # According to the Cederqvist manual, in node "Tags", tag
348 # names must start with an uppercase or lowercase letter and
349 # can contain uppercase and lowercase letters, digits, `-',
350 # and `_'. However, it's not our place to enforce that, so
351 # we'll allow anything CVS hands us to be a tag:
352 /^\s+([^:]+): ([\d.]+)$/;
353 my $tag_name = $1;
354 my $tag_rev = $2;
355
356 # A branch number either has an odd number of digit sections
357 # (and hence an even number of dots), or has ".0." as the
358 # second-to-last digit section. Test for these conditions.
359 my $real_branch_rev = "";
360 if (($tag_rev =~ /^(\d+\.\d+\.)+\d+$/) # Even number of dots...
361 and (! ($tag_rev =~ /^(1\.)+1$/))) # ...but not "1.[1.]1"
362 {
363 $real_branch_rev = $tag_rev;
364 }
365 elsif ($tag_rev =~ /(\d+\.(\d+\.)+)0.(\d+)/) # Has ".0."
366 {
367 $real_branch_rev = $1 . $3;
368 }
369 # If we got a branch, record its number.
370 if ($real_branch_rev)
371 {
372 $branch_names{$real_branch_rev} = $tag_name;
373 if (@Follow_Branches) {
374 if (grep ($_ eq $tag_name, @Follow_Branches)) {
375 $branch_numbers{$tag_name} = $real_branch_rev;
376 }
377 }
378 }
379 else {
380 # Else it's just a regular (non-branch) tag.
381 push (@{$symbolic_names{$tag_rev}}, $tag_name);
382 }
383 }
384 }
385 # End of code for collecting tag names.
386
387 # If have file name, but not revision, and see revision, then grab
388 # it. (We collect unconditionally, even though we may or may not
389 # ever use it.)
390 if ((! (defined $revision)) and (/^revision (\d+\.[\d.]+)/))
391 {
392 $revision = $1;
393
394 if (@Follow_Branches)
395 {
396 foreach my $branch (@Follow_Branches)
397 {
398 # Special case for following trunk revisions
399 if (($branch =~ /^trunk$/i) and ($revision =~ /^[0-9]+\.[0-9]+$/))
400 {
401 goto dengo;
402 }
403
404 my $branch_number = $branch_numbers{$branch};
405 if ($branch_number)
406 {
407 # Are we on one of the follow branches or an ancestor of
408 # same?
409 #
410 # If this revision is a prefix of the branch number, or
411 # possibly is less in the minormost number, OR if this
412 # branch number is a prefix of the revision, then yes.
413 # Otherwise, no.
414 #
415 # So below, we determine if any of those conditions are
416 # met.
417
418 # Trivial case: is this revision on the branch?
419 # (Compare this way to avoid regexps that screw up Emacs
420 # indentation, argh.)
421 if ((substr ($revision, 0, ((length ($branch_number)) + 1)))
422 eq ($branch_number . "."))
423 {
424 goto dengo;
425 }
426 # Non-trivial case: check if rev is ancestral to branch
427 elsif ((length ($branch_number)) > (length ($revision)))
428 {
429 $revision =~ /^((?:\d+\.)+)(\d+)$/;
430 my $r_left = $1; # still has the trailing "."
431 my $r_end = $2;
432
433 $branch_number =~ /^((?:\d+\.)+)(\d+)\.\d+$/;
434 my $b_left = $1; # still has trailing "."
435 my $b_mid = $2; # has no trailing "."
436
437 if (($r_left eq $b_left)
438 && ($r_end <= $b_mid))
439 {
440 goto dengo;
441 }
442 }
443 }
444 }
445 }
446 else # (! @Follow_Branches)
447 {
448 next;
449 }
450
451 # Else we are following branches, but this revision isn't on the
452 # path. So skip it.
453 undef $revision;
454 dengo:
455 next;
456 }
457
458 # If we don't have a revision right now, we couldn't possibly
459 # be looking at anything useful.
460 if (! (defined ($revision))) {
461 $detected_file_separator = /^$file_separator$/o;
462 if ($detected_file_separator) {
463 # No revisions for this file; can happen, e.g. "cvs log -d DATE"
464 goto CLEAR;
465 }
466 else {
467 next;
468 }
469 }
470
471 # If have file name but not date and author, and see date or
472 # author, then grab them:
473 unless (defined $time)
474 {
475 if (/^date: .*/)
476 {
477 ($time, $author) = &parse_date_and_author ($_);
478 if (defined ($usermap{$author}) and $usermap{$author}) {
479 $author = $usermap{$author};
480 }
481 }
482 else {
483 $detected_file_separator = /^$file_separator$/o;
484 if ($detected_file_separator) {
485 # No revisions for this file; can happen, e.g. "cvs log -d DATE"
486 goto CLEAR;
487 }
488 }
489 # If the date/time/author hasn't been found yet, we couldn't
490 # possibly care about anything we see. So skip:
491 next;
492 }
493
494 # A "branches: ..." line here indicates that one or more branches
495 # are rooted at this revision. If we're showing branches, then we
496 # want to show that fact as well, so we collect all the branches
497 # that this is the latest ancestor of and store them in
498 # @branch_roots. Just for reference, the format of the line we're
499 # seeing at this point is:
500 #
501 # branches: 1.5.2; 1.5.4; ...;
502 #
503 # Okay, here goes:
504
505 if (/^branches:\s+(.*);$/)
506 {
507 if ($Show_Branches)
508 {
509 my $lst = $1;
510 $lst =~ s/(1\.)+1;|(1\.)+1$//; # ignore the trivial branch 1.1.1
511 if ($lst) {
512 @branch_roots = split (/;\s+/, $lst);
513 }
514 else {
515 undef @branch_roots;
516 }
517 next;
518 }
519 else
520 {
521 # Ugh. This really bothers me. Suppose we see a log entry
522 # like this:
523 #
524 # ----------------------------
525 # revision 1.1
526 # date: 1999/10/17 03:07:38; author: jrandom; state: Exp;
527 # branches: 1.1.2;
528 # Intended first line of log message begins here.
529 # ----------------------------
530 #
531 # The question is, how we can tell the difference between that
532 # log message and a *two*-line log message whose first line is
533 #
534 # "branches: 1.1.2;"
535 #
536 # See the problem? The output of "cvs log" is inherently
537 # ambiguous.
538 #
539 # For now, we punt: we liberally assume that people don't
540 # write log messages like that, and just toss a "branches:"
541 # line if we see it but are not showing branches. I hope no
542 # one ever loses real log data because of this.
543 next;
544 }
545 }
546
547 # If have file name, time, and author, then we're just grabbing
548 # log message texts:
549 $detected_file_separator = /^$file_separator$/o;
550 if ($detected_file_separator && ! (defined $revision)) {
551 # No revisions for this file; can happen, e.g. "cvs log -d DATE"
552 goto CLEAR;
553 }
554 unless ($detected_file_separator || /^$logmsg_separator$/o)
555 {
556 $msg_txt .= $_; # Normally, just accumulate the message...
