Re:Refactoring vs. Rewrite
by richardw on Wednesday November 18, 2009 @11:54AM
(#948575)
Joel wrote an article back in 2000 on why you should never rewrite, mentioning Netscape. While I'm far less sure than he is about it (see below), there are some awesome points in there.
http://www.joelonsoftware.com/articles/fog0000000069.html
While rewriting I've often been surprised how much I've had to re-discover and re-think, and how many mistakes I'll make second time around. On the other hand, on one occasion a rewrite saved the day because it pulled together 5 different apps in different languages in a new up-do-date platform. A few months of work and the sum was infinitely better than the whole and saved the business much more than it cost.
Re:Refactoring vs. Rewrite
by btilly on Wednesday November 18, 2009 @01:56PM
(#948844)
On the flip side during an extended major refactor your code looks like a construction site. You have some stuff that has been rewritten, some that hasn't, and no consistency. Then during the refactor people will rethink how they want it written, and you get more code styles.
Frequently the dust and mess gets so discouraging that people start being driven away. And then the rewriting effort peters out and people just get used to living in a half-demolished house.
The moral? Keep refactors small. If you want to do a major redesign, I would suggest rewriting a small prototype of a piece of it. If the prototype goes well then expand it until it can replace the previous system. (I've seen this prototype approach succeed several times. I haven't seen such good results with major refactors, or the "Let's rewrite from scratch!" approach.)
Re:Refactoring vs. Rewrite
by firefalcon on Wednesday November 18, 2009 @02:04PM
(#948866)
I also don't think that full rewrite from scratch is a viable strategy. Much better is to rewrite by pieces and integrate them with existing system via interfaces/adapters/etc. For example, you may start new Data Layer and move code from old app to it slowly, but constantly. Sometime you will have completely new DL.
Re:Refactoring vs. Rewrite
by btilly on Wednesday November 18, 2009 @04:50PM
(#949210)
Believe it or not I have seen multiple people succeed with a full rewrite from scratch. The ones I have been close to did it with a prototype they evolved.
Re:Refactoring vs. Rewrite
by jlouis on Wednesday November 18, 2009 @01:10PM
(#948710)
Interestingly however, much of the Linux kernel is based on the idea of pulling something off and then substituting it completely. For what it is worth, it seems to work for them.
Re:Refactoring vs. Rewrite
by btilly on Wednesday November 18, 2009 @01:58PM
(#948847)
IE 4 was competing with NS 4, which also was largely rewritten. That leveled the playing field quite a bit. (Particularly since the NS rewrite introduced a lot of problems.)
Re:Refactoring vs. Rewrite
by jcapote on Wednesday November 18, 2009 @01:34PM
(#948780)
I wish they had rewritten netscape, maybe firefox wouldn't be total garbage now. This advice, while great for some things, needs to be taken with a grain of salt. Sometimes, a full rewrite is the only option.
With all the maintainers doing this at the code, database and design level you shouldn't need to do a rewrite.
The problem is with 'immovable things' that necessarily accrete layer upon layer of work-arounds. If maintenance work allows 'immovable things' to persist in the application, at whatever level, the whole structure gradually degrades and a costly rewrite is eventually forced.
Re:Refactoring vs. Rewrite
by barrkel on Wednesday November 18, 2009 @11:40AM
(#948548)
It's not nearly so simple in practice. If your codebase is old and large, and no one person knows it enough to change significant amounts of it, there are simply some fixes that you never have the time to risk.
Re:Refactoring vs. Rewrite
by by on Wednesday November 18, 2009 @12:25PM
(#948638)
Yes I agree if your 'immovable thing', for example the problem that would need a risky fix that nobody fully understands, can remain without much impact for the rest of the application's life.
Sometimes I find I can move towards fixing something difficult in multiple steps as I apply successive features etc. I find that even just documenting the problem in the affected parts of the code can be a good first step. If something changes once it often changes again and I'll move it forward a bit more next time. My theory is that the movement is the important thing, not the arrival at some particular point.
Re:Refactoring vs. Rewrite
by igrekel on Wednesday November 18, 2009 @11:54AM
(#948572)
That is very short, the subject has been explored much more deeply in the past.
I'd suggest Brian Foote's Big ball of mud where he talks about software going through phases of expansion and consolidation, he also discusses the total rewrite in much greater lengths. Although it too is lacks hard data and I find parts of the format and discourse annoying (can't exactly put my finger on why), I think it is still more interesting.
http://www.laputan.org/mud/mud.html
Re:Refactoring vs. Rewrite
by zyb09 on Wednesday November 18, 2009 @07:09PM
(#949445)
Now I'm the middle of rewriting a good-sized software system. The reason for this is, we had to a add a completely new security level on top of our data (user accounts, groups, individual column access).
There's just no way we could've refactored our already tucked-together legacy code. This thing was passed along 4 different developers, each constantly adding new features and I'm surprised it hasn't exploded yet.We're getting calls every day cause something broke again.
So yea sometimes rewrite definitively makes sense and I'm sure our customers will love it.
While rewriting I've often been surprised how much I've had to re-discover and re-think, and how many mistakes I'll make second time around. On the other hand, on one occasion a rewrite saved the day because it pulled together 5 different apps in different languages in a new up-do-date platform. A few months of work and the sum was infinitely better than the whole and saved the business much more than it cost.
Frequently the dust and mess gets so discouraging that people start being driven away. And then the rewriting effort peters out and people just get used to living in a half-demolished house.
The moral? Keep refactors small. If you want to do a major redesign, I would suggest rewriting a small prototype of a piece of it. If the prototype goes well then expand it until it can replace the previous system. (I've seen this prototype approach succeed several times. I haven't seen such good results with major refactors, or the "Let's rewrite from scratch!" approach.)
http://www.moserware.com/2008/10/boy-scout-check-ins.html
With all the maintainers doing this at the code, database and design level you shouldn't need to do a rewrite.
The problem is with 'immovable things' that necessarily accrete layer upon layer of work-arounds. If maintenance work allows 'immovable things' to persist in the application, at whatever level, the whole structure gradually degrades and a costly rewrite is eventually forced.
Sometimes I find I can move towards fixing something difficult in multiple steps as I apply successive features etc. I find that even just documenting the problem in the affected parts of the code can be a good first step. If something changes once it often changes again and I'll move it forward a bit more next time. My theory is that the movement is the important thing, not the arrival at some particular point.
I'd suggest Brian Foote's Big ball of mud where he talks about software going through phases of expansion and consolidation, he also discusses the total rewrite in much greater lengths. Although it too is lacks hard data and I find parts of the format and discourse annoying (can't exactly put my finger on why), I think it is still more interesting. http://www.laputan.org/mud/mud.html
This would have been a very interesting article if it was backed by actual data. It's not. Shame.
There's just no way we could've refactored our already tucked-together legacy code. This thing was passed along 4 different developers, each constantly adding new features and I'm surprised it hasn't exploded yet.We're getting calls every day cause something broke again.
So yea sometimes rewrite definitively makes sense and I'm sure our customers will love it.