From: Tom Newton (tom.newton@convergys.com)
Date: Mon Mar 25 2002 - 12:53:13 EST
Being a lurker, I don't have much right to comment, but here is my tuppence
worth:
> -----Original Message-----
> From: Jesper Skov [mailto:jskov@zoftcorp.dk]
> Sent: 25 March 2002 16:47
> Cc: abiword-dev@nyorp.abisource.com
> Subject: A new start? [Re: Alan's excellent idea]
[snip]
> Still here? :) OK. I must admit that I've had doubts about
> my continued
> use of time on AbiWord. Not the project itself or the gang - which I
> really love - but I'm feeling we're not getting anywhere
> (FAOD I'm fully
> aware of my personal contribution the last months which is
> very close to
> zero, AWN aside). Bugzilla keeps fighting back every attempt
> at reducing
> the Bug counts, and doing QA, I see the same old bugs reappear after a
> while.
>
> Maybe it's just because writing a WP is tricky. Maybe it's because the
> backend is stretched beyond its limits.
Probably both. But if there is one thing that can save you, it's regression
tests. I firmly believe these are one of the major (and sometimes
overlooked) things that contributes to successful, stable software.
Regression tests can be a PITA to maintain, but if correctly organised, they
can significantly reduce the kind of repeat bugs you describe. The idea is
that you encode all your test cases into a script of some description and
record the output somewhere (and put it into CVS) after verifying that the
output is correct. The next time you make a code change, you run the
regression tests and they compare the output from last time to the output
with your modified code. You then use a diff tool to verify that any
differences between the "good" file and your result are ones that you would
expect from your changes. And because the existing output is known to be
"good", you only have to check the differences.
It is obviously harder to create such a system with GUI based applications
(and particularly with XP ones) where the interactions are very difficult to
record in a file in a meaningful and repeatable way, but it must be
doable(!). Even if you only did this for the various file format conversion
parts of the code (which are all file based and should be deterministic) I
suspect it would help dramatically.
The tests can be run automatically as part of or immediately after a build,
and part of the release procedure should be to make sure that 100%
regression tests pass.
cheers
Tom
(still talking and not doing)
--NOTICE: The information contained in this electronic mail transmission is intended by Convergys Corporation for the use of the named individual or entity to which it is directed and may contain information that is privileged or otherwise confidential. If you have received this electronic mail transmission in error, please delete it from your system without copying or forwarding it, and notify the sender of the error by reply email or by telephone (collect), so that the sender's address records can be corrected.
This archive was generated by hypermail 2.1.4 : Mon Mar 25 2002 - 12:54:36 EST