557 next;
558 }
559 # ... until a msg separator is encountered:
560 # Ensure the message contains something:
561 if ((! $msg_txt)
562 || ($msg_txt =~ /^\s*\.\s*$|^\s*$/)
563 || ($msg_txt =~ /\*\*\* empty log message \*\*\*/))
564 {
565 if ($Prune_Empty_Msgs) {
566 goto CLEAR;
567 }
568 # else
569 $msg_txt = "[no log message]\n";
570 }
571
572 ### Store it all in the Grand Poobah:
573 {
574 my $dir_key; # key into %grand_poobah
575 my %qunk; # complicated little jobbie, see below
576
577 # Each revision of a file has a little data structure (a `qunk')
578 # associated with it. That data structure holds not only the
579 # file's name, but any additional information about the file
580 # that might be needed in the output, such as the revision
581 # number, tags, branches, etc. The reason to have these things
582 # arranged in a data structure, instead of just appending them
583 # textually to the file's name, is that we may want to do a
584 # little rearranging later as we write the output. For example,
585 # all the files on a given tag/branch will go together, followed
586 # by the tag in parentheses (so trunk or otherwise non-tagged
587 # files would go at the end of the file list for a given log
588 # message). This rearrangement is a lot easier to do if we
589 # don't have to reparse the text.
590 #
591 # A qunk looks like this:
592 #
593 # {
594 # filename => "hello.c",
595 # revision => "1.4.3.2",
596 # time => a timegm() return value (moment of commit)
597 # tags => [ "tag1", "tag2", ... ],
598 # branch => "branchname" # There should be only one, right?
599 # branchroots => [ "branchtag1", "branchtag2", ... ]
600 # }
601
602 if ($Distributed) {
603 # Just the basename, don't include the path.
604 ($qunk{'filename'}, $dir_key, undef) = fileparse ($file_full_path);
605 }
606 else {
607 $dir_key = "./";
608 $qunk{'filename'} = $file_full_path;
609 }
610
611 # This may someday be used in a more sophisticated calculation
612 # of what other files are involved in this commit. For now, we
613 # don't use it much except for delta mode, because the
614 # common-commit-detection algorithm is hypothesized to be
615 # "good enough" as it stands.
616 $qunk{'time'} = $time;
617
618 # We might be including revision numbers and/or tags and/or
619 # branch names in the output. Most of the code from here to
620 # loop-end deals with organizing these in qunk.
621
622 $qunk{'revision'} = $revision;
623
624 # Grab the branch, even though we may or may not need it:
625 $qunk{'revision'} =~ /((?:\d+\.)+)\d+/;
626 my $branch_prefix = $1;
627 $branch_prefix =~ s/\.$//; # strip off final dot
628 if ($branch_names{$branch_prefix}) {
629 $qunk{'branch'} = $branch_names{$branch_prefix};
630 }
631
632 # If there's anything in the @branch_roots array, then this
633 # revision is the root of at least one branch. We'll display
634 # them as branch names instead of revision numbers, the
635 # substitution for which is done directly in the array:
636 if (@branch_roots) {
637 my @roots = map { $branch_names{$_} } @branch_roots;
638 $qunk{'branchroots'} = \@roots;
639 }
640
641 # Save tags too.
642 if (defined ($symbolic_names{$revision})) {
643 $qunk{'tags'} = $symbolic_names{$revision};
644 delete $symbolic_names{$revision};
645
646 # If we're in 'delta' mode, update the latest observed
647 # times for the beginning and ending tags, and
648 # when we get around to printing output, we will simply restrict
649 # ourselves to that timeframe...
650
651 if ($Delta_Mode) {
652 if (($time > $Delta_StartTime) &&
653 (grep { $_ eq $Delta_From } @{$qunk{'tags'}}))
654 {
655 $Delta_StartTime = $time;
656 }
657
658 if (($time > $Delta_EndTime) &&
659 (grep { $_ eq $Delta_To } @{$qunk{'tags'}}))
660 {
661 $Delta_EndTime = $time;
662 }
663 }
664 }
665
666 # Add this file to the list
667 # (We use many spoonfuls of autovivication magic. Hashes and arrays
668 # will spring into existence if they aren't there already.)
669
670 &debug ("(pushing log msg for ${dir_key}$qunk{'filename'})\n");
671
672 # Store with the files in this commit. Later we'll loop through
673 # again, making sure that revisions with the same log message
674 # and nearby commit times are grouped together as one commit.
675 push (@{$grand_poobah{$dir_key}{$author}{$time}{$msg_txt}}, \%qunk);
676 }
677
678 CLEAR:
679 # Make way for the next message
680 undef $msg_txt;
681 undef $time;
682 undef $revision;
683 undef $author;
684 undef @branch_roots;
685
686 # Maybe even make way for the next file:
687 if ($detected_file_separator) {
688 undef $file_full_path;
689 undef %branch_names;
690 undef %branch_numbers;
691 undef %symbolic_names;
692 }
693 }
694
695 close (LOG_SOURCE);
696
697 ### Process each ChangeLog
698
699 while (my ($dir,$authorhash) = each %grand_poobah)
700 {
701 &debug ("DOING DIR: $dir\n");
702
703 # Here we twist our hash around, from being
704 # author => time => message => filelist
705 # in %$authorhash to
706 # time => author => message => filelist
707 # in %changelog.
708 #
709 # This is also where we merge entries. The algorithm proceeds
710 # through the timeline of the changelog with a sliding window of
711 # $Max_Checkin_Duration seconds; within that window, entries that
712 # have the same log message are merged.
713 #
714 # (To save space, we zap %$authorhash after we've copied
715 # everything out of it.)
716
717 my %changelog;
718 while (my ($author,$timehash) = each %$authorhash)
719 {
720 my $lasttime;
721 my %stamptime;
722 foreach my $time (sort {$main::a <=> $main::b} (keys %$timehash))
723 {
724 my $msghash = $timehash->{$time};
725 while (my ($msg,$qunklist) = each %$msghash)
726 {
727 my $stamptime = $stamptime{$msg};
728 if ((defined $stamptime)
729 and (($time - $stamptime) < $Max_Checkin_Duration)
730 and (defined $changelog{$stamptime}{$author}{$msg}))
731 {
732 push(@{$changelog{$stamptime}{$author}{$msg}}, @$qunklist);
733 }
734 else {
735 $changelog{$time}{$author}{$msg} = $qunklist;
736 $stamptime{$msg} = $time;
737 }
738 }
739 }
740 }
741 undef (%$authorhash);
742
743 ### Now we can write out the ChangeLog!
744
745 my ($logfile_here, $logfile_bak, $tmpfile);
746
747 if (! $Output_To_Stdout) {
748 $logfile_here = $dir . $Log_File_Name;
749 $logfile_here =~ s/^\.\/\//\//; # fix any leading ".//" problem
750 $tmpfile = "${logfile_here}.cvs2cl$$.tmp";
751 $logfile_bak = "${logfile_here}.bak";
752
753 open (LOG_OUT, ">$tmpfile") or die "Unable to open \"$tmpfile\"";
754 }
755 else {
756 open (LOG_OUT, ">-") or die "Unable to open stdout for writing";
757 }
758
759 print LOG_OUT $ChangeLog_Header;
760
761 if ($XML_Output) {
762 my $encoding =
763 length $XML_Encoding ? qq'encoding="$XML_Encoding"' : '';
764 my $version = 'version="1.0"';
765 my $declaration =
766 sprintf '<?xml %s?>', join ' ', grep length, $version, $encoding;
767 my $root =
768 '<changelog xmlns="http://www.red-bean.com/xmlns/cvs2cl/">';
769 print LOG_OUT "$declaration\n\n$root\n\n";
770 }
771
772 foreach my $time (sort {$main::b <=> $main::a} (keys %changelog))
773 {
774 next if ($Delta_Mode &&
775 (($time <= $Delta_StartTime) ||
776 ($time > $Delta_EndTime && $Delta_EndTime)));
777
778 # Set up the date/author line.
779 # kff todo: do some more XML munging here, on the header
780 # part of the entry:
781 my ($ignore,$min,$hour,$mday,$mon,$year,$wday)
782 = $UTC_Times ? gmtime($time) : localtime($time);
783
784 # XML output includes everything else, we might as well make
785 # it always include Day Of Week too, for consistency.
786 if ($Show_Day_Of_Week or $XML_Output) {
787 $wday = ("Sunday", "Monday", "Tuesday", "Wednesday",
788 "Thursday", "Friday", "Saturday")[$wday];
789 $wday = ($XML_Output) ? "<weekday>${wday}</weekday>\n" : " $wday";
790 }
791 else {
792 $wday = "";
793 }
794
795 my $authorhash = $changelog{$time};
796 if ($Show_Tag_Dates) {
797 my %tags;
798 while (my ($author,$mesghash) = each %$authorhash) {
799 while (my ($msg,$qunk) = each %$mesghash) {
800 foreach my $qunkref2 (@$qunk) {
801 if (defined ($$qunkref2{'tags'})) {
802 foreach my $tag (@{$$qunkref2{'tags'}}) {
803 $tags{$tag} = 1;
804 }
805 }
806 }
807 }
808 }
809 foreach my $tag (keys %tags) {
810 if (!defined $tag_date_printed{$tag}) {
811 $tag_date_printed{$tag} = $time;
812 if ($XML_Output) {
813 # NOT YET DONE
814 }
815 else {
816 if ($Show_Times) {
817 printf LOG_OUT ("%4u-%02u-%02u${wday} %02u:%02u tag %s\n\n",
818 $year+1900, $mon+1, $mday, $hour, $min, $tag);
819 } else {
820 printf LOG_OUT ("%4u-%02u-%02u${wday} tag %s\n\n",
821 $year+1900, $mon+1, $mday, $tag);
822 }
823 }
824 }
825 }
826 }
827 while (my ($author,$mesghash) = each %$authorhash)
828 {
829 # If XML, escape in outer loop to avoid compound quoting:
830 if ($XML_Output) {
831 $author = &xml_escape ($author);
832 }
833
834 FOOBIE:
835 while (my ($msg,$qunklist) = each %$mesghash)
836 {
837 ## MJP: 19.xii.01 : Exclude @ignore_tags
838 for my $ignore_tag (@ignore_tags) {
839 next FOOBIE
840 if grep $_ eq $ignore_tag, map(@{$_->{tags}},
841 grep(defined $_->{tags},
842 @$qunklist));
843 }
844 ## MJP: 19.xii.01 : End exclude @ignore_tags
845
846 my $files = &pretty_file_list ($qunklist);
847 my $header_line; # date and author
848 my $body; # see below
849 my $wholething; # $header_line + $body
850
851 if ($XML_Output) {
852 $header_line =
853 sprintf ("<date>%4u-%02u-%02u</date>\n"
854 . "${wday}"
855 . "<time>%02u:%02u</time>\n"
856 . "<author>%s</author>\n",
857 $year+1900, $mon+1, $mday, $hour, $min, $author);
858 }
859 else {
860 if ($Show_Times) {
861 $header_line =
862 sprintf ("%4u-%02u-%02u${wday} %02u:%02u %s\n\n",
863 $year+1900, $mon+1, $mday, $hour, $min, $author);
864 } else {
865 $header_line =
866 sprintf ("%4u-%02u-%02u${wday} %s\n\n",
867 $year+1900, $mon+1, $mday, $author);
868 }
869 }
870
871 $Text::Wrap::huge = 'overflow'
872 if $Text::Wrap::VERSION >= 2001.0130;
873 # Reshape the body according to user preferences.
874 if ($XML_Output)
875 {
876 $msg = &preprocess_msg_text ($msg);
877 $body = $files . $msg;
878 }
879 elsif ($No_Wrap)
880 {
881 $msg = &preprocess_msg_text ($msg);
882 $files = wrap ("\t", " ", "$files");
883 $msg =~ s/\n(.*)/\n\t$1/g;
884 unless ($After_Header eq " ") {
885 $msg =~ s/^(.*)/\t$1/g;
886 }
887 $body = $files . $After_Header . $msg;
888 }
889 else # do wrapping, either FSF-style or regular
890 {
891 if ($FSF_Style)
892 {
893 $files = wrap ("\t", " ", "$files");
894
895 my $files_last_line_len = 0;
896 if ($After_Header eq " ")
897 {
898 $files_last_line_len = &last_line_len ($files);
899 $files_last_line_len += 1; # for $After_Header
900 }
901
902 $msg = &wrap_log_entry
903 ($msg, "\t", 69 - $files_last_line_len, 69);
904 $body = $files . $After_Header . $msg;
905 }
906 else # not FSF-style
907 {
908 $msg = &preprocess_msg_text ($msg);
909 $body = $files . $After_Header . $msg;
910 $body = wrap ("\t", " ", "$body");
911 }
912 }
913
914 $wholething = $header_line . $body;
915
916 if ($XML_Output) {
917 $wholething = "<entry>\n${wholething}</entry>\n";
918 }
919
920 # One last check: make sure it passes the regexp test, if the
921 # user asked for that. We have to do it here, so that the
922 # test can match against information in the header as well
923 # as in the text of the log message.
924
925 # How annoying to duplicate so much code just because I
926 # can't figure out a way to evaluate scalars on the trailing
927 # operator portion of a regular expression. Grrr.
928 if ($Case_Insensitive) {
929 unless ($Regexp_Gate && ($wholething !~ /$Regexp_Gate/oi)) {
930 print LOG_OUT "${wholething}\n";
931 }
932 }
933 else {
934 unless ($Regexp_Gate && ($wholething !~ /$Regexp_Gate/o)) {
935 print LOG_OUT "${wholething}\n";
936 }
937 }
938 }
939 }
940 }
941
942 if ($XML_Output) {
943 print LOG_OUT "</changelog>\n";
944 }
945
946 close (LOG_OUT);
947
948 if (! $Output_To_Stdout)
949 {
950 # If accumulating, append old data to new before renaming. But
951 # don't append the most recent entry, since it's already in the
952 # new log due to CVS's idiosyncratic interpretation of "log -d".
953 if ($Cumulative && -f $logfile_here)
954 {
955 open (NEW_LOG, ">>$tmpfile")
956 or die "trouble appending to $tmpfile ($!)";
957
958 open (OLD_LOG, "<$logfile_here")
959 or die "trouble reading from $logfile_here ($!)";
960
961 my $started_first_entry = 0;
962 my $passed_first_entry = 0;
963 while (<OLD_LOG>)
964 {
965 if (! $passed_first_entry)
966 {
967 if ((! $started_first_entry)
968 && /^(\d\d\d\d-\d\d-\d\d\s+\d\d:\d\d)/) {
969 $started_first_entry = 1;
970 }
971 elsif (/^(\d\d\d\d-\d\d-\d\d\s+\d\d:\d\d)/) {
972 $passed_first_entry = 1;
973 print NEW_LOG $_;
974 }
975 }
976 else {
977 print NEW_LOG $_;
978 }
979 }
980
981 close (NEW_LOG);
982 close (OLD_LOG);
983 }
984
985 if (-f $logfile_here) {
986 rename ($logfile_here, $logfile_bak);
987 }
988 rename ($tmpfile, $logfile_here);
989 }
990 }
991 }
992
993 sub parse_date_and_author ()
994 {
995 # Parses the date/time and author out of a line like:
996 #
997 # date: 1999/02/19 23:29:05; author: apharris; state: Exp;
998
999 my $line = shift;
1000
1001 my ($year, $mon, $mday, $hours, $min, $secs, $author) = $line =~
1002 m#(\d+)/(\d+)/(\d+)\s+(\d+):(\d+):(\d+);\s+author:\s+([^;]+);#
1003 or die "Couldn't parse date ``$line''";
1004 die "Bad date or Y2K issues" unless ($year > 1969 and $year < 2258);
1005 # Kinda arbitrary, but useful as a sanity check
1006 my $time = timegm($secs,$min,$hours,$mday,$mon-1,$year-1900);
1007
1008 return ($time, $author);
1009 }
1010
1011 # Here we take a bunch of qunks and convert them into printed
1012 # summary that will include all the information the user asked for.
