Re: libevent upgrade from 1.3e to 1.4.5-stable

["Samuel J. Greear" <dragonflybsd@xxxxxxxxxxxx>]

"Antonio Huete Jimenez" <tuxillo@quantumachine.net> wrote in message 
486bf4f3$0$850$415eb37d@crater_reader.dragonflybsd.org">news:486bf4f3$0$850$415eb37d@crater_reader.dragonflybsd.org...
> Hi
>
> Here's a working patch for upgrading libevent. It compiles just fine and 
> I'm currently using in one of my DFBSD machines.
>
> http://leaf.dragonflybsd.org/~tuxillo/patches/libevent145.diff
>
> Bug fixes can be found in:
>
> http://levent.svn.sourceforge.net/viewvc/levent/branches/patches-1.4/libevent/ChangeLog?revision=885&view=markup
>
> I don't have the oportunity of checking with bluetooth because I don't 
> have any device, so if someone has devices and time, please check.
>
> Joerg,
> About ABI changes, there are some functions that have been removed, and 
> some parameters slightly changed:

. ..

> Samuel Greear (sjg), I invite you to give your opinion here just like at 
> IRC ;-)
>
>
> --
> * Please, consider your environment before printing this email
> Cheers
> tuxillo at EFNet in #dragonflybsd


Considering the primary rationale for importing libevent (to my knowledge) 
is to
support libbluetooth, and given that it is readily available via pkgsrc and 
is
known to have an unstable ABI. Wouldn't it make the most sense (short of
rewriting libbluetooth not to use libevent) to either move libevent into a
subdirectory of libbluetooth and link it directly into libbluetooth, 
potentially
dropping the manpages. Or, expose it as dflibevent or similar noting 
prominently
that it is generally intended for base system use and third party binary
applications should not expect it to maintain a stable ABI? My preference 
would
be toward the former more than the latter, at least until some other 
consumer of
libevent hits base. Either way it would pretty much eliminate any ABI 
breakage
concerns which fly in the face of a preference to upgrade the library upon 
every
new release. I would think the ideal would be for third party applications 
to
prefer pkgsrc for something like this. ABI breakage is probably no big deal 
at
current but could get bumpy down the road (I'm not sure how this would 
factor
into process checkpointing in the case of restoring a libevent-dependant 
process
elsewhere, where libevent exposes a different binary interface). In any 
case,
maintenance and updates of select and poll support for libevent can likely 
be
pulled from the tree.

I can prepare a patch if there is some kind of consensus.

Sam