Re: Shipping Vera with 2.4

[Sander Vesik <sander_traveling yahoo co uk>]

Jody Goldberg wrote:
> On Wed, Mar 05, 2003 at 02:11:13PM +0000, Sander Vesik wrote:
>
> > Havoc Pennington wrote:
> >
> > > If you had shared IPC, then you could have just specified a message to
> > > add a new file, and a broadcast when the list changed, and that's it.
> >
> > Saying 'and to add entries into the recently used files list you need to 
> > link to this stack of libraries and use these APIs' is in fact 
> > coinsiderably more complex code wise than specifying file locking. And also 
> > raises teh question of how much impact will various bugs in the layer you 
> > are linking to exposing you to and how you test it and so on.
>
>
> I've got to disagree with you there Sander.  In my experience it is
> far far better to have a single implementation that can be
> fixed as new subtleties arise, than a myriad of one off versions.
> There are two key advantages to a shared implementation :
>     1) fixing 1 bug fixes all programs
>     2) having several programs using the same code accelerates the
>        finding of bugs.
>
> Look at something as deceptively simple as opening a stdio FILE for
> output and saving things.  Most apps did not handle things like
> setgid directories, or symlinks smoothly.  Gedit added some support
> for them but it did not work well on Solaris.  I raided the code and
> with lots of assistance polished it up and stuck it in libgsf.  Now
> all the apps using gsf get to benefit.
>
> Having a librecentfiles that exposed a trivial interface and
> handled file locking and contention smoothly seems like a much
> better choice than waving our hands at the issue by writing a
> detailed specification that boils down to
>     'do the right thing (please)'

If the librecentfiles - or similar - is a simple self contained library 
that does one thing, does it well, does not contain non-trivial amounts of 
copy-paste code and can be used with dlopen with no major hurdles from one 
place, then I obviously agree. This changes largely for linking to a 
multipurpose stack of libraries to get (essentialy) one function.

I'm not exactly a fan of everybody linking to a static expat library, but 
its a sort of workable solution - the same would not apply to an evolving 
infrastructure.

> Not that a spec is a bad thing, but a solid implementation trumps it.

You normaly need a spec for a solid implementation, otherwise, how do you 
tell its is doing what it is supposed to?

__________________________________________________
Do You Yahoo!?
Everything you'll ever need on one web page
from News and Sport to Email and Music Charts
http://uk.my.yahoo.com