rcs-fast-export.rb

#!/usr/bin/ruby

=begin
RCS fast export: run the script with the `--help` option for further
information.

No installation needed: you can run it from anywhere, including the git
checkout directory. For extra comfort, symlink it to some directory in
your PATH. I myself have this symlink:

	~/bin/rcs-fast-export -> ~/src/rcs-fast-export/rcs-fast-export.rb

allowing me to run `rcs-fast-export` from anywhere.
=end

=begin
TODO
	* Refactor commit coalescing
	* Add --strict-symbol-check to only coalesce commits if their symbol lists are equal
	* Add support for commitid for coalescing commits
	* Further coalescing options? (e.g. small logfile differences)
	* Proper branching support in multi-file export
	* Optimize memory usage by discarding unneeded text
	* Provide an option that marks a file as deleted based on symbolic revisions
=end

require 'pp'
require 'set'

class NoBranchSupport < NotImplementedError ; end

# Integer#odd? was introduced in Ruby 1.8.7, backport it to
# older versions
unless 2.respond_to? :odd?
	class Integer
		def odd?
			self % 2 == 1
		end
	end
end

# Set standard output to binary mode: git fast-import doesn't like Windows
# line-endings, and this ensures that the line termination will be a simple 0x0a
# on Windows too (it expands to 0x0D 0x0A otherwise).
STDOUT.binmode
STDIN.binmode

=begin
RCS fast-export version: set to `git` in the repository, but can be overridden
by packagers, e.g. based on the latest tag, git description, custom packager
patches or whatever.

When the version is set to `git`, we make a little effort to find more information
about which commit we are at.
=end

RFE_VERSION="git"

def version
	if RFE_VERSION == "git"
		nolinkfile = File.readlink(__FILE__) rescue __FILE__
		Dir.chdir File.expand_path File.dirname nolinkfile

		if File.exists? '.git' ; begin
			git_out = `git log -1 --pretty="%h %H%n%ai" | git name-rev --stdin`.split("\n")
			hash=git_out.first.split.first
			branch=git_out.first.split('(').last.chomp(')')
			date=git_out.last.split.first
			changed=`git diff --no-ext-diff --quiet --exit-code`
			branch << "*" unless $?.success?
			info=" [#{branch}] #{hash} (#{date})"
		rescue
			info=" (no info)"
		end ; end

		STDERR.puts "#{$0}: RCS fast-export, #{RFE_VERSION} version#{info}"
	else
		STDERR.puts "#{$0}: RCS fast-export, version #{RFE_VERSION}"
	end
end

def usage
	$stdout.flush
	STDERR.puts <<EOM
#{$0} [options] file [file ...]

Fast-export the RCS history of one or more files. If a directory is specified,
all RCS-tracked files in the directory and its descendants are exported.

When importing single files, their pathname is discarded during import. When
importing directories, only the specified directory component is discarded.

When importing a single file, RCS commits are converted one by one. Otherwise,
some heuristics is used to determine how to coalesce commits touching different
files.

Currently, commits are coalesced if they share the exact same log and if their
date differs by no more than the user-specified fuzziness. Additionally, the
symbols in one of the commit must be a subset of the symbols in the other
commit, unless --no-symbol-check is specified or rcs.symbolCheck is set to
false in the git configuration.

Typical usage:
    git init && rcs-fast-export.rb . | git fast-import && git reset

Options:
	--help, -h, -?		display this help text
	--authors-file, -A	specify a file containing username = Full Name <email> mappings
	--[no-]author-is-committer
				use the author name and date as committer identity
	--ignore		ignore the specified files (shell pattern)
	--log-encoding          specify the encoding of log messages, for transcoding to UTF-8
	--rcs-commit-fuzz	fuzziness in RCS commits to be considered a single one when
				importing multiple files
				(in seconds, defaults to 300, i.e. 5 minutes)
	--[no-]warn-missing-authors
				[do not] warn about usernames missing from the map file
	--[no-]symbol-check	[do not] check symbols when coalescing commits
	--[no-]tag-each-rev	[do not] create a lightweight tag for each RCS revision when
				importing a single file
	--[no-]log-filename	[do not] prepend the filename to the commit log when importing
				a single file
	--skip-branches		when exporting multiple files with a branched history, export
				the main branch only instead of aborting due to the lack of
				support for branched multi-file history export



Config options:
	rcs.authorsFile		for --authors-file
	rcs.authorIsCommitter	for --author-is-committer
	rcs.tagEachRev		for --tag-each-rev
	rcs.logFilename		for --log-filename
	rcs.commitFuzz		for --rcs-commit-fuzz
	rcs.warnMissingAuthors  for --warn-missing-authors
	rcs.symbolCheck		for --rcs-symbol-check
	rcs.tagFuzz		for --rcs-tag-fuzz

