Hi,
On Fri, 4 Feb 2005, Larry McVoy wrote:
> > > > > CVS BitKeeper [*]
> > > > > Deltas 235,956 280,212
>
> You need to rethink your math, you are way off. I'll explain it so that
> the rest of the people can see this is just pure FUD.
>
> To make sure this was apples to apples I went to the BK and CVS trees
> which are in lock step and reran the numbers:
>
> CVS BitKeeper % in CVS
> file deltas 210,609 218,742 96%
> changsets 26,603 59,220 44%
>
> You are not factoring out the ChangeSet deltas, which are BK metadata,
> they aren't part of the source tree. If you remove those then the
> difference between CVS and BK revision history is 209K deltas vs
> 221K deltas.
>
> In other words, the CVS tree is missing no more than 4% of the deltas
> to the source files.
>
> READ THAT AGAIN, PLEASE.
>
> The CVS tree has 96% of all the deltas to all your source files. 96%.
Before you start shouting, how about you get your numbers right? Your
first numbers were wrong and the second numbers are still wrong:
$ find -name \*,v -a ! -path ./BitKeeper\* -a ! -name ChangeSet,v | xargs rlog
| egrep -e '^[0-9]{4}(/[0-9]{2}){2} [0-9]{2}(:[0-9]{2}){2}' -e '\(Logical
change 1.[0-9]+\)' | wc -l
219242
$ find -name \*,v -a ! -path ./BitKeeper\* -a ! -name ChangeSet,v | xargs rlog
| egrep '\(Logical change 1.[0-9]+\)' | wc -l
187576
That would bring us back to 85% and I leave it to you to find the bug.
> My good friend Stelian would have you believe that you are missing 50%
> of your data when in fact you are missing NONE of your data, you have
> ALL of your data in an almost the identical form.
The 44% number deserves an explanation, which our good friend Larry would
have you believe is completely unimportant and thinking otherwise is just
"FUD". Actually on the contrary it is the more important number.
The 85% number means that the history of a _single_ file can be restored
with a precision of 85% via bkcvs. If one is only interested in the
history of a single file, that 85% might be sufficient. OTOH a change to
the kernel is rarely just a change to a single file. A patch can consist
of changes to multiple files and this relationship between multiple single
file changes can only be restored with a precision of 44% via bkcvs.
IOW if one wants to not just look at some random file history, but
actually restore and use some of it, one can only get 44% of the 59220
snapshots of kernel history.
> What Stelian is complaining about is the patch set which is easily
> extractable from CVS is at a coarser granularity than the patch set
> extractable from BK. That's true but so what? Before BK you only
> a handful of patches between releases, now you have thousands.
Hmm, that argument is not exactly original anymore. If Linus would say,
that kernel history has to be exclusively recorded via bk and that the
history available via bkcvs is the maximum that must be made available to
other users, you would have a point.
So it's quite legitimate to at least ask occasionally how the 44%
granularity can be improved. CVS is obviously not really capable of it,
but RCS on which CVS is based on is. I attached an example script which
can convert the history of a bk file into RCS. The only thing that is
missing is to 1) generate a list revisions with their parents, 2) checkout
a given revision, 3) generate the log for a given revision. (*) I'm quite
sure that's trivial to do for you and this only took me a few hours
including testing and cleanup, it only takes a little more work to make
CVS happy and allow incremental updates, so your "3 month" number is
absolutely ridiculous.
Any bk user can now easily prove or disprove this by completing my example
script, I think the "bk prs" man page should be very helpful here. If you
want to complete it into your very own bk2cvs script, I'd be happy to
help.
