| 1 |
What has been done this week: |
| 2 |
|
| 3 |
- added the stage of building in a file access event log record |
| 4 |
- done smart handling of logs. Delete duplicate records if that records |
| 5 |
belongs to one stage |
| 6 |
- implemented a fast <local file>-><name of installed package> mapping |
| 7 |
- an util for user which show him packages having at least |
| 8 |
one file accessed has been developed. |
| 9 |
Example of its report is here: http://dl.dropbox.com/u/7603377/report_cowsay.txt |
| 10 |
Report of manual launching ./configure for bash: |
| 11 |
http://dl.dropbox.com/u/7603377/report_configure.txt |
| 12 |
Report of manual launching ./make for bash: |
| 13 |
http://dl.dropbox.com/u/7603377/report_make.txt |
| 14 |
|
| 15 |
-changed a [C part]<->[python part] protocol. There is a stage in it |
| 16 |
instead of pid. In past pid was in protocol and I paused the program |
| 17 |
while python part is analysing pid. Now all pid analyse is in C part; |
| 18 |
Early all fields in protocol message was escaped like filename\ with\ |
| 19 |
spaces. I profiled the python part. Unescaping was one of bottleneck. |
| 20 |
Now it is NULL byte between fields and two NULL bytes between records. |
| 21 |
|
| 22 |
What I plan do do this week: |
| 23 |
- finish ld_preload approach(apply small changes in protocol) |
| 24 |
- add simple --block <pkg> key in autodep util. It will block an |
| 25 |
access to all <pkg> files |
| 26 |
- make a report more human-readable |
| 27 |
- make a building stage determining, and a blocking logic for |
| 28 |
non-child of initial process faster. |
| 29 |
|
| 30 |
Best, |
| 31 |
|
| 32 |
Alexander Bersenev |
Archives