Tinderbox hacking

I was talking with sprout earlier and he was asking me about how I would envision getting performance information out of the build logs and into something that could be mined for trends or other data. Basically the idea is to get into the log the test name and it's value and prepend a marker so it can be pulled out when the tinderbox2 processmail_builds perl script is run (it's called for each email a build client sends to the tinderbox server.)

During this script run a number of items are scanned for and one is used by the Mozilla team to extract bloat-related data, and this is the pattern I mimic'd for this hack.

Windows installer

My first thoughts were to use the InnoSetup^? open source install tool but after researching the very latest version of NSIS I happened upon a feature it has that makes it very useful to OSAF: the script compiler can be run on Linux. This will allow the maximum flexibility on how/where/when the installers are built.

I'm working on a prototype now to compare against InnoSetup^? for look and feel issues.

Twisted install size

First pass at bug:2216 - what makes it tough is that I could not use the output of py2depgraph.py like I thought - I'll have to see why it didn't catch some of the import references.

Anywho, the current build image contains the following twisted code:

• 60K ./application
• 48K ./conch/client
• 16K ./conch/insults
• 12K ./conch/openssh_compat
• 56K ./conch/scripts
• 180K ./conch/ssh
• 20K ./conch/ui
• 412K ./conch
• 68K ./cred
• 64K ./enterprise
• 76K ./flow
• 64K ./internet/iocpreactor
• 24K ./internet/serialport
• 584K ./internet
• 16K ./lore/scripts
• 132K ./lore
• 16K ./mail/scripts
• 372K ./mail
• 76K ./manhole/ui
• 188K ./manhole
• 112K ./names
• 84K ./news
• 28K ./pair
• 20K ./persisted/journal
• 100K ./persisted
• 32K ./protocols/gps
• 8.0K ./protocols/mice
• 388K ./protocols
• 344K ./python
• 56K ./runner
• 104K ./scripts
• 64K ./spread/ui
• 296K ./spread
• 24K ./tap
• 60K ./trial
• 8.0K ./web/scripts
• 276K ./web/woven
• 640K ./web
• 152K ./words/im
• 32K ./words/protocols/jabber
• 288K ./words/protocols
• 8.0K ./words/scripts
• 564K ./words
• 76K ./xish

for a total of 4.7M and after some snooping/testing I seemed to have gotten it down to:

• 68K ./cred
• 64K ./internet/iocpreactor
• 560K ./internet
• 16K ./mail/scripts
• 372K ./mail
• 80K ./persisted
• 348K ./protocols
• 344K ./python

for a total of 1.7M

I think I may have to run a script to scan for *.pyo and remove any directories that don't have any after a successful chandler run - may be the safest.

-- MikeT - 23 Nov 2004