EOM
end

def warning(msg)
	$stdout.flush
	STDERR.puts msg
end

def not_found(arg)
	warning "Could not find #{arg}"
end

def emit_committer(opts, author, date)
	if opts[:author_is_committer]
		committer = "#{author} #{date}"
	else
		committer = `git var GIT_COMMITTER_IDENT`.chomp
	end
	puts "committer #{committer}"
end

# returns a hash that maps usernames to author names & emails
def load_authors_file(fn)
	hash = {}
	begin
		File.open(File.expand_path(fn)) do |io|
			io.each_line do |line|
				uname, author = line.split('=', 2)
				uname.strip!
				author.strip!
				warning "Username #{uname} redefined to #{author}" if hash.has_key? uname
				hash[uname] = author
			end
		end
	rescue
		not_found(fn)
	end
	return hash
end

def username_to_author(name, opts)
	map = opts[:authors]
	raise "no authors map defined" unless map and Hash === map

	# if name is not found in map, provide a default one, optionally giving a warning (once)
	unless map.key? name
		warning "no author found for #{name}" if opts[:warn_missing_authors]
		map[name] = "#{name} <empty>"
	end
	return map[name]
end

# display a message about a (recoverable) error
def alert(msg, action)
	STDERR.puts "ERROR:\t#{msg}"
	STDERR.puts "\t#{action}"
end

class Time
	def Time.rcs(string)
		fields = string.split('.')
		raise ArgumentError, "wrong number of fields for RCS date #{string}" unless fields.length == 6
		# in Ruby 1.9, '99' is interpreted as year 99, not year 1999
		if fields.first.length < 3
			fields.first.insert 0, '19'
		end
		Time.utc(*fields)
	end
end

module RCS
	# strip an optional final ;
	def RCS.clean(arg)
		arg.chomp(';')
	end

	# strip the first and last @, and de-double @@s
	def RCS.sanitize(arg)
		case arg
		when Array
			ret = arg.dup
			raise 'malformed first line' unless ret.first[0,1] == '@'
			raise 'malformed last line' unless ret.last[-1,1] == '@'
			ret.first.sub!(/^@/,'')
			ret.last.sub!(/@$/,'')
			ret.map { |l| l.gsub('@@','@') }
		when String
			arg.chomp('@').sub(/^@/,'').gsub('@@','@')
		else
			raise
		end
	end

	# clean and sanitize
	def RCS.at_clean(arg)
		RCS.sanitize RCS.clean(arg)
	end

	def RCS.mark(key)
		@@marks ||= {}
		if @@marks.key? key
			@@marks[key]
		else
			@@marks[key] = @@marks.length + 1
		end
	end

	def RCS.blob(file, rev)
		RCS.mark([file, rev])
	end

	def RCS.commit(commit)
		RCS.mark(commit)
	end

	class File
		attr_accessor :head, :comment, :desc, :revision, :fname, :mode
		def initialize(fname, executable)
			@fname = fname.dup
			@head = nil
			@comment = nil
			@desc = []
			@revision = Hash.new { |h, r| h[r] = Revision.new(self, r) }
			@mode = executable ? '755' : '644'
		end

		def has_revision?(rev)
			@revision.has_key?(rev) and not @revision[rev].author.nil?
		end

		def export_commits(opts={})
			counter = 0
			exported = []
			log_enc = opts[:log_encoding]
			until @revision.empty?
				counter += 1

				# a string sort is a very good candidate for
				# export order, getting a miss only for
				# multi-digit revision components
				keys = @revision.keys.sort

				warning "commit export loop ##{counter}"
				warning "\t#{exported.length} commits exported so far: #{exported.join(', ')}" unless exported.empty?
				warning "\t#{keys.size} to export: #{keys.join(', ')}"

				keys.each do |key|
					rev = @revision[key]
					# the parent commit is rev.next if we're on the
					# master branch (rev.branch is nil) or
					# rev.diff_base otherwise
					from = rev.branch.nil? ? rev.next : rev.diff_base
					# A commit can only be exported if it has no
					# parent, or if the parent has been exported
					# already. Skip this commit otherwise
					if from and not exported.include? from
						next
					end

					branch = rev.branch || 'master'
					author = username_to_author(rev.author, opts)
					date = "#{rev.date.tv_sec} +0000"
					log = String.new
					if opts[:log_filename]
						log << @fname << ": "
					end
					if log_enc
						# git fast-import expects logs to be in UTF-8, so if a different log encoding
						# is specified for the log we transcode from whatever was specified to UTF-8.
						# we then mark the string as ASCII-8BIT (as everything else) so that string
						# lengths are computed in bytes
						log << rev.log.join.encode('UTF-8', log_enc).force_encoding('ASCII-8BIT')
					else
						log << rev.log.join
					end

					puts "commit refs/heads/#{branch}"
					puts "mark :#{RCS.commit key}"
					puts "author #{author} #{date}"
					emit_committer(opts, author, date)
					puts "data #{log.length}"
					puts log unless log.empty?
					puts "from :#{RCS.commit from}" if from
					puts "M #{@mode} :#{RCS.blob @fname, key} #{@fname}"

