On Thursday 04 February 2010 15:51:03 CeDeROM wrote:
> On Thu, Feb 4, 2010 at 8:14 PM, Mike Frysinger wrote:
> > i'm more than familiar with shared library policies, but for something
> > that has grown organically like urjtag, i dont think it's reasonable to
> > try and get it "perfect" the first time.  if we screw up, we bump the
> > SONAME and then we dont have to keep baggage around.  i honestly dont
> > think urjtag is an important enough piece of software atm that bumping
> > the SONAME is going to cause enormous grief.
> 
> Well, to be honest this kind of attitude "we can bump it" made me
> leave Linux in favour of FreeBSD, where almost everything in API and
> config looks the same for years and some new functionalities are
> added/expanded but not replacing the old ones each time they show up

are you sure ?  first of all, FreeBSD user<->kernel is the exact opposite of 
Linux -- they have evertyhing intertwined so that they can break ABI as they 
like.  g'luck taking a FreeBSD-4 binary and running it with FreeBSD-7.  
however, any Linux userspace binary should continue to work because the 
syscall ABI has been set in stone.

plus, the default FreeBSD library policy is terrible.  just look at the soname 
conventions used in libtool.  that's why for the Gentoo/FreeBSD port, we 
actually change the convention to use Linux SONAME policies.  we get sane 
upgrade paths.

as for the "we can bump it" attitude, that really isnt specific to any Linux 
distribution.  that is coming from the people packaging the software directly 
and they're OS independent (just like urjtag).

> UrJTAG is pretty mature as for now I think and we can try to
> figure out some generic API that won't have to change with the
> underlying stuff.

yes, UrJTAG as command line software is pretty mature, but that doesnt have 
any bearing on the library API/ABI.  we havent made a single release with the 
library interface in it, so any claims about its maturity is, frankly, 
premature.

> I do agree that UrJTAG may not the broadway star
> across the world software (for me that is okay), so some screwups can
> happen (as always), but they should not be our assumption - the
> assumption should be interoperability, generic interface,
> expandability and backward compatibility... I am not talking about
> total reorganization of the current API, but also not to changing it
> every bump, possibly to avoid the bumps at all by expanding library
> with some helper functions - maybe by working on these basic
> functions...

i'm not talking about bumping it every time or taking the approach loosely.  i 
am talking about the first release or two being "developer previews" without 
any stable guarantee.

we also dont have any symbol versioning support in here, so until that's done, 
i dont think talking about a stable ABI is appropriate.

> I dream about UrJTAG library to be a set of functions loadable by a
> nice Python script with GUI that can always work the same way, even
> with the next 10th or 50th release of urjtag library.

sure, but not the 1st or 2nd :P

> I think this is a good time to start such discussion and try to
> implement it in one of the next releases - there are many people
> already that I know awaiting for the UrJTAG as a library :-)

here's a concrete statement for you to cut your teeth on: any stable ABI 
discussion is completely moot as long as internal storage is directly 
accessible and the layout of structures are not hidden completely.  the only 
thing that may be exported are functions.
-mike

Attachment: signature.asc
Description: This is a digitally signed message part.

------------------------------------------------------------------------------
The Planet: dedicated and managed hosting, cloud storage, colocation
Stay online with enterprise data centers and the best network in the business
Choose flexible plans and management services without long-term contracts
Personal 24x7 support from experience hosting pros just a phone call away.
http://p.sf.net/sfu/theplanet-com
_______________________________________________
UrJTAG-development mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/urjtag-development

Reply via email to