1013 sub pretty_file_list ()
1014 {
1015 if ($Hide_Filenames and (! $XML_Output)) {
1016 return "";
1017 }
1018
1019 my $qunksref = shift;
1020 my @qunkrefs = @$qunksref;
1021 my @filenames;
1022 my $beauty = ""; # The accumulating header string for this entry.
1023 my %non_unanimous_tags; # Tags found in a proper subset of qunks
1024 my %unanimous_tags; # Tags found in all qunks
1025 my %all_branches; # Branches found in any qunk
1026 my $common_dir = undef; # Dir prefix common to all files ("" if none)
1027 my $fbegun = 0; # Did we begin printing filenames yet?
1028
1029 # First, loop over the qunks gathering all the tag/branch names.
1030 # We'll put them all in non_unanimous_tags, and take out the
1031 # unanimous ones later.
1032 QUNKREF:
1033 foreach my $qunkref (@qunkrefs)
1034 {
1035 ## MJP: 19.xii.01 : Exclude @ignore_tags
1036 for my $ignore_tag (@ignore_tags) {
1037 next QUNKREF
1038 if grep $_ eq $ignore_tag, @{$$qunkref{'tags'}};
1039 }
1040 ## MJP: 19.xii.01 : End exclude @ignore_tags
1041
1042 # Keep track of whether all the files in this commit were in the
1043 # same directory, and memorize it if so. We can make the output a
1044 # little more compact by mentioning the directory only once.
1045 if ((scalar (@qunkrefs)) > 1)
1046 {
1047 if (! (defined ($common_dir)))
1048 {
1049 my ($base, $dir);
1050 ($base, $dir, undef) = fileparse ($$qunkref{'filename'});
1051
1052 if ((! (defined ($dir))) # this first case is sheer paranoia
1053 or ($dir eq "")
1054 or ($dir eq "./")
1055 or ($dir eq ".\\"))
1056 {
1057 $common_dir = "";
1058 }
1059 else
1060 {
1061 $common_dir = $dir;
1062 }
1063 }
1064 elsif ($common_dir ne "")
1065 {
1066 # Already have a common dir prefix, so how much of it can we preserve?
1067 $common_dir = &common_path_prefix ($$qunkref{'filename'}, $common_dir);
1068 }
1069 }
1070 else # only one file in this entry anyway, so common dir not an issue
1071 {
1072 $common_dir = "";
1073 }
1074
1075 if (defined ($$qunkref{'branch'})) {
1076 $all_branches{$$qunkref{'branch'}} = 1;
1077 }
1078 if (defined ($$qunkref{'tags'})) {
1079 foreach my $tag (@{$$qunkref{'tags'}}) {
1080 $non_unanimous_tags{$tag} = 1;
1081 }
1082 }
1083 }
1084
1085 # Any tag held by all qunks will be printed specially... but only if
1086 # there are multiple qunks in the first place!
1087 if ((scalar (@qunkrefs)) > 1) {
1088 foreach my $tag (keys (%non_unanimous_tags)) {
1089 my $everyone_has_this_tag = 1;
1090 foreach my $qunkref (@qunkrefs) {
1091 if ((! (defined ($$qunkref{'tags'})))
1092 or (! (grep ($_ eq $tag, @{$$qunkref{'tags'}})))) {
1093 $everyone_has_this_tag = 0;
1094 }
1095 }
1096 if ($everyone_has_this_tag) {
1097 $unanimous_tags{$tag} = 1;
1098 delete $non_unanimous_tags{$tag};
1099 }
1100 }
1101 }
1102
1103 if ($XML_Output)
1104 {
1105 # If outputting XML, then our task is pretty simple, because we
1106 # don't have to detect common dir, common tags, branch prefixing,
1107 # etc. We just output exactly what we have, and don't worry about
1108 # redundancy or readability.
1109
1110 foreach my $qunkref (@qunkrefs)
1111 {
1112 my $filename = $$qunkref{'filename'};
1113 my $revision = $$qunkref{'revision'};
1114 my $tags = $$qunkref{'tags'};
1115 my $branch = $$qunkref{'branch'};
1116 my $branchroots = $$qunkref{'branchroots'};
1117
1118 $filename = &xml_escape ($filename); # probably paranoia
1119 $revision = &xml_escape ($revision); # definitely paranoia
1120
1121 $beauty .= "<file>\n";
1122 $beauty .= "<name>${filename}</name>\n";
1123 $beauty .= "<revision>${revision}</revision>\n";
1124 if ($branch) {
1125 $branch = &xml_escape ($branch); # more paranoia
1126 $beauty .= "<branch>${branch}</branch>\n";
1127 }
1128 foreach my $tag (@$tags) {
1129 $tag = &xml_escape ($tag); # by now you're used to the paranoia
1130 $beauty .= "<tag>${tag}</tag>\n";
1131 }
1132 foreach my $root (@$branchroots) {
1133 $root = &xml_escape ($root); # which is good, because it will continue
1134 $beauty .= "<branchroot>${root}</branchroot>\n";
1135 }
1136 $beauty .= "</file>\n";
1137 }
1138
1139 # Theoretically, we could go home now. But as long as we're here,
1140 # let's print out the common_dir and utags, as a convenience to
1141 # the receiver (after all, earlier code calculated that stuff
1142 # anyway, so we might as well take advantage of it).
1143
1144 if ((scalar (keys (%unanimous_tags))) > 1) {
1145 foreach my $utag ((keys (%unanimous_tags))) {
1146 $utag = &xml_escape ($utag); # the usual paranoia
1147 $beauty .= "<utag>${utag}</utag>\n";
1148 }
1149 }
1150 if ($common_dir) {
1151 $common_dir = &xml_escape ($common_dir);
1152 $beauty .= "<commondir>${common_dir}</commondir>\n";
1153 }
1154
1155 # That's enough for XML, time to go home:
1156 return $beauty;
1157 }
1158
1159 # Else not XML output, so complexly compactify for chordate
1160 # consumption. At this point we have enough global information
1161 # about all the qunks to organize them non-redundantly for output.
1162
1163 if ($common_dir) {
1164 # Note that $common_dir still has its trailing slash
1165 $beauty .= "$common_dir: ";
1166 }
1167
1168 if ($Show_Branches)
1169 {
1170 # For trailing revision numbers.
1171 my @brevisions;
1172
1173 foreach my $branch (keys (%all_branches))
1174 {
1175 foreach my $qunkref (@qunkrefs)
1176 {
1177 if ((defined ($$qunkref{'branch'}))
1178 and ($$qunkref{'branch'} eq $branch))
1179 {
1180 if ($fbegun) {
1181 # kff todo: comma-delimited in XML too? Sure.
1182 $beauty .= ", ";
1183 }
1184 else {
1185 $fbegun = 1;
1186 }
1187 my $fname = substr ($$qunkref{'filename'}, length ($common_dir));
1188 $beauty .= $fname;
1189 $$qunkref{'printed'} = 1; # Just setting a mark bit, basically
1190
1191 if ($Show_Tags && (defined @{$$qunkref{'tags'}})) {
1192 my @tags = grep ($non_unanimous_tags{$_}, @{$$qunkref{'tags'}});
1193
1194 if (@tags) {
1195 $beauty .= " (tags: ";
1196 $beauty .= join (', ', @tags);
1197 $beauty .= ")";
1198 }
1199 }
1200
1201 if ($Show_Revisions) {
1202 # Collect the revision numbers' last components, but don't
1203 # print them -- they'll get printed with the branch name
1204 # later.