					# TODO FIXME this *should* be safe, in
					# that it should not unduly move
					# branches back in time, but I'm not
					# 100% sure ...
					rev.branches.each do |sym|
						puts "reset refs/heads/#{sym}"
						puts "from :#{RCS.commit key}"
					end
					rev.symbols.each do |sym|
						puts "reset refs/tags/#{sym}"
						puts "from :#{RCS.commit key}"
					end
					if opts[:tag_each_rev]
						puts "reset refs/tags/#{key}"
						puts "from :#{RCS.commit key}"
					end

					exported.push key
				end
				exported.each { |k| @revision.delete(k) }
			end
		end
	end

	class Revision
		attr_accessor :rev, :author, :state, :next
		attr_accessor :branches, :log, :text, :symbols
		attr_accessor :branch, :diff_base, :branch_point
		attr_reader   :date
		def initialize(file, rev)
			@file = file
			@rev = rev
			@author = nil
			@date = nil
			@state = nil
			@next = nil
			@branches = Set.new
			@branch = nil
			@branch_point = nil
			@diff_base = nil
			@log = []
			@text = []
			@symbols = Set.new
		end

		def date=(str)
			@date = Time.rcs(str)
		end

		def blob
			str = @text.join('')
			ret = "blob\nmark :#{RCS.blob @file.fname, @rev}\ndata #{str.length}\n#{str}\n"
			ret
		end
	end

	# TODO: what if a revision does not end with newline?
	# TODO this should be done internally, not piping out to RCS
	def RCS.expand_keywords(rcsfile, revision)
		ret = ::File.read("|co -q -p#{revision} #{rcsfile}")
		lines = []
		ret.each_line do |line|
			lines << line
		end
		lines
	end

	def RCS.parse(fname, rcsfile, opts={})
		csrcsType = 0
		rcs = RCS::File.new(fname, ::File.executable?(rcsfile))

		::File.open(rcsfile, 'r:ASCII-8BIT') do |file|
			status = [:basic]
			rev = nil
			lines = []
			difflines = []
			file.each_line do |line|
				case status.last
				when :basic
					command, args = line.split($;,2)
					next if command.empty?

					if command.chomp!(';')
						warning "Skipping empty command #{command.inspect}" if $DEBUG
						next
					end

					case command
					when 'head'
						rcs.head = RCS.clean(args.chomp)
					when 'symbols'
						status.push :symbols
						next if args.empty?
						line = args; redo
					when 'comment'
						rcs.comment = RCS.at_clean(args.chomp)
					when /^[0-9.]+$/
						rev = command.dup
						if rcs.has_revision?(rev)
							status.push :revision_data
						else
							status.push :new_revision
						end
					when 'desc'
						status.push :desc
						lines.clear
						status.push :read_lines
					when 'branch', 'access', 'locks', 'expand'
						warning "Skipping unhandled command #{command.inspect}" if $DEBUG
						status.push :skipping_lines
						next if args.empty?
						line = args; redo
					else
						raise "Unknown command #{command.inspect}"
					end
				when :skipping_lines
					status.pop if line.strip.chomp!(';')
				when :symbols
					# we can have multiple symbols per line
					pairs = line.strip.split($;)
					pairs.each do |pair|
						sym, rev = pair.strip.split(':',2);
						if rev
							status.pop if rev.chomp!(';')
							rcs.revision[rev].symbols << sym
						else
							status.pop
						end
					end
				when :desc
					rcs.desc.replace lines.dup
					status.pop
				when :read_lines
					# we sanitize lines as we read them

					actual_line = line.dup

					# the first line must begin with a @, which we strip
					if lines.empty?
						ats = line.match(/^@+/)
						raise 'malformed line' unless ats
						actual_line.replace line.sub(/^@/,'')
					end

