gitstat-devel

[Gitstat-devel] Fwd: Current gitstat broken?

[Soon-Son Kwon(Shawn) <ks...@kl...>]

Oops... we should check it.

Paul: (Not Paul Mundt, but Paul Vriens huhu)
could you check this? We don't have to feel lonely.
At least there shold be several people watching
this project. If we have more bug, then more people
will show more interest. (just kidding) :-)

Paul Mindt: Thank you for letting us know.
I hope you could join the devel list and
drop us some note from time to time...


2007/11/12, Paul Mundt <pau...@re...>:
> My current statistics on gitstat seem to be completely miscalculated, and
> have been broken eve since the -rc1 release. Could you see if there is a
> bug here?
>
> git shortlog gets the numbers right, at least:
>
> $ git shortlog v2.6.23..v2.6.24-rc2  | grep 'Paul Mundt'
> Paul Mundt (100):
> ...
>
> while gitstat for some reason only reports 17?
>


-- 
http://kldp.org/~kss

Re: [Gitstat-devel] Current gitstat broken?

[<jun...@gm...>]

Hi,
In my opinion, gitstat has some bug,

Now, when generate some chart with current gitstat,
1. find start and end date of Tags.
2. and count changest and author in this period.
3. Make chart.

But we could know It's not good solution.
see below page.

http://tree.celinuxforum.org/gitstat/commit-detail.php?commit=1100b47ea557e1cd6d9b71fe7ba1a189ed8bf276

This changeset is included in tag(v2.6.24-rc1), but it was commited at
2007/06/28 20:15 !!

  	v2.6.24-rc2    2007/11/06 21:57:46
	v2.6.24-rc1    2007/10/24 04:50:57
	v2.6.23       	 2007/10/09 21:31:38

So, we should make other logic for making chart.

And the chart has wrong data rely on server-setting, too.
Now CELF server is setted on PST.
so, we should set server-time or make some modification on gitstat.

Thanks,all.

On Nov 12, 2007 10:04 AM, Soon-Son Kwon(Shawn) <ks...@kl...> wrote:
> Oops... we should check it.
>
> Paul: (Not Paul Mundt, but Paul Vriens huhu)
> could you check this? We don't have to feel lonely.
> At least there shold be several people watching
> this project. If we have more bug, then more people
> will show more interest. (just kidding) :-)
>
> Paul Mindt: Thank you for letting us know.
> I hope you could join the devel list and
> drop us some note from time to time...
>
>
> 2007/11/12, Paul Mundt <pau...@re...>:
> > My current statistics on gitstat seem to be completely miscalculated, and
> > have been broken eve since the -rc1 release. Could you see if there is a
> > bug here?
> >
> > git shortlog gets the numbers right, at least:
> >
> > $ git shortlog v2.6.23..v2.6.24-rc2  | grep 'Paul Mundt'
> > Paul Mundt (100):
> > ...
> >
> > while gitstat for some reason only reports 17?
> >
>
>
> --
> http://kldp.org/~kss
>

Re: [Gitstat-devel] Current gitstat broken?

[Paul M. <le...@li...>]

On Mon, Nov 12, 2007 at 03:19:05PM +0900, ?????? wrote:
> In my opinion, gitstat has some bug,
> 
Yes, it's totally broken. Here's a stupid comparison script I hacked up,
maybe it's useful for someone:

#!/bin/bash
#
# gitstat-cmp.sh
#
# Sample gitstat input:
#   1 Adrian Bunk                45
#   2 Ralf Baechle               42
#   3 Alan Cox                   25
#   4 David S. Miller            24
#   5 Jeff Garzik                24
#   6 Al Viro                    19
#   7 Pavel Emelyanov            19
#   8 Tejun Heo                  18
#   9 Paul Mundt                 17
#
# Sample gitstat-cmp.sh output:
#
# Bogus entry for "Adrian Bunk" (gitstat 45 kernel 137)
# Bogus entry for "Ralf Baechle" (gitstat 42 kernel 143)
# Bogus entry for "Alan Cox" (gitstat 25 kernel 70)
# ...
#