1205 $$qunkref{'revision'} =~ /.+\.([\d]+)$/;
1206 push (@brevisions, $1);
1207
1208 # todo: we're still collecting branch roots, but we're not
1209 # showing them anywhere. If we do show them, it would be
1210 # nifty to just call them revision "0" on a the branch.
1211 # Yeah, that's the ticket.
1212 }
1213 }
1214 }
1215 $beauty .= " ($branch";
1216 if (@brevisions) {
1217 if ((scalar (@brevisions)) > 1) {
1218 $beauty .= ".[";
1219 $beauty .= (join (',', @brevisions));
1220 $beauty .= "]";
1221 }
1222 else {
1223 # Square brackets are spurious here, since there's no range to
1224 # encapsulate
1225 $beauty .= ".$brevisions[0]";
1226 }
1227 }
1228 $beauty .= ")";
1229 }
1230 }
1231
1232 # Okay; any qunks that were done according to branch are taken care
1233 # of, and marked as printed. Now print everyone else.
1234
1235 foreach my $qunkref (@qunkrefs)
1236 {
1237 next if (defined ($$qunkref{'printed'})); # skip if already printed
1238
1239 if ($fbegun) {
1240 $beauty .= ", ";
1241 }
1242 else {
1243 $fbegun = 1;
1244 }
1245 $beauty .= substr ($$qunkref{'filename'}, length ($common_dir));
1246 # todo: Shlomo's change was this:
1247 # $beauty .= substr ($$qunkref{'filename'},
1248 # (($common_dir eq "./") ? "" : length ($common_dir)));
1249 $$qunkref{'printed'} = 1; # Set a mark bit.
1250
1251 if ($Show_Revisions || $Show_Tags)
1252 {
1253 my $started_addendum = 0;
1254
1255 if ($Show_Revisions) {
1256 $started_addendum = 1;
1257 $beauty .= " (";
1258 $beauty .= "$$qunkref{'revision'}";
1259 }
1260 if ($Show_Tags && (defined $$qunkref{'tags'})) {
1261 my @tags = grep ($non_unanimous_tags{$_}, @{$$qunkref{'tags'}});
1262 if ((scalar (@tags)) > 0) {
1263 if ($started_addendum) {
1264 $beauty .= ", ";
1265 }
1266 else {
1267 $beauty .= " (tags: ";
1268 }
1269 $beauty .= join (', ', @tags);
1270 $started_addendum = 1;
1271 }
1272 }
1273 if ($started_addendum) {
1274 $beauty .= ")";
1275 }
1276 }
1277 }
1278
1279 # Unanimous tags always come last.
1280 if ($Show_Tags && %unanimous_tags)
1281 {
1282 $beauty .= " (utags: ";
1283 $beauty .= join (', ', sort keys (%unanimous_tags));
1284 $beauty .= ")";
1285 }
1286
1287 # todo: still have to take care of branch_roots?
1288
1289 $beauty = "* $beauty:";
1290
1291 return $beauty;
1292 }
1293
1294 sub common_path_prefix ()
1295 {
1296 my $path1 = shift;
1297 my $path2 = shift;
1298
1299 my ($dir1, $dir2);
1300 (undef, $dir1, undef) = fileparse ($path1);
1301 (undef, $dir2, undef) = fileparse ($path2);
1302
1303 # Transmogrify Windows filenames to look like Unix.
1304 # (It is far more likely that someone is running cvs2cl.pl under
1305 # Windows than that they would genuinely have backslashes in their
1306 # filenames.)
1307 $dir1 =~ tr#\\#/#;
1308 $dir2 =~ tr#\\#/#;
1309
1310 my $accum1 = "";
1311 my $accum2 = "";
1312 my $last_common_prefix = "";
1313
1314 while ($accum1 eq $accum2)
1315 {
1316 $last_common_prefix = $accum1;
1317 last if ($accum1 eq $dir1);
1318 my ($tmp1) = split (/\//, (substr ($dir1, length ($accum1))));
1319 my ($tmp2) = split (/\//, (substr ($dir2, length ($accum2))));
1320 $accum1 .= "$tmp1/" if (defined $tmp1 and $tmp1 ne '');
1321 $accum2 .= "$tmp2/" if (defined $tmp2 and $tmp2 ne '');
1322 }
1323
1324 return $last_common_prefix;
1325 }
1326
1327 sub preprocess_msg_text ()
1328 {
1329 my $text = shift;
1330
1331 # Strip out carriage returns (as they probably result from DOSsy editors).
1332 $text =~ s/\r\n/\n/g;
1333
1334 # If it *looks* like two newlines, make it *be* two newlines:
1335 $text =~ s/\n\s*\n/\n\n/g;
1336
1337 if ($XML_Output)
1338 {
1339 $text = &xml_escape ($text);
1340 $text = "<msg>${text}</msg>\n";
1341 }
1342 elsif (! $No_Wrap)
1343 {
1344 # Strip off lone newlines, but only for lines that don't begin with
1345 # whitespace or a mail-quoting character, since we want to preserve
1346 # that kind of formatting. Also don't strip newlines that follow a
1347 # period; we handle those specially next. And don't strip
1348 # newlines that precede an open paren.
1349 1 while ($text =~ s/(^|\n)([^>\s].*[^.\n])\n([^>\n])/$1$2 $3/g);
1350
1351 # If a newline follows a period, make sure that when we bring up the
1352 # bottom sentence, it begins with two spaces.
1353 1 while ($text =~ s/(^|\n)([^>\s].*)\n([^>\n])/$1$2 $3/g);
1354 }
1355
1356 return $text;
1357 }
1358
1359 sub last_line_len ()
1360 {
1361 my $files_list = shift;
1362 my @lines = split (/\n/, $files_list);
1363 my $last_line = pop (@lines);
1364 return length ($last_line);
1365 }
1366
1367 # A custom wrap function, sensitive to some common constructs used in
1368 # log entries.
1369 sub wrap_log_entry ()
1370 {
1371 my $text = shift; # The text to wrap.
1372 my $left_pad_str = shift; # String to pad with on the left.
1373
1374 # These do NOT take left_pad_str into account:
1375 my $length_remaining = shift; # Amount left on current line.
1376 my $max_line_length = shift; # Amount left for a blank line.
1377
1378 my $wrapped_text = ""; # The accumulating wrapped entry.
1379 my $user_indent = ""; # Inherited user_indent from prev line.
1380
1381 my $first_time = 1; # First iteration of the loop?
1382 my $suppress_line_start_match = 0; # Set to disable line start checks.
1383
1384 my @lines = split (/\n/, $text);
1385 while (@lines) # Don't use `foreach' here, it won't work.
1386 {
1387 my $this_line = shift (@lines);
1388 chomp $this_line;
1389
1390 if ($this_line =~ /^(\s+)/) {
1391 $user_indent = $1;
1392 }
1393 else {
1394 $user_indent = "";
1395 }
1396
1397 # If it matches any of the line-start regexps, print a newline now...