					# if the line ends with an ODD number of @, it's the
					# last line -- we work on actual_line so that content
					# such as @\n or @ work correctly (they would be
					# encoded respectively as ['@@@\n','@\n'] and
					# ['@@@@\n']
					ats = actual_line.chomp.match(/@+$/)
					if nomore = (ats && Regexp.last_match(0).length.odd?)
						actual_line.replace actual_line.chomp.sub(/@$/,'')
					end
					lines << actual_line.gsub('@@','@')
					if nomore
						status.pop
						redo
					end
				when :new_revision
					case line.chomp
					when /^date\s+(\S+);\s+author\s+(\S+);\s+state\s+(\S+);$/
						rcs.revision[rev].date = $1
						rcs.revision[rev].author = $2
						rcs.revision[rev].state = $3
					when /^branches\s*;/
						next
					when /^branches(?:\s+|$)/
						status.push :branches
						if line.index(';')
							line = line.sub(/^branches\s+/,'')
							redo
						end
					when /^next\s+(\S+)?;$/
						nxt = rcs.revision[rev].next = $1
						next unless nxt
						raise "multiple diff_bases for #{nxt}" unless rcs.revision[nxt].diff_base.nil?
						rcs.revision[nxt].diff_base = rev
						rcs.revision[nxt].branch = rcs.revision[rev].branch
					else
						status.pop
					end
				when :branches
					candidate = line.split(';',2)
					candidate.first.strip.split.each do |branch|
						raise "multiple diff_bases for #{branch}" unless rcs.revision[branch].diff_base.nil?
						rcs.revision[branch].diff_base = rev
						# we drop the last number from the branch name
						rcs.revision[branch].branch = branch.sub(/\.\d+$/,'.x')
						rcs.revision[branch].branch_point = rev
					end
					status.pop if candidate.length > 1
				when :revision_data
					case line.chomp
					when 'log'
						status.push :log
						lines.clear
						status.push :read_lines
					when 'text'
						if rev == rcs.head
							status.push :head
						else
							status.push :diff
						end
						lines.clear
						status.push :read_lines
					else
						status.pop
					end
				when :log
					rcs.revision[rev].log.replace lines.dup
					status.pop
				when :head
#AMJ - Begin changes for CS-RCS
# You can probably do something clever with regex matching here, but simple string compare is fine.
					if lines.first.chomp.eql?('))[(###%%%~~~StoreCompleteRevision))[(###%%%~~~')
						# This can be modularised, because the code repeats, I just wanted it to work.
						csrcsType = 1 # Remember what type this is
						myBinFile = rcsfile + ',f/' + rev + '_1'
						myBinLines = ::File.open(myBinFile, 'rb') { |f| f.read }
						rcs.revision[rev].text.replace [myBinLines]
					elsif lines.first.chomp.eql?('))[(###%%%~~~StoreZippedRevision))[(###%%%~~~')
						csrcsType = 2 # Remember what type this is
						myZipFile = rcsfile + ',f/' + rev + '_2'
						myZipBase = ::File::basename(rcsfile, ',v')
						myZipCmd = "unzip -p #{myZipFile} #{myZipBase}"
						myZipLines = IO.binread("|#{myZipCmd}")
						rcs.revision[rev].text.replace [myZipLines]
#AMJ - End changes
					elsif opts[:expand_keywords]
						rcs.revision[rev].text.replace RCS.expand_keywords(rcsfile, rev)
					else
						rcs.revision[rev].text.replace lines.dup
					end
					puts rcs.revision[rev].blob
					status.pop
				when :diff
#AMJ - Begin changes for CS-RCS
# A bit lazy here - I'm not checking that the "diff" is actually "@@", because it always is,
# in the files I've seen. I don't think this will likely be false, but an error in case it is?
					if csrcsType == 1
						# Again, you could modularise this. Exactly the same as the code above.
						myBinFile = rcsfile + ',f/' + rev + '_1'
						myBinLines = ::File.open(myBinFile, 'rb') { |f| f.read }
						rcs.revision[rev].text.replace [myBinLines]
					elsif csrcsType == 2
						myZipFile = rcsfile + ',f/' + rev + '_2'
						myZipBase = ::File::basename(rcsfile, ',v')
						myZipCmd = "unzip -p #{myZipFile} #{myZipBase}"
						myZipLines = IO.binread("|#{myZipCmd}")
						rcs.revision[rev].text.replace [myZipLines]
#AMJ - End changes
					elsif opts[:expand_keywords]
						rcs.revision[rev].text.replace RCS.expand_keywords(rcsfile, rev)
					else
						difflines.replace lines.dup
						difflines.pop if difflines.last.empty?
						if difflines.first.chomp.empty?
							alert "malformed diff: empty initial line @ #{rcsfile}:#{file.lineno-difflines.length-1}", "skipping"
							difflines.shift
						end unless difflines.empty?
						base = rcs.revision[rev].diff_base
						unless rcs.revision[base].text
							pp rcs
							puts rev, base
							raise 'no diff base!'
						end
						# deep copy
						buffer = []
						rcs.revision[base].text.each { |l| buffer << [l.dup] }

