Add blame support to waterfall.sh

Add blame support to waterfall.sh

BUG=51584
TEST=none

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=55711

[jochen, 2010-08-11 10:31:54]

please review.

Here's a demo output:


./waterfall.sh blame "Linux Tests (valgrind)(3)" ui 152F078A
Searching for latest revision without this hash........
Hash #152F078A# occurs in build #6327 for the first time.
Receiving blame list.
Revision: 55620
Blamelist:
cmasone@google.com
dhollowa@chromium.org
evan@chromium.org
jianli@chromium.org
mirandac@chromium.org 
nshkrob@chromium.org
satish@chromium.org
sky@chromium.org
stuartmorgan@chromium.org

[Timur Iskhodzhanov, 2010-08-11 10:52:12]

Thank you very much for working on this feature!

See the inline comments

(Looking at the size of the new waterfall.sh I think we should try re-writing it
on python :-) )

http://codereview.chromium.org/3132006/diff/1/2
File tools/valgrind/waterfall.sh (right):

http://codereview.chromium.org/3132006/diff/1/2#newcode100
tools/valgrind/waterfall.sh:100: # $1 - Name of the builder.
What I'm slightly worried about is a rare situation when a report "moves"
between different shards.
Also, a report with the same hash may appear on a few builders.

Is it possible to search through the whole waterfall (keeping the "test name"
restriction) or is it way too long for 'blame'?

http://codereview.chromium.org/3132006/diff/1/2#newcode106
tools/valgrind/waterfall.sh:106: if [ "x$1" = "x" -o "x$2" = "x" -o "x$3" = "x"
]
I think using 'x'es for comparison bash is required only if you don't use
quotes.
if [ "$1" == "" -o ... ]
should be enough.

[jochen, 2010-08-11 11:19:18]

http://codereview.chromium.org/3132006/diff/1/2
File tools/valgrind/waterfall.sh (right):

http://codereview.chromium.org/3132006/diff/1/2#newcode100
tools/valgrind/waterfall.sh:100: # $1 - Name of the builder.
On 2010/08/11 10:52:12, Timur Iskhodzhanov wrote:
> What I'm slightly worried about is a rare situation when a report "moves"
> between different shards.
> Also, a report with the same hash may appear on a few builders.
> 
> Is it possible to search through the whole waterfall (keeping the "test name"
> restriction) or is it way too long for 'blame'?

Sure, that would be possible. However, the load on the waterfall will be n-times
larger then.

http://codereview.chromium.org/3132006/diff/1/2#newcode106
tools/valgrind/waterfall.sh:106: if [ "x$1" = "x" -o "x$2" = "x" -o "x$3" = "x"
]
On 2010/08/11 10:52:12, Timur Iskhodzhanov wrote:
> I think using 'x'es for comparison bash is required only if you don't use
> quotes.
> if [ "$1" == "" -o ... ]
> should be enough.

Done.

[Timur Iskhodzhanov, 2010-08-11 11:53:09]

LGTM
searching the whole waterfall is definitely a lower priority feature 

On 2010/08/11 11:19:18, jochen wrote:
> http://codereview.chromium.org/3132006/diff/1/2
> File tools/valgrind/waterfall.sh (right):
> 
> http://codereview.chromium.org/3132006/diff/1/2#newcode100
> tools/valgrind/waterfall.sh:100: # $1 - Name of the builder.
> On 2010/08/11 10:52:12, Timur Iskhodzhanov wrote:
> > What I'm slightly worried about is a rare situation when a report "moves"
> > between different shards.
> > Also, a report with the same hash may appear on a few builders.
> > 
> > Is it possible to search through the whole waterfall (keeping the "test
name"
> > restriction) or is it way too long for 'blame'?
> 
> Sure, that would be possible. However, the load on the waterfall will be
n-times
> larger then.
> 
> http://codereview.chromium.org/3132006/diff/1/2#newcode106
> tools/valgrind/waterfall.sh:106: if [ "x$1" = "x" -o "x$2" = "x" -o "x$3" =
"x"
> ]
> On 2010/08/11 10:52:12, Timur Iskhodzhanov wrote:
> > I think using 'x'es for comparison bash is required only if you don't use
> > quotes.
> > if [ "$1" == "" -o ... ]
> > should be enough.
> 
> Done.

[M-A Ruel, 2010-08-11 13:14:40]

http://codereview.chromium.org/3132006/diff/5001/6001
File tools/valgrind/waterfall.sh (right):

http://codereview.chromium.org/3132006/diff/5001/6001#newcode15
tools/valgrind/waterfall.sh:15:
WATERFALL_PAGE="http://build.chromium.org/buildbot/memory/builders"
Why is this script bash? I think it's starting to become time to convert it in
python.

Using simplejson.loads( urllib.urlopen(
'http://build.chromium.org/buildbot/memory/json/builders' ) ) looks simpler to
me.