1398 if ($suppress_line_start_match)
1399 {
1400 $suppress_line_start_match = 0;
1401 }
1402 elsif (($this_line =~ /^(\s*)\*\s+[a-zA-Z0-9]/)
1403 || ($this_line =~ /^(\s*)\* [a-zA-Z0-9_\.\/\+-]+/)
1404 || ($this_line =~ /^(\s*)\([a-zA-Z0-9_\.\/\+-]+(\)|,\s*)/)
1405 || ($this_line =~ /^(\s+)(\S+)/)
1406 || ($this_line =~ /^(\s*)- +/)
1407 || ($this_line =~ /^()\s*$/)
1408 || ($this_line =~ /^(\s*)\*\) +/)
1409 || ($this_line =~ /^(\s*)[a-zA-Z0-9](\)|\.|\:) +/))
1410 {
1411 # Make a line break immediately, unless header separator is set
1412 # and this line is the first line in the entry, in which case
1413 # we're getting the blank line for free already and shouldn't
1414 # add an extra one.
1415 unless (($After_Header ne " ") and ($first_time))
1416 {
1417 if ($this_line =~ /^()\s*$/) {
1418 $suppress_line_start_match = 1;
1419 $wrapped_text .= "\n${left_pad_str}";
1420 }
1421
1422 $wrapped_text .= "\n${left_pad_str}";
1423 }
1424
1425 $length_remaining = $max_line_length - (length ($user_indent));
1426 }
1427
1428 # Now that any user_indent has been preserved, strip off leading
1429 # whitespace, so up-folding has no ugly side-effects.
1430 $this_line =~ s/^\s*//;
1431
1432 # Accumulate the line, and adjust parameters for next line.
1433 my $this_len = length ($this_line);
1434 if ($this_len == 0)
1435 {
1436 # Blank lines should cancel any user_indent level.
1437 $user_indent = "";
1438 $length_remaining = $max_line_length;
1439 }
1440 elsif ($this_len >= $length_remaining) # Line too long, try breaking it.
1441 {
1442 # Walk backwards from the end. At first acceptable spot, break
1443 # a new line.
1444 my $idx = $length_remaining - 1;
1445 if ($idx < 0) { $idx = 0 };
1446 while ($idx > 0)
1447 {
1448 if (substr ($this_line, $idx, 1) =~ /\s/)
1449 {
1450 my $line_now = substr ($this_line, 0, $idx);
1451 my $next_line = substr ($this_line, $idx);
1452 $this_line = $line_now;
1453
1454 # Clean whitespace off the end.
1455 chomp $this_line;
1456
1457 # The current line is ready to be printed.
1458 $this_line .= "\n${left_pad_str}";
1459
1460 # Make sure the next line is allowed full room.
1461 $length_remaining = $max_line_length - (length ($user_indent));
1462
1463 # Strip next_line, but then preserve any user_indent.
1464 $next_line =~ s/^\s*//;
1465
1466 # Sneak a peek at the user_indent of the upcoming line, so
1467 # $next_line (which will now precede it) can inherit that
1468 # indent level. Otherwise, use whatever user_indent level
1469 # we currently have, which might be none.
1470 my $next_next_line = shift (@lines);
1471 if ((defined ($next_next_line)) && ($next_next_line =~ /^(\s+)/)) {
1472 $next_line = $1 . $next_line if (defined ($1));
1473 # $length_remaining = $max_line_length - (length ($1));
1474 $next_next_line =~ s/^\s*//;
1475 }
1476 else {
1477 $next_line = $user_indent . $next_line;
1478 }
1479 if (defined ($next_next_line)) {
1480 unshift (@lines, $next_next_line);
1481 }
1482 unshift (@lines, $next_line);
1483
1484 # Our new next line might, coincidentally, begin with one of
1485 # the line-start regexps, so we temporarily turn off
1486 # sensitivity to that until we're past the line.
1487 $suppress_line_start_match = 1;
1488
1489 last;
1490 }
1491 else
1492 {
1493 $idx--;
1494 }
1495 }
1496
1497 if ($idx == 0)
1498 {
1499 # We bottomed out because the line is longer than the
1500 # available space. But that could be because the space is
1501 # small, or because the line is longer than even the maximum
1502 # possible space. Handle both cases below.
1503
1504 if ($length_remaining == ($max_line_length - (length ($user_indent))))
1505 {
1506 # The line is simply too long -- there is no hope of ever
1507 # breaking it nicely, so just insert it verbatim, with
1508 # appropriate padding.
1509 $this_line = "\n${left_pad_str}${this_line}";
1510 }
1511 else
1512 {
1513 # Can't break it here, but may be able to on the next round...
1514 unshift (@lines, $this_line);
1515 $length_remaining = $max_line_length - (length ($user_indent));
1516 $this_line = "\n${left_pad_str}";
1517 }
1518 }
1519 }
1520 else # $this_len < $length_remaining, so tack on what we can.
1521 {
1522 # Leave a note for the next iteration.
1523 $length_remaining = $length_remaining - $this_len;
1524
1525 if ($this_line =~ /\.$/)
1526 {
1527 $this_line .= " ";
1528 $length_remaining -= 2;
1529 }
1530 else # not a sentence end
1531 {
1532 $this_line .= " ";
1533 $length_remaining -= 1;
1534 }
1535 }
1536
1537 # Unconditionally indicate that loop has run at least once.
1538 $first_time = 0;
1539
1540 $wrapped_text .= "${user_indent}${this_line}";
1541 }
1542
1543 # One last bit of padding.
1544 $wrapped_text .= "\n";
1545
1546 return $wrapped_text;
1547 }
1548
1549 sub xml_escape ()
1550 {
1551 my $txt = shift;
1552 $txt =~ s/&/&amp;/g;
1553 $txt =~ s/</&lt;/g;
1554 $txt =~ s/>/&gt;/g;
1555 return $txt;
1556 }
1557
1558 sub maybe_read_user_map_file ()
1559 {
1560 my %expansions;
1561
1562 if ($User_Map_File)
1563 {
1564 open (MAPFILE, "<$User_Map_File")
1565 or die ("Unable to open $User_Map_File ($!)");
1566
1567 while (<MAPFILE>)
1568 {
1569 next if /^\s*#/; # Skip comment lines.
1570 next if not /:/; # Skip lines without colons.
1571
1572 # It is now safe to split on ':'.
1573 my ($username, $expansion) = split ':';
1574 chomp $expansion;
1575 $expansion =~ s/^'(.*)'$/$1/;
1576 $expansion =~ s/^"(.*)"$/$1/;
1577
1578 # If it looks like the expansion has a real name already, then
1579 # we toss the username we got from CVS log. Otherwise, keep
1580 # it to use in combination with the email address.
1581
1582 if ($expansion =~ /^\s*<{0,1}\S+@.*/) {
1583 # Also, add angle brackets if none present
1584 if (! ($expansion =~ /<\S+@\S+>/)) {
1585 $expansions{$username} = "$username <$expansion>";
1586 }
1587 else {
1588 $expansions{$username} = "$username $expansion";
1589 }
1590 }
1591 else {
1592 $expansions{$username} = $expansion;
1593 }
1594 }
1595
1596 close (MAPFILE);
1597 }
1598
1599 return %expansions;
1600 }
1601
1602 sub parse_options ()
1603 {
1604 # Check this internally before setting the global variable.
1605 my $output_file;
1606
1607 # If this gets set, we encountered unknown options and will exit at
1608 # the end of this subroutine.