						adding = false
						index = nil
						count = nil

						while l = difflines.shift
							if adding
								raise 'negative index during insertion' if index < 0
								raise 'negative count during insertion' if count < 0
								adding << l
								count -= 1
								# collected all the lines, put the before
								unless count > 0
									unless buffer[index]
										buffer[index] = []
									end
									buffer[index].unshift(*adding)
									adding = false
								end
								next
							end

							l.chomp!
							raise "malformed diff @ #{rcsfile}:#{file.lineno-difflines.length-1} `#{l}`" unless l =~ /^([ad])(\d+) (\d+)$/
							diff_cmd = $1.intern
							index = $2.to_i
							count = $3.to_i
							case diff_cmd
							when :d
								# for deletion, index 1 is the first index, so the Ruby
								# index is one less than the diff one
								index -= 1
								# we replace them with empty string so that 'a' commands
								# referring to the same line work properly
								while count > 0
									buffer[index].clear
									index += 1
									count -= 1
								end
							when :a
								# addition will prepend the appropriate lines
								# to the given index, and in this case Ruby
								# and diff indices are the same
								adding = []
							end
						end

						# turn the buffer into an array of lines, deleting the empty ones
						buffer.delete_if { |l| l.empty? }
						buffer.flatten!

						rcs.revision[rev].text = buffer
					end
					puts rcs.revision[rev].blob
					status.pop
				else
					raise "Unknown status #{status.last}"
				end
			end
		end

		# clean up the symbols/branches: look for revisions that have
		# one or more symbols but no dates, and make them into
		# branches, pointing to the highest commit with that key
		branches = []
		keys = rcs.revision.keys
		rcs.revision.each do |key, rev|
			if rev.date.nil? and not rev.symbols.empty?
				top = keys.select { |k| k.match(/^#{key}\./) }.sort.last
				tr = rcs.revision[top]
				raise "unhandled complex branch structure met: #{rev.inspect} refers #{tr.inspect}" if tr.date.nil?
				tr.branches |= rev.symbols
				branches << key
			end
		end
		branches.each { |k| rcs.revision.delete k }

		return rcs
	end

	class Tree
		def initialize(commit)
			@commit = commit
			@files = Hash.new
		end

		def merge!(tree)
			testfiles = @files.dup
			tree.each { |rcs, rev| self.add(rcs, rev, testfiles) }
			# the next line is only reached if all the adds were
			# successful, so the merge is atomic
			@files.replace testfiles
		end

		def add(rcs, rev, file_list=@files)
			if file_list.key? rcs
				prev = file_list[rcs]
				if prev.log == rev.log
					str = "re-adding existing file #{rcs.fname} (old: #{prev.rev}, new: #{rev.rev})"
				else
					str = "re-adding existing file #{rcs.fname} (old: #{[prev.rev, prev.log.to_s].inspect}, new: #{[rev.rev, rev.log.to_s].inspect})"
				end
				if prev.text != rev.text
					raise str
				else
					@commit.warn_about str
				end
			end
			file_list[rcs] = rev
		end

		def each &block
			@files.each &block
		end

		def to_a
			files = []
			@files.map do |rcs, rev|
				if rev.state.downcase == "dead"
					files << "D #{rcs.fname}"
				else
					files << "M #{rcs.mode} :#{RCS.blob rcs.fname, rev.rev} #{rcs.fname}"
				end
			end
			files
		end

		def filenames
			@files.map { |rcs, rev| rcs.fname }
		end

		def to_s
			self.to_a.join("\n")
		end
	end

	class Commit
		attr_accessor :date, :log, :symbols, :author, :branch
		attr_accessor :tree
		attr_accessor :min_date, :max_date
		def initialize(rcs, rev)
			raise NoBranchSupport if rev.branch
			self.date = rev.date.dup
			self.min_date = self.max_date = self.date
			self.log = rev.log.dup
			self.symbols = rev.symbols.dup
			self.author = rev.author
			self.branch = rev.branch

			self.tree = Tree.new self
			self.tree.add rcs, rev
		end

		def to_a
			[self.min_date, self.date, self.max_date, self.branch, self.symbols, self.author, self.log, self.tree.to_a]
		end

		def warn_about(str)
			warn str + " for commit on #{self.date}"
		end

		# Sort by date and then by number of symbols
		def <=>(other)
			ds = self.date <=> other.date
			if ds != 0
				return ds
			else
				return self.symbols.length <=> other.symbols.length
			end
		end

		def merge!(commit)
			self.tree.merge! commit.tree
			if commit.max_date > self.max_date
				self.max_date = commit.max_date
			end
			if commit.min_date < self.min_date
				self.min_date = commit.min_date
			end
			self.symbols.merge commit.symbols
		end

		def export(opts={})
			xbranch = self.branch || 'master'
			xauthor = username_to_author(self.author, opts)
			xlog = self.log.join
			numdate = self.date.tv_sec
			xdate = "#{numdate} +0000"
			key = numdate.to_s

			puts "commit refs/heads/#{xbranch}"
			puts "mark :#{RCS.commit key}"
			puts "author #{xauthor} #{xdate}"
			emit_committer(opts, xauthor, xdate)
			puts "data #{xlog.length}"
			puts xlog unless xlog.empty?
			# TODO branching support for multi-file export
			# puts "from :#{RCS.commit from}" if self.branch_point
			puts self.tree.to_s