bye, Roman
(*) In case anyone thinks this is simple without bk, try to find the
parents for this changeset
http://linux.bkbits.net:8080/linux-2.6/[EMAIL PROTECTED]#!/usr/bin/ruby
class RcsFile
attr_reader :path
def initialize(path)
@path = path
end
def checkout(rev)
system("co -q -ko -r#{rev} [EMAIL PROTECTED]")
end
def checkin(rev, log)
if rev == "1.1"
open("|rcs -q -U -i -kb [EMAIL PROTECTED]", "w") { |p| }
system("ci -q -f1.1 #{path}")
open("|-") do |p|
exec("rcs", "-q", "-m1.1:#{log}", @path) if !p
print p.read
end
else
open("|ci -q -f#{rev} [EMAIL PROTECTED]", "w") do |p|
p.write log
end
end
end
end
class RcsSrc < RcsFile
def initialize(path)
super(path)
end
def rev_list
open("[EMAIL PROTECTED],m") do |f|
@rev_list = f.readlines
@rev_list.collect! do |l|
l.split
end
end
end
def checkin(rev, *par)
open(@path, "w") do |f|
f.write "data #{rev}\n"
end
super(rev, log(rev))
open("[EMAIL PROTECTED],m", "a") do |f|
f.write "#{rev} #{par.join(" ")}\n"
end
end
def log(rev)
"checkin #{rev}"
end
end
class RcsConv < RcsFile
attr_reader :src, :rev_map, :merge_map
def initialize(path, src)
super(path)
@src = src
@rev_map = {}
@merge_map = {}
if test(?f, "#{path},m")
open("#{path},m") do |f|
f.each do |l|
a = l.split
@rev_map[a[0]] = a[1]
@merge_map[a[1]] = a[2] if a[2]
end
end
end
end
def finish
open("#{path},m", "w") do |f|
@rev_map.each do |k,v|
f.print("#{k} #{v} [EMAIL PROTECTED]")
end
end
end
def checkout(rev)
src.checkout(rev)
File.rename(@src.path, @path)
end
def checkin(par, key)
if !par
rev = "1.1"
else
rev = par.sub(/\d*$/) { |v| v.succ }
if @rev_map.has_value?(rev)
i = "0"
begin
rev = "#{par}.#{i.succ!}.1"
end until [EMAIL PROTECTED](rev)
end
end
super(rev, src.log(key))
print "#{key} -> #{rev}\n"
@rev_map[key] = rev
end
def import(key, *par)
return true if @rev_map.has_key?(key)
if !par[0]
checkout(key)
checkin(nil, key)
return true
end
return false if [EMAIL PROTECTED](par[0])
if !par[1]
checkout(key)
checkin(@rev_map[par[0]], key)
return true
end
return false if [EMAIL PROTECTED](par[1])
checkout(key)
par1 = @rev_map[par[0]]
par2 = @rev_map[par[1]]
par1, par2 = par2, par1 if par1.count(".") > par2.count(".")
@merge_map[checkin(par1, key)] = par2
return true
end
end
system("rm -f test,? test2,?")
src = RcsSrc.new("test")
src.checkin "1.1"
src.checkin "1.2", "1.1"
src.checkin "1.3", "1.2"
src.checkin "1.3.1.1", "1.3"
src.checkin "1.3.1.2", "1.3.1.1"
src.checkin "1.3.2.1", "1.3"
src.checkin "1.3.2.2", "1.3.2.1"
src.checkin "1.4", "1.3"
src.checkin "1.5", "1.4", "1.3.1.2"
src.checkin "1.5.1.1", "1.5"
src.checkin "1.5.1.2", "1.5.1.1"
src.checkin "1.5.2.1", "1.5"
src.checkin "1.5.2.2", "1.5.2.1"
src.checkin "1.6", "1.5", "1.3.2.2"
src.checkin "1.7", "1.6"
src.checkin "1.8", "1.7", "1.5.1.2"
src.checkin "1.8.1.1", "1.8", "1.5.2.2"
src.checkin "1.8.1.2", "1.8.1.1"
src.checkin "1.9", "1.8"
src.checkin "1.10", "1.9", "1.8.1.2"
rcs = RcsConv.new("test2", src)
a = src.rev_list.dup
# just for fun we do it in random order
while !a.empty?
i = rand(a.size)
a.delete_at(i) if rcs.import(*a[i])
end
rcs.finish
src = RcsSrc.new("test")
src.checkin "1.8.2.1", "1.8"
src.checkin "1.8.2.2", "1.8.2.1"
src.checkin "1.11", "1.10"
src.checkin "1.12", "1.11", "1.8.2.2"
src.checkin "1.13", "1.12"
src.checkin "1.8.2.1.1.1", "1.8.2.1"
src.checkin "1.8.2.1.1.2", "1.8.2.1.1.1"
src.checkin "1.11.1.1", "1.11", "1.8.2.1.1.2"
src.checkin "1.14", "1.13", "1.11.1.1"
rcs = RcsConv.new("test2", src)
a = src.rev_list.dup
while !a.empty?
i = rand(a.size)
a.delete_at(i) if rcs.import(*a[i])
end
rcs.finish
