[maemo-community] Periodic cleanup of Extras-Devel?

From: Dave Neary dneary at maemo.org
Date: Fri Jan 15 17:36:24 EET 2010
Hi,

Simon Pickering wrote:
> Interesting. I think that if a package uses some library and it works,
> the whole lot should be pushed to testing. E.g. I've got a bunch of
> mathematics libraries sitting in extras-devel for a lack of any
> applications using them. When I upload Octave that will use them and
> then they can all get promoted to extras-testing, but if someone else
> compiles e.g. R, which will also use the same libraries, they should be
> able to promote the whole lot if it all appears to work IMO.

Let's say I write MyCalculatorApp, which uses libbuggymathlibrary
version 1.0 (libbml.so.1.0 for short)  for adding, subtracting,
multiplying and dividing numbers.

Now let's say that libbml.so.1.0 is also used by ScientificCal, which
also calls a number of other functions: geometric functions, square
roots, logs, etc.

In a new version of libbml, libbml.so.1.1, the author moves to ane
xperimental Taylor's series calculation for trigs, logs and Nth roots;
The results are off by 20% because he's not actually any good with
maths. But anyway...

Now I upload MyCalculatorApp, which works perfectly, and gets promoted
to testing, and drags the new version of libbml with it.

Unless ScientificCal has specified a specific version of the library to
take, won't it also pick up the new libbml, and start having bad results
for its calculations? I'm no expert, of course, but if the libbml.so
author didn't change the API or ABI, he wouldn't be required by
convention to bump the major number, and could thus introduce bugs which
affect part of the functionality of the library not used by the app
getting promoted to testing.

Is this (contrived) example completely detached from reality?

Cheers,
Dave.

-- 
maemo.org docsmaster
Email: dneary at maemo.org
Jabber: bolsh at jabber.org


More information about the maemo-community mailing list