			# TODO branching support for multi-file export
			# rev.branches.each do |sym|
			# 	puts "reset refs/heads/#{sym}"
			# 	puts "from :#{RCS.commit key}"
			# end

			self.symbols.each do |sym|
				puts "reset refs/tags/#{sym}"
				puts "from :#{RCS.commit key}"
			end

		end
	end
end

require 'getoptlong'

opts = GetoptLong.new(
	# Authors file, like git-svn and git-cvsimport, more than one can be
	# specified
	['--authors-file', '-A', GetoptLong::REQUIRED_ARGUMENT],
	# Use author identity as committer identity?
	['--author-is-committer', GetoptLong::NO_ARGUMENT],
	['--no-author-is-committer', GetoptLong::NO_ARGUMENT],
	# Use "co" to obtain the actual revision with keywords expanded.
	['--expand-keywords', GetoptLong::NO_ARGUMENT],
	# RCS file suffix, like RCS
	['--rcs-suffixes', '-x', GetoptLong::REQUIRED_ARGUMENT],
	# Shell pattern to identify files to be ignored
	['--ignore', GetoptLong::REQUIRED_ARGUMENT],
	# Encoding of log messages in the RCS files
	['--log-encoding', GetoptLong::REQUIRED_ARGUMENT],
	# Date fuzziness for commits to be considered the same (in seconds)
	['--rcs-commit-fuzz', GetoptLong::REQUIRED_ARGUMENT],
	# warn about usernames missing in authors file map?
	['--warn-missing-authors', GetoptLong::NO_ARGUMENT],
	['--no-warn-missing-authors', GetoptLong::NO_ARGUMENT],
	# check symbols when coalescing?
	['--symbol-check', GetoptLong::NO_ARGUMENT],
	['--no-symbol-check', GetoptLong::NO_ARGUMENT],
	# tag each revision?
	['--tag-each-rev', GetoptLong::NO_ARGUMENT],
	['--no-tag-each-rev', GetoptLong::NO_ARGUMENT],
	# prepend filenames to commit logs?
	['--log-filename', GetoptLong::NO_ARGUMENT],
	['--no-log-filename', GetoptLong::NO_ARGUMENT],
	# skip branches when exporting a whole tree?
	['--skip-branches', GetoptLong::NO_ARGUMENT],
	# show current version
	['--version', '-v', GetoptLong::NO_ARGUMENT],
	# show help/usage
	['--help', '-h', '-?', GetoptLong::NO_ARGUMENT]
)

# We read options in order, but they apply to all passed parameters.
# TODO maybe they should only apply to the following, unless there's only one
# file?
opts.ordering = GetoptLong::RETURN_IN_ORDER

file_list = []
parse_options = {
	:authors => Hash.new,
	:ignore => Array.new,
	:commit_fuzz => 300,
	:tag_fuzz => -1,
}

# Read config options
`git config --get-all rcs.authorsfile`.each_line do |fn|
	parse_options[:authors].merge! load_authors_file(fn.chomp)
end

parse_options[:author_is_committer] = (
	`git config --bool rcs.authoriscommitter`.chomp == 'false'
) ? false : true

parse_options[:tag_each_rev] = (
	`git config --bool rcs.tageachrev`.chomp == 'true'
) ? true : false

parse_options[:log_filename] = (
	`git config --bool rcs.logfilename`.chomp == 'true'
) ? true : false

fuzz = `git config --int rcs.commitFuzz`.chomp
parse_options[:commit_fuzz] = fuzz.to_i unless fuzz.empty?

fuzz = `git config --int rcs.tagFuzz`.chomp
parse_options[:tag_fuzz] = fuzz.to_i unless fuzz.empty?