# Example usage: sh ./gitstat-cmp.sh gitstat.txt . v2.6.23 v2.6.24-rc2
[ "$#" -ne "4" ] && \
	echo "Usage: $0 <gitstat list> <kernel tree> <tag1> <tag2>" && exit 1

IFS="
"
pushd $2 > /dev/null || exit 1

for line in $(sed -e 's/^[ \t]*[0-9]* //' < $1); do
	name=$(echo $line | sed -e 's/  .*$//;s/\t.*//')
	count=$(echo $line | sed -e "s/$name//g;s/^[ \t]*//;s/[ \t]*$//")

	for srcline in $(git shortlog $3..$4 | grep "^$name"); do
		srccount=$(echo $srcline | sed -e 's/.*(\(.*\)):.*$/\1/g')
		if [ "$count" -ne "$srccount" ]; then
			echo "Bogus entry for \"$name\" (gitstat $count kernel $srccount)"
		fi
	done
done

popd > /dev/null

Re: [Gitstat-devel] Current gitstat broken?

[Paul V. <pau...@gm...>]

??? wrote:
> Hi,
> In my opinion, gitstat has some bug,
> 
> Now, when generate some chart with current gitstat,
> 1. find start and end date of Tags.
> 2. and count changest and author in this period.
> 3. Make chart.
> 
> But we could know It's not good solution.
> see below page.
> 
> http://tree.celinuxforum.org/gitstat/commit-detail.php?commit=1100b47ea557e1cd6d9b71fe7ba1a189ed8bf276
> 
> This changeset is included in tag(v2.6.24-rc1), but it was commited at
> 2007/06/28 20:15 !!
> 
>   	v2.6.24-rc2    2007/11/06 21:57:46
> 	v2.6.24-rc1    2007/10/24 04:50:57
> 	v2.6.23       	 2007/10/09 21:31:38
> 
> So, we should make other logic for making chart.
> 
> And the chart has wrong data rely on server-setting, too.
> Now CELF server is setted on PST.
> so, we should set server-time or make some modification on gitstat.
> 
> Thanks,all.
> 
> On Nov 12, 2007 10:04 AM, Soon-Son Kwon(Shawn) <ks...@kl...> wrote:
>> Oops... we should check it.
>>
>> Paul: (Not Paul Mundt, but Paul Vriens huhu)
>> could you check this? We don't have to feel lonely.
>> At least there shold be several people watching
>> this project. If we have more bug, then more people
>> will show more interest. (just kidding) :-)
>>
>> Paul Mindt: Thank you for letting us know.
>> I hope you could join the devel list and
>> drop us some note from time to time...
>>
>>
>> 2007/11/12, Paul Mundt <pau...@re...>:
>>> My current statistics on gitstat seem to be completely miscalculated, and
>>> have been broken eve since the -rc1 release. Could you see if there is a
>>> bug here?
>>>
>>> git shortlog gets the numbers right, at least:
>>>
>>> $ git shortlog v2.6.23..v2.6.24-rc2  | grep 'Paul Mundt'
>>> Paul Mundt (100):
>>> ...
>>>
>>> while gitstat for some reason only reports 17?
>>>
>>
>> --
>> http://kldp.org/~kss
>>
> 
Hi,

Yes stupid thinko(s).

Using the date for the tags was a bad idea in this case. We picked the date from 
the tags and used that as limits. What basically happened was:

get date of last tag : Dec-1-2007
get date of the tag before that : Nov-1-2007

and then we calculate the commits between those dates. That's obviously wrong as 
the date marks the 'end of a release'.

I will sent a patch that corrects that.

Another thing that's important in this particular case is that we now also can 
show release candidates (which we do). This means that getting the "Top 
contributors (Kernel Release) for 2.6.24-rc2" will by default be compared 
against 2.6.24-rc1 and not 2.6.23. The gitstat administrator can decide which 
tags can be used.