1609 my $exit_with_admonishment = 0;
1610
1611 while (my $arg = shift (@ARGV))
1612 {
1613 if ($arg =~ /^-h$|^-help$|^--help$|^--usage$|^-?$/) {
1614 $Print_Usage = 1;
1615 }
1616 elsif ($arg =~ /^--delta$/) {
1617 my $narg = shift(@ARGV) || die "$arg needs argument.\n";
1618 if ($narg =~ /^([A-Za-z][A-Za-z0-9_\-]*):([A-Za-z][A-Za-z0-9_\-]*)$/) {
1619 $Delta_From = $1;
1620 $Delta_To = $2;
1621 $Delta_Mode = 1;
1622 } else {
1623 die "--delta FROM_TAG:TO_TAG is what you meant to say.\n";
1624 }
1625 }
1626 elsif ($arg =~ /^--debug$/) { # unadvertised option, heh
1627 $Debug = 1;
1628 }
1629 elsif ($arg =~ /^--version$/) {
1630 $Print_Version = 1;
1631 }
1632 elsif ($arg =~ /^-g$|^--global-opts$/) {
1633 my $narg = shift (@ARGV) || die "$arg needs argument.\n";
1634 # Don't assume CVS is called "cvs" on the user's system:
1635 $Log_Source_Command =~ s/(^\S*)/$1 $narg/;
1636 }
1637 elsif ($arg =~ /^-l$|^--log-opts$/) {
1638 my $narg = shift (@ARGV) || die "$arg needs argument.\n";
1639 $Log_Source_Command .= " $narg";
1640 }
1641 elsif ($arg =~ /^-f$|^--file$/) {
1642 my $narg = shift (@ARGV) || die "$arg needs argument.\n";
1643 $output_file = $narg;
1644 }
1645 elsif ($arg =~ /^--accum$/) {
1646 $Cumulative = 1;
1647 }
1648 elsif ($arg =~ /^--fsf$/) {
1649 $FSF_Style = 1;
1650 }
1651 elsif ($arg =~ /^-U$|^--usermap$/) {
1652 my $narg = shift (@ARGV) || die "$arg needs argument.\n";
1653 $User_Map_File = $narg;
1654 }
1655 elsif ($arg =~ /^-W$|^--window$/) {
1656 defined(my $narg = shift (@ARGV)) || die "$arg needs argument.\n";
1657 $Max_Checkin_Duration = $narg;
1658 }
1659 elsif ($arg =~ /^-I$|^--ignore$/) {
1660 my $narg = shift (@ARGV) || die "$arg needs argument.\n";
1661 push (@Ignore_Files, $narg);
1662 }
1663 elsif ($arg =~ /^-C$|^--case-insensitive$/) {
1664 $Case_Insensitive = 1;
1665 }
1666 elsif ($arg =~ /^-R$|^--regexp$/) {
1667 my $narg = shift (@ARGV) || die "$arg needs argument.\n";
1668 $Regexp_Gate = $narg;
1669 }
1670 elsif ($arg =~ /^--stdout$/) {
1671 $Output_To_Stdout = 1;
1672 }
1673 elsif ($arg =~ /^--version$/) {
1674 $Print_Version = 1;
1675 }
1676 elsif ($arg =~ /^-d$|^--distributed$/) {
1677 $Distributed = 1;
1678 }
1679 elsif ($arg =~ /^-P$|^--prune$/) {
1680 $Prune_Empty_Msgs = 1;
1681 }
1682 elsif ($arg =~ /^-S$|^--separate-header$/) {
1683 $After_Header = "\n\n";
1684 }
1685 elsif ($arg =~ /^--no-wrap$/) {
1686 $No_Wrap = 1;
1687 }
1688 elsif ($arg =~ /^--gmt$|^--utc$/) {
1689 $UTC_Times = 1;
1690 }
1691 elsif ($arg =~ /^-w$|^--day-of-week$/) {
1692 $Show_Day_Of_Week = 1;
1693 }
1694 elsif ($arg =~ /^--no-times$/) {
1695 $Show_Times = 0;
1696 }
1697 elsif ($arg =~ /^-r$|^--revisions$/) {
1698 $Show_Revisions = 1;
1699 }
1700 elsif ($arg =~ /^-t$|^--tags$/) {
1701 $Show_Tags = 1;
1702 }
1703 elsif ($arg =~ /^-T$|^--tagdates$/) {
1704 $Show_Tag_Dates = 1;
1705 }
1706 elsif ($arg =~ /^-b$|^--branches$/) {
1707 $Show_Branches = 1;
1708 }
1709 elsif ($arg =~ /^-F$|^--follow$/) {
1710 my $narg = shift (@ARGV) || die "$arg needs argument.\n";
1711 push (@Follow_Branches, $narg);
1712 }
1713 elsif ($arg =~ /^--stdin$/) {
1714 $Input_From_Stdin = 1;
1715 }
1716 elsif ($arg =~ /^--header$/) {
1717 my $narg = shift (@ARGV) || die "$arg needs argument.\n";
1718 $ChangeLog_Header = &slurp_file ($narg);
1719 if (! defined ($ChangeLog_Header)) {
1720 $ChangeLog_Header = "";
1721 }
1722 }
1723 elsif ($arg =~ /^--xml-encoding$/) {
1724 my $narg = shift (@ARGV) || die "$arg needs argument.\n";
1725 $XML_Encoding = $narg ;
1726 }
1727 elsif ($arg =~ /^--xml$/) {
1728 $XML_Output = 1;
1729 }
1730 elsif ($arg =~ /^--hide-filenames$/) {
1731 $Hide_Filenames = 1;
1732 $After_Header = "";
1733 }
1734 elsif ($arg =~ /^--ignore-tag$/ ) {
1735 die "$arg needs argument.\n"
1736 unless @ARGV;
1737 push @ignore_tags, shift @ARGV;
1738 }
1739 else {
1740 # Just add a filename as argument to the log command
1741 $Log_Source_Command .= " '$arg'";
1742 }
1743 }
1744
1745 ## Check for contradictions...
1746
1747 if ($Output_To_Stdout && $Distributed) {
1748 print STDERR "cannot pass both --stdout and --distributed\n";
1749 $exit_with_admonishment = 1;
1750 }
1751
1752 if ($Output_To_Stdout && $output_file) {
1753 print STDERR "cannot pass both --stdout and --file\n";
1754 $exit_with_admonishment = 1;
1755 }
1756
1757 if ($XML_Output && $Cumulative) {
1758 print STDERR "cannot pass both --xml and --accum\n";
1759 $exit_with_admonishment = 1;
1760 }
1761
1762 # Or if any other error message has already been printed out, we
1763 # just leave now:
1764 if ($exit_with_admonishment) {
1765 &usage ();
1766 exit (1);
1767 }
1768 elsif ($Print_Usage) {
1769 &usage ();
1770 exit (0);
1771 }
1772 elsif ($Print_Version) {
1773 &version ();
1774 exit (0);
1775 }
1776
1777 ## Else no problems, so proceed.
1778
1779 if ($output_file) {
1780 $Log_File_Name = $output_file;
1781 }
1782 }
1783
1784 sub slurp_file ()
1785 {
1786 my $filename = shift || die ("no filename passed to slurp_file()");
1787 my $retstr;
1788
1789 open (SLURPEE, "<${filename}") or die ("unable to open $filename ($!)");
1790 my $saved_sep = $/;
1791 undef $/;
1792 $retstr = <SLURPEE>;
1793 $/ = $saved_sep;
1794 close (SLURPEE);
1795 return $retstr;
1796 }
1797
1798 sub debug ()
1799 {
1800 if ($Debug) {
1801 my $msg = shift;
1802 print STDERR $msg;
1803 }
1804 }
1805
1806 sub version ()
1807 {
1808 print "cvs2cl.pl version ${VERSION}; distributed under the GNU GPL.\n";
1809 }
1810
1811 sub usage ()
1812 {
1813 &version ();
1814 print <<'END_OF_INFO';
1815 Generate GNU-style ChangeLogs in CVS working copies.
1816
1817 Notes about the output format(s):
1818
1819 The default output of cvs2cl.pl is designed to be compact, formally
1820 unambiguous, but still easy for humans to read. It is largely
1821 self-explanatory, I hope; the one abbreviation that might not be
1822 obvious is "utags". That stands for "universal tags" -- a
1823 universal tag is one held by all the files in a given change entry.