parse_options[:symbol_check] = (
	`git config --bool rcs.symbolcheck`.chomp == 'false'
) ? false : true

parse_options[:warn_missing_authors] = (
	`git config --bool rcs.warnmissingauthors`.chomp == 'false'
) ? false : true

opts.each do |opt, arg|
	case opt
	when '--authors-file'
		authors = load_authors_file(arg)
		redef = parse_options[:authors].keys & authors.keys
		warning "Authors file #{arg} redefines #{redef.join(', ')}" unless redef.empty?
		parse_options[:authors].merge!(authors)
	when '--author-is-committer'
		parse_options[:author_is_committer] = true
	when '--no-author-is-committer'
		parse_options[:author_is_committer] = false
	when '--expand-keywords'
		parse_options[:expand_keywords] = true
	when '--rcs-suffixes'
		# TODO
	when '--ignore'
		parse_options[:ignore] << arg
	when '--log-encoding'
		parse_options[:log_encoding] = Encoding.find(arg)
	when '--rcs-commit-fuzz'
		parse_options[:commit_fuzz] = arg.to_i
	when '--rcs-tag-fuzz'
		parse_options[:tag_fuzz] = arg.to_i
	when '--symbol-check'
		parse_options[:symbol_check] = true
	when '--no-symbol-check'
		parse_options[:symbol_check] = false
	when '--tag-each-rev'
		parse_options[:tag_each_rev] = true
	when '--no-tag-each-rev'
		# this is the default, which is fine since the missing key
		# (default) returns nil which is false in Ruby
		parse_options[:tag_each_rev] = false
	when '--log-filename'
		parse_options[:log_filename] = true
	when '--no-log-filename'
		# this is the default, which is fine since the missing key
		# (default) returns nil which is false in Ruby
		parse_options[:log_filename] = false
	when '--skip-branches'
		parse_options[:skip_branches] = true
	when ''
		file_list << arg
	when '--version'
		version
		exit
	when '--help'
		usage
		exit
	end
end

if parse_options[:tag_fuzz] < parse_options[:commit_fuzz]
	parse_options[:tag_fuzz] = parse_options[:commit_fuzz]
end

require 'etc'

user = Etc.getlogin || ENV['USER']

# steal username/email data from other init files that may contain the
# information
def steal_username
	[
		# the user's .hgrc file for a username field
		['~/.hgrc',   /^\s*username\s*=\s*(["'])?(.*)\1$/,       2],
		# the user's .(g)vimrc for a changelog_username setting
		['~/.vimrc',  /changelog_username\s*=\s*(["'])?(.*)\1$/, 2],
		['~/.gvimrc', /changelog_username\s*=\s*(["'])?(.*)\1$/, 2],
	].each do |fn, rx, idx|
		file = File.expand_path fn
		if File.readable?(file) and File.read(file) =~ rx
			parse_options[:authors][user] = Regexp.last_match(idx).strip
			break
		end
	end
end

if user and not user.empty? and not parse_options[:authors].has_key?(user)
	name = ENV['GIT_AUTHOR_NAME'] || ''
	name.replace(`git config user.name`.chomp) if name.empty?
	name.replace(Etc.getpwnam(user).gecos) if name.empty?

	if name.empty?
		# couldn't find a name, try to steal data from other sources
		steal_username
	else
		# if we found a name, try to find an email too
		email = ENV['GIT_AUTHOR_EMAIL'] || ''
		email.replace(`git config user.email`.chomp) if email.empty?

		if email.empty?
			# couldn't find an email, try to steal data too
			steal_username
		else
			# we got both a name and email, fill the info
			parse_options[:authors][user] = "#{name} <#{email}>"
		end
	end
end

if file_list.empty?
	usage
	exit 1
end

SFX = ',v'

status = 0

rcs = []
file_list.each do |arg|
	case ftype = File.ftype(arg)
	when 'file'
		if arg[-2,2] == SFX
			if File.exists? arg
				rcsfile = arg.dup
			else
				not_found "RCS file #{arg}"
				status |= 1
			end
			filename = File.basename(arg, SFX)
		else
			filename = File.basename(arg)
			path = File.dirname(arg)
			rcsfile = File.join(path, 'RCS', filename) + SFX
			unless File.exists? rcsfile
				rcsfile.replace File.join(path, filename) + SFX
				unless File.exists? rcsfile
					not_found "RCS file for #{filename} in #{path}"
				end
			end
		end
		rcs << RCS.parse(filename, rcsfile, parse_options)
	when 'directory'
		argdirname = arg.chomp(File::SEPARATOR)
		pattern = File.join(argdirname, '**', '*' + SFX)
		Dir.glob(pattern, File::FNM_DOTMATCH).each do |rcsfile|
			filename = File.basename(rcsfile, SFX)
			path = File.dirname(rcsfile)
			# strip trailing "/RCS" if present, or "RCS" if that's
			# the full path
			path.sub!(/(^|#{File::SEPARATOR})RCS$/, '')
			# strip off the portion of the path specified
			# on the command line from the front of the path
			# (or delete the path completely if it is the same
			# as the specified directory)
			path.sub!(/^#{Regexp.escape argdirname}(#{File::SEPARATOR}|$)/, '')
			filename = File.join(path, filename) unless path.empty?