So:

a) I'll sent a patch
b) I'll change http://tree.celinuxforum.org so that 2.6.24-rc1 is left out of 
the equation.

Cheers,

Paul.

P.S. I guess we have this issue in other charts as well so I will have a look 
(later).

Re: [Gitstat-devel] Current gitstat broken?

[<jun...@gm...>]

I received and applied that patch.
Plz, Check about it.
Thanks,all.

On Nov 12, 2007 4:59 PM, Paul Vriens <pau...@gm...> wrote:
>
> ??? wrote:
> > Hi,
> > In my opinion, gitstat has some bug,
> >
> > Now, when generate some chart with current gitstat,
> > 1. find start and end date of Tags.
> > 2. and count changest and author in this period.
> > 3. Make chart.
> >
> > But we could know It's not good solution.
> > see below page.
> >
> > http://tree.celinuxforum.org/gitstat/commit-detail.php?commit=1100b47ea557e1cd6d9b71fe7ba1a189ed8bf276
> >
> > This changeset is included in tag(v2.6.24-rc1), but it was commited at
> > 2007/06/28 20:15 !!
> >
> >       v2.6.24-rc2    2007/11/06 21:57:46
> >       v2.6.24-rc1    2007/10/24 04:50:57
> >       v2.6.23          2007/10/09 21:31:38
> >
> > So, we should make other logic for making chart.
> >
> > And the chart has wrong data rely on server-setting, too.
> > Now CELF server is setted on PST.
> > so, we should set server-time or make some modification on gitstat.
> >
> > Thanks,all.
> >
> > On Nov 12, 2007 10:04 AM, Soon-Son Kwon(Shawn) <ks...@kl...> wrote:
> >> Oops... we should check it.
> >>
> >> Paul: (Not Paul Mundt, but Paul Vriens huhu)
> >> could you check this? We don't have to feel lonely.
> >> At least there shold be several people watching
> >> this project. If we have more bug, then more people
> >> will show more interest. (just kidding) :-)
> >>
> >> Paul Mindt: Thank you for letting us know.
> >> I hope you could join the devel list and
> >> drop us some note from time to time...
> >>
> >>
> >> 2007/11/12, Paul Mundt <pau...@re...>:
> >>> My current statistics on gitstat seem to be completely miscalculated, and
> >>> have been broken eve since the -rc1 release. Could you see if there is a
> >>> bug here?
> >>>
> >>> git shortlog gets the numbers right, at least:
> >>>
> >>> $ git shortlog v2.6.23..v2.6.24-rc2  | grep 'Paul Mundt'
> >>> Paul Mundt (100):
> >>> ...
> >>>
> >>> while gitstat for some reason only reports 17?
> >>>
> >>
> >> --
> >> http://kldp.org/~kss
> >>
> >
> Hi,
>
> Yes stupid thinko(s).
>
> Using the date for the tags was a bad idea in this case. We picked the date from
> the tags and used that as limits. What basically happened was:
>
> get date of last tag : Dec-1-2007
> get date of the tag before that : Nov-1-2007
>
> and then we calculate the commits between those dates. That's obviously wrong as
> the date marks the 'end of a release'.
>
> I will sent a patch that corrects that.
>
> Another thing that's important in this particular case is that we now also can
> show release candidates (which we do). This means that getting the "Top
> contributors (Kernel Release) for 2.6.24-rc2" will by default be compared
> against 2.6.24-rc1 and not 2.6.23. The gitstat administrator can decide which
> tags can be used.
>
> So:
>
> a) I'll sent a patch
> b) I'll change http://tree.celinuxforum.org so that 2.6.24-rc1 is left out of
> the equation.
>
> Cheers,
>
> Paul.
>
> P.S. I guess we have this issue in other charts as well so I will have a look
> (later).
>
>
>
>
>

Re: [Gitstat-devel] Current gitstat broken?