1824
1825 If you need output that's easy for a program to parse, use the
1826 --xml option. Note that with XML output, just about all available
1827 information is included with each change entry, whether you asked
1828 for it or not, on the theory that your parser can ignore anything
1829 it's not looking for.
1830
1831 Notes about the options and arguments (the actual options are listed
1832 last in this usage message):
1833
1834 * The -I and -F options may appear multiple times.
1835
1836 * To follow trunk revisions, use "-F trunk" ("-F TRUNK" also works).
1837 This is okay because no would ever, ever be crazy enough to name a
1838 branch "trunk", right? Right.
1839
1840 * For the -U option, the UFILE should be formatted like
1841 CVSROOT/users. That is, each line of UFILE looks like this
1842 jrandom:jrandom@red-bean.com
1843 or maybe even like this
1844 jrandom:'Jesse Q. Random <jrandom@red-bean.com>'
1845 Don't forget to quote the portion after the colon if necessary.
1846
1847 * Many people want to filter by date. To do so, invoke cvs2cl.pl
1848 like this:
1849 cvs2cl.pl -l "-d'DATESPEC'"
1850 where DATESPEC is any date specification valid for "cvs log -d".
1851 (Note that CVS 1.10.7 and below requires there be no space between
1852 -d and its argument).
1853
1854 Options/Arguments:
1855
1856 -h, -help, --help, or -? Show this usage and exit
1857 --version Show version and exit
1858 -r, --revisions Show revision numbers in output
1859 -b, --branches Show branch names in revisions when possible
1860 -t, --tags Show tags (symbolic names) in output
1861 -T, --tagdates Show tags in output on their first occurance
1862 --stdin Read from stdin, don't run cvs log
1863 --stdout Output to stdout not to ChangeLog
1864 -d, --distributed Put ChangeLogs in subdirs
1865 -f FILE, --file FILE Write to FILE instead of "ChangeLog"
1866 --fsf Use this if log data is in FSF ChangeLog style
1867 -W SECS, --window SECS Window of time within which log entries unify
1868 -U UFILE, --usermap UFILE Expand usernames to email addresses from UFILE
1869 -R REGEXP, --regexp REGEXP Include only entries that match REGEXP
1870 -I REGEXP, --ignore REGEXP Ignore files whose names match REGEXP
1871 -C, --case-insensitive Any regexp matching is done case-insensitively
1872 -F BRANCH, --follow BRANCH Show only revisions on or ancestral to BRANCH
1873 -S, --separate-header Blank line between each header and log message
1874 --no-wrap Don't auto-wrap log message (recommend -S also)
1875 --gmt, --utc Show times in GMT/UTC instead of local time
1876 --accum Add to an existing ChangeLog (incompat w/ --xml)
1877 -w, --day-of-week Show day of week
1878 --no-times Don't show times in output
1879 --header FILE Get ChangeLog header from FILE ("-" means stdin)
1880 --xml Output XML instead of ChangeLog format
1881 --xml-encoding ENCODING Insert encoding clause in XML header
1882 --hide-filenames Don't show filenames (ignored for XML output)
1883 -P, --prune Don't show empty log messages
1884 -g OPTS, --global-opts OPTS Invoke like this "cvs OPTS log ..."
1885 -l OPTS, --log-opts OPTS Invoke like this "cvs ... log OPTS"
1886 FILE1 [FILE2 ...] Show only log information for the named FILE(s)
1887
1888 See http://www.red-bean.com/cvs2cl for maintenance and bug info.
1889 END_OF_INFO
1890 }
1891
1892 __END__
1893
1894 =head1 NAME
1895
1896 cvs2cl.pl - produces GNU-style ChangeLogs in CVS working copies, by
1897 running "cvs log" and parsing the output. Shared log entries are
1898 unified in an intuitive way.
1899
1900 =head1 DESCRIPTION
1901
1902 This script generates GNU-style ChangeLog files from CVS log
1903 information. Basic usage: just run it inside a working copy and a
1904 ChangeLog will appear. It requires repository access (i.e., 'cvs log'
1905 must work). Run "cvs2cl.pl --help" to see more advanced options.
1906
1907 See http://www.red-bean.com/cvs2cl for updates, and for instructions
1908 on getting anonymous CVS access to this script.
1909
1910 Maintainer: Karl Fogel <kfogel@red-bean.com>
1911 Please report bugs to <bug-cvs2cl@red-bean.com>.
1912
1913 =head1 README
1914
1915 This script generates GNU-style ChangeLog files from CVS log
1916 information. Basic usage: just run it inside a working copy and a
1917 ChangeLog will appear. It requires repository access (i.e., 'cvs log'
1918 must work). Run "cvs2cl.pl --help" to see more advanced options.
1919
1920 See http://www.red-bean.com/cvs2cl for updates, and for instructions
1921 on getting anonymous CVS access to this script.
1922
1923 Maintainer: Karl Fogel <kfogel@red-bean.com>
1924 Please report bugs to <bug-cvs2cl@red-bean.com>.
1925
1926 =head1 PREREQUISITES
1927
1928 This script requires C<Text::Wrap>, C<Time::Local>, and
1929 C<File::Basename>.
1930 It also seems to require C<Perl 5.004_04> or higher.
1931
1932 =pod OSNAMES
1933
1934 any
1935
1936 =pod SCRIPT CATEGORIES
1937
1938 Version_Control/CVS
1939
1940 =cut
1941
1942 -*- -*- -*- -*- -*- -*- -*- -*- -*- -*- -*- -*- -*- -*- -*- -*- -*- -*-
1943
1944 Note about a bug-slash-opportunity:
1945 -----------------------------------
1946
1947 There's a bug in Text::Wrap, which affects cvs2cl. This script
1948 reveals it:
1949
1950 #!/usr/bin/perl -w
1951
1952 use Text::Wrap;
1953
1954 my $test_text =
1955 "This script demonstrates a bug in Text::Wrap. The very long line
1956 following this paragraph will be relocated relative to the surrounding
1957 text:
1958
1959 ====================================================================
1960
1961 See? When the bug happens, we'll get the line of equal signs below
1962 this paragraph, even though it should be above.";
1963
1964 # Print out the test text with no wrapping:
1965 print "$test_text";
1966 print "\n";
1967 print "\n";
1968
1969 # Now print it out wrapped, and see the bug:
1970 print wrap ("\t", " ", "$test_text");
1971 print "\n";
1972 print "\n";
1973
1974 If the line of equal signs were one shorter, then the bug doesn't
1975 happen. Interesting.
1976
1977 Anyway, rather than fix this in Text::Wrap, we might as well write a
1978 new wrap() which has the following much-needed features:
1979
1980 * initial indentation, like current Text::Wrap()
1981 * subsequent line indentation, like current Text::Wrap()
1982 * user chooses among: force-break long words, leave them alone, or die()?
1983 * preserve existing indentation: chopped chunks from an indented line
1984 are indented by same (like this line, not counting the asterisk!)
1985 * optional list of things to preserve on line starts, default ">"
1986
1987 Note that the last two are essentially the same concept, so unify in
1988 implementation and give a good interface to controlling them.
1989
1990 And how about:
1991
1992 Optionally, when encounter a line pre-indented by same as previous
1993 line, then strip the newline and refill, but indent by the same.
1994 Yeah...
1995