			# skip file if it's to be ignored
			unless parse_options[:ignore].empty?
				ignored = false
				parse_options[:ignore].each do |pat|
					if File.fnmatch?(pat, filename, File::FNM_PATHNAME)
						ignored = true
						break
					end
				end
				next if ignored
			end

			# proceed
			begin
				rcs << RCS.parse(filename, rcsfile, parse_options)
			rescue Exception => e
				warning "Failed to parse #{filename} @ #{rcsfile}:#{$.}"
				raise e
			end
		end
	else
		warning "Cannot handle #{arg} of #{ftype} type"
		status |= 1
	end
end

if rcs.length == 1
	rcs.first.export_commits(parse_options)
else
	warning "Preparing commits"

	commits = []

	rcs.each do |r|
		r.revision.each do |k, rev|
			begin
				commits << RCS::Commit.new(r, rev)
			rescue NoBranchSupport
				if parse_options[:skip_branches]
					warning "Skipping revision #{rev.rev} for #{r.fname} (branch)"
				else raise
				end
			end
		end
	end

	warning "Sorting by date"

	commits.sort!

	if $DEBUG
		warning "RAW commits (#{commits.length}):"
		commits.each do |c|
			PP.pp c.to_a, $stderr
		end
	else
		warning "#{commits.length} single-file commits"
	end

	warning "Coalescing [1] by date with fuzz #{parse_options[:commit_fuzz]}"

	thisindex = commits.size
	commits.reverse_each do |c|
		nextindex = thisindex
		thisindex -= 1

		cfiles = Set.new c.tree.filenames
		ofiles = Set.new

		mergeable = []

		# test for mergeable commits by looking at following commits
		while nextindex < commits.size
			k = commits[nextindex]
			nextindex += 1

			# commits are date-sorted, so we know we can quit early if we are too far
			# for coalescing to work
			break if k.min_date > c.max_date + parse_options[:commit_fuzz]

			skipthis = false

			kfiles = Set.new k.tree.filenames

			if c.log != k.log or c.author != k.author or c.branch != k.branch
				skipthis = true
			end

			unless c.symbols.subset?(k.symbols) or k.symbols.subset?(c.symbols)
				cflist = cfiles.to_a.join(', ')
				kflist = kfiles.to_a.join(', ')
				if parse_options[:symbol_check]
					warning "Not coalescing #{c.log.inspect}\n\tfor (#{cflist})\n\tand (#{kflist})"
					warning "\tbecause their symbols disagree:\n\t#{c.symbols.to_a.inspect} and #{k.symbols.to_a.inspect} disagree on #{(c.symbols ^ k.symbols).to_a.inspect}"
					warning "\tretry with the --no-symbol-check option if you want to merge these commits anyway"
					skipthis = true
				elsif $DEBUG
					warning "Coalescing #{c.log.inspect}\n\tfor (#{cflist})\n\tand (#{kflist})"
					warning "\twith disagreeing symbols:\n\t#{c.symbols.to_a.inspect} and #{k.symbols.to_a.inspect} disagree on #{(c.symbols ^ k.symbols).to_a.inspect}"
				end
			end

			# keep track of filenames touched by commits we are not merging with,
			# since we don't want to merge with commits that touch them, to preserve
			# the monotonicity of history for each file
			# TODO we could forward-merge with them, unless some of our files were
			# touched too.
			if skipthis
				# if the candidate touches any file already in the commit,
				# we can stop looking forward
				break unless cfiles.intersection(kfiles).empty?
				ofiles |= kfiles
				next
			end

			# the candidate has the same log, author, branch and appropriate symbols
			# does it touch anything in ofiles?
			unless ofiles.intersection(kfiles).empty?
				if $DEBUG
					cflist = cfiles.to_a.join(', ')
					kflist = kfiles.to_a.join(', ')
					oflist = ofiles.to_a.join(', ')
					warning "Not coalescing #{c.log.inspect}\n\tfor (#{cflist})\n\tand (#{kflist})"
					warning "\tbecause the latter intersects #{oflist} in #{(ofiles & kfiles).to_a.inspect}"
				end
				next
			end

			mergeable << k
		end

		mergeable.each do |k|
			begin
				c.merge! k
			rescue RuntimeError => err
				fuzz = c.date - k.date
				warning "Fuzzy commit coalescing failed: #{err}"
				warning "\tretry with commit fuzz < #{fuzz} if you don't want to see this message"
				break
			end
			commits.delete k
		end
	end

	if $DEBUG
		warning "[1] commits (#{commits.length}):"
		commits.each do |c|
			PP.pp c.to_a, $stderr
		end
	else
		warning "#{commits.length} coalesced commits"
	end

	commits.each { |c| c.export(parse_options) }

end

exit status