User: Password:
|
|
Subscribe / Log in / New account

KS2010: ABI status for tracepoints

KS2010: ABI status for tracepoints

Posted Nov 2, 2010 12:24 UTC (Tue) by Cyberax (✭ supporter ✭, #52523)
Parent article: KS2010: ABI status for tracepoints

Why not make tracepoints 'semi-stable'?

For example, guarantee that they won't change for at least 2 years after kernel is released. That way, it'll still be possible to remove old tracepoints once they are not necessary.

Besides, tracepoint clients should by design follow kernel development. It's not like the situation with ioctls where we have ancient programs that we still need to support.


(Log in to post comments)

KS2010: ABI status for tracepoints

Posted Nov 2, 2010 13:38 UTC (Tue) by ccurtis (guest, #49713) [Link]

It's not like the situation with ioctls where we have ancient programs that we still need to support.

... today. Once tracepoints become 'stable', they'll be integrated into things like advanced IDEs and profilers. Netbeans uses DTrace on Solaris, for example.

KS2010: ABI status for tracepoints

Posted Nov 2, 2010 13:46 UTC (Tue) by Cyberax (✭ supporter ✭, #52523) [Link]

So? IDEs are updated fairly frequently (and usually automatically). And anyway, if a tracepoint becomes obsolete its usefulness can actually be negative. They are tied to kernel internals by their design.

With ioctls it's different. Old user-space programs might not be tied to kernel at all.

KS2010: ABI status for tracepoints

Posted Nov 2, 2010 18:50 UTC (Tue) by NAR (subscriber, #1313) [Link]

Maybe tracepoints should be stable for a distribution. E.g. Netbeans for RHEL 6 or SuSE 11 could support a different but still stable set of tracepoints.

KS2010: ABI status for tracepoints

Posted Nov 2, 2010 19:07 UTC (Tue) by Cyberax (✭ supporter ✭, #52523) [Link]

That would put a burden on both maintainers of NetBeans and RHEL/SuSE since their release cycles are not likely to be the same.

It's better to just guarantee that for N stable releases tracepoints will remain stable.

KS2010: ABI status for tracepoints

Posted Nov 3, 2010 8:35 UTC (Wed) by NAR (subscriber, #1313) [Link]

Maybe it is time to synchronize their release cycles? It is possibly wise to use the same longer supported kernel (like 2.6.27 or 2.6.32) in longer supported distributions anyway, so the tracepoints would be the same. At this point they just have to pick the same NetBeans version...

KS2010: ABI status for tracepoints

Posted Nov 3, 2010 11:50 UTC (Wed) by zdzichu (subscriber, #17118) [Link]

Exactly. PowerTOP on Solaris was implemented using its tracepoints implementation (DTrace). It was 2.5 years ago.

KS2010: ABI status for tracepoints

Posted Nov 4, 2010 16:14 UTC (Thu) by glikely (subscriber, #39601) [Link]

Linus made it absolutely clear that once something is deemed an ABI, then it is stable. Period. The pain is just too far-reaching on users with old userland to allow ABI churn (and there are many), even assuming a 2 year guarantee. That is why it is so important to be careful about how new ABIs are defined.


Copyright © 2017, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds