On Wed, 24 Oct 2007, James G. Sack (jim) wrote:
Nevertheless, I am attracted to the idea of
only claim (eg in specs) that which you have a [unit] test for
I wonder if any significant project has ever come close to achieving this?
The code at: http://www.trading-shim.org/ and the doco in the
manual.pdf are completely underlaid by test driven
development, with the tests pulled in and out as a given area
is refactored.
After introductory and interstical material, the meat of the
doco itself is written from test covered behaviours, and
particularly the edge diagrams and so forth are pulled
mechanically from the headers to generate the .dot diagrams
with graphviz on the fly. Those are massaged into .eps's and
thence via TeX, into the manual.
One cannot leave the full test suite enabled at all times,
because it turns a 2 minute test run, looking at one section
of changed code, into a 20 minute one. One loses the
'context' of where one was working when the cycle is too long.
We've spend a couple of years full time on the project, and
have essentially completed what we think of as our 1.0 release
featureset. We consider it 'substantial'. GPL v 3
-- Russ Herrold
--
[email protected]
http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list
