incubator-ooo-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Joost Andrae <>
Subject Off-topic: Re: Please add postgresql-sdbc to Oracle source code grant to Apache Software Foundation
Date Mon, 05 Sep 2011 22:44:52 GMT
Dear TJ,

I doubt this backend can or needs to be released and I'm not sure if 
AOOo will benefit from having such a system.

What do you do to read a callstack?

Take the upper lines of it and read their function calls. Somewhere 
within the upper part of the callstack you will find the address where 
the program called a signal (easiest case SEGV) which caused the crash. 
Just before this signal you will mostly find the root cause (a function 
call that is not behaving well) of the crash. Often you need to find 
other reasons by manually reading the callstack in combination with the 
sourcecode. Often you need to consult another developer to get 
information about correlation to other code parts. If you know the 
memory addresses of this cryptic (and mostly stripped) callstack then 
you can resolve these addresses against code segments within your code. 
All this can be done manually and by using your brain and possible 
automation is not an illusion.

Comparing build-id (part of the data) matching callstacks in combination 
with knowing the relative memory addresses (sent with the crash report) 
against a matching debug lib (archived within the build system) makes it 
quite easy for an experienced C++/Assembler programmer to build a 
database that counts a cumulative percentage of the upper stack 
addresses of the callstack which helps to detect duplicates to detect 
those crashes with the highest frequency and to automatically generate 
bugreports to those developers responsible to the exact code snipplet. 
This can only be processed if all build systems, debug systems, 
database, etc. are centrally hosted within a local network and all 
developers are known and are responsible for their code parts to be able 
to submit bug reports to them. In my opinion AOOo wouldn't have any 
benefit to host the legacy OOo Hamburg build related data because it's 
tied to the quite complex build system and to those builds created 
within this build system. If a build system is de-centralized (what I 
suppose it to be for AOOo) then those who want to provide a crash report 
enabled build need to provide their individual backend infrastructure to 
process callstacks of their builds against (to be archived) debug libs 
of their builds.

Have you thought about the effort it takes to host such a system outside 
of a corporate environment ? It's at least *very* expensive...

Am 05.09.2011 15:35, schrieb TJ Frazier:
> Another SGA question: what about the back-room, never-released code to
> process crash dumps? Details at
> /tj/

Just some thoughts

Kind regards, Joost

View raw message