[Paul V. <pau...@gm...>]

??? wrote:
> I received and applied that patch.
> Plz, Check about it.
> Thanks,all.
> 

Numbers look better now. It's strange though that my system show a 100 commits 
for Paul Mundt (and that's what Paul expected) and yours shows 101.

Needs more investigation (when time permits).

Cheers.

Paul.

Re: [Gitstat-devel] Current gitstat broken?

[Paul V. <pau...@gm...>]

Paul Vriens wrote:
> Hi,
> 
> Yes stupid thinko(s).
> 
> Using the date for the tags was a bad idea in this case. We picked the 
> date from the tags and used that as limits. What basically happened was:
> 
> get date of last tag : Dec-1-2007
> get date of the tag before that : Nov-1-2007
> 
> and then we calculate the commits between those dates. That's obviously 
> wrong as the date marks the 'end of a release'.
> 
> 
Please forget the bogus explanation. The fix I committed is correct but for 
different reasons as stated above.

We know to what tag a commit belongs so it's fine to use that for the graph.

The reason that the numbers were not correct is purely the date of the commits 
(AFAIK). If I take for example commit e7bd34a15b85655f24d1b45edbe3bdfebf9d027e:

[paul@penguin linux-2.6.git]$ git show e7bd34a15b85655f24d1b45edbe3bdfebf9d027e
commit e7bd34a15b85655f24d1b45edbe3bdfebf9d027e
Author: Paul Mundt <le...@li...>
Date:   Tue Jul 31 17:07:28 2007 +0900

     sh: Support explicit L1 cache disabling.

     This reworks the cache mode configuration in Kconfig, and allows for
     explicit selection of write-back/write-through/off configurations.
     All of the cache flushing routines are optimized away for the off
     case.

     Signed-off-by: Paul Mundt <le...@li...>

This patch dates Jul-31 whereas 2.6.23 came out on Oct-9.

git-gui shows for this commit:

Follow : v2.6.23-rc7
Precedes : v2.6.24-rc1

Cheers,

Paul.

Re: [Gitstat-devel] Current gitstat broken?

[Paul M. <le...@li...>]

On Mon, Nov 12, 2007 at 10:27:05AM +0100, Paul Vriens wrote:
> We know to what tag a commit belongs so it's fine to use that for the graph.
> 
> The reason that the numbers were not correct is purely the date of the 
> commits (AFAIK). If I take for example commit 
> e7bd34a15b85655f24d1b45edbe3bdfebf9d027e:
> 
[snip]

> This patch dates Jul-31 whereas 2.6.23 came out on Oct-9.
> 
I don't know if it matters or not, but I do frequently rebase my trees,
which results in a different timestamp for commit and author. Perhaps
this is something that tripped up gitstat?

Re: [Gitstat-devel] Current gitstat broken?

[Paul V. <pau...@gm...>]

Paul Mundt wrote:
> On Mon, Nov 12, 2007 at 10:27:05AM +0100, Paul Vriens wrote:
>> We know to what tag a commit belongs so it's fine to use that for the graph.
>>
>> The reason that the numbers were not correct is purely the date of the 
>> commits (AFAIK). If I take for example commit 
>> e7bd34a15b85655f24d1b45edbe3bdfebf9d027e:
>>
> [snip]
> 
>> This patch dates Jul-31 whereas 2.6.23 came out on Oct-9.
>>
> I don't know if it matters or not, but I do frequently rebase my trees,
> which results in a different timestamp for commit and author. Perhaps
> this is something that tripped up gitstat?
> 
That was my thought as well. We have seen several cases now where using the date 
is a bad idea so we need to find some other means. As we already had the tag 
with each commit (when it's stored in the DB) this was an easy one.

The problem has been there all along btw. It's just that with the latest release 
we also included the possibility to show release candidates which made the 
number of commits between tags less so more obvious to spot.

Thanks for the report anyway and I hope you continue to use gitstat. We have 
lots more changes planned and if you have any ideas/wishes just shout (or create 
a request on the SourceForge site).

Cheers,

Paul.