Things You Should Never Do, Part I (2000)
Netscape 6.0 is finally going into its first public beta. There never was a version 5.0. The last major release, version 4.0, was released almost three years ago. Three years is an awfully long tim…
Yet the world is full of code that was replaced with less work than it would take to fix a single bug on the broken original.
It's almost like if universal rules for software engineering doesn't exist... but no, that's crazy.
Replacing code is fine. Replacing an entire product is the mistake.
Yet the world is full of code that was replaced with less work than it would take to fix a single bug on the broken original.
Can you point out a concrete example? Because you're commenting on a discussion on an essay that documents several notorious examples that demonstrate the opposite point.
I’m not extremely familiar with it, but I think X11 qualifies. I think it was determined that HDR support would be basically impossible to implement.
Ah yes those projects exist, that just are perfect and.. Or wait was that a dream maybe. Anyway, I'm sure they exist. After all, there's no universial rules, society and science doesn't constantly evolve and improve by searching for and defining such rules since the dawn of time.. Or was that a dream maybe. Anyway, I am certain this is wrong because my code is perfect
It's mostly stopping the world to re-write everything from scratch that is a mistake.
Good practice is to refactor the codebase until it's modular (if it's not already sufficiently modular), and then proceed to replace modules gradually until you reach your desired state of re-writtenness.
This has the benefits of actually getting you up to speed with maintenance over time without having to halt new feature development, giving you the best of both worlds.
Boy do I ever disagree with this.
For big projects, with multiple people and man-years of work, sure. Don’t start from scratch. But in my humble opinion, those projects shouldn’t really exist. Instead they should be atomic, made up of small page-length units which individually can be scraped and rebuilt.
For small projects, rewriting is often superb. It allows us to reorganize a mess, apply new knowledge, add neat features and doodads, etc.
But in my humble opinion, those projects shouldn’t really exist.
What's the point of your opinion if not only do these projects exist but they are also pervasive?
You cannot wish things away and pretend reality is something different.
those projects shouldn’t really exist
You think web browsers should not exist? How do you write Google Chrome, and all of it's dependencies, in one page of code?
I think you're miss-understanding the article. Joel didn't say you should never rewrite an individual component in your code, he was saying you should never throw out an entire project (all of the components) and start from scratch.
He also wasn't talking about "multiple people and man-years of work". He was talking much larger projects. How many people have contributed Chrome? Not just direct contributions writing lines of code, but indirect contributions such as reporting bugs or writing documentation on how it works?
If Google were to start over, all of that would be thrown out. It just can't be done.
All you can really do is what Microsoft did with IE / Edge. Edge was a fork of Chromium which was a fork of WebKit which was a fork of KTHML which was a fork of the KDE HTML Widget. Which dates back to 1996. Internet Explorer started in 1995. Microsoft didn't start Edge from scratch, they basically shifted their team of developers over to another project that was the same age.
The smaller the project, the easier it is to do a full rewrite but realistically it's almost never a good idea unless your product is very young.
If Google were to start over, all of that would be thrown out. It just can’t be done.
To stress the importance of this very basic fact, people need to understand that even Google, a company with virtually limitless resources to rearchitect and rewrite any and all type of software project, made the call to avoid using major features offered by some programming languages, such as C++'s exceptions, because it could have unintended consequences on the company's legacy code base which they could not rewrite.
And here we are, reading fantastic claims over how complete rewrites are reasonable things while flipping compiler flags to harden legacy projects is unheard of.
They actually did somewhat start Edge from scratch originally. They made EdgeHTML as a rewrite of the IE 11 trident engine.
In the end they abandoned it and moved over to chromium. One of the reasons being Google intentionally breaking their sites for EdgeHTML.
You think web browsers should not exist?
Yes. Multiple historical layers, each giving the creators way too much power over presentation, while they are still supposed to handle stuff like accessibility themselves, making a company webpage a thousand manhours project. Browsers being monoliths is only one page of the book.
How do you write Google Chrome, and all of it's dependencies, in one page of code?
So we agree that browsers are way too complex?
This is a weird take given that the majority of projects relevant to this article are massive projects with hundreds or thousands of developers working on them, over time periods that can measure in decades.
Pretending those don't exist and imagining fantasy scenarios where all large projects are made up of small modular pieces (while conveniently making no mention to all of the new problems this raises in practice).
Replace functions replace files and rewrite modules, that's expected and healthy for any project. This article is referring to the tendency for programmers to believe that an entire project should be scrapped and rewritten from scratch. Which seems to have nothing to do with your comment...?
I've found that big projects need refactoring without replacement, as Joel says. Bind and sudo and even sendmail have value in this world (the latter for versatility) that cannot be easily replicated without massive cost and decades of maintenance amid real attacks.
But yeah, trivial stuff can be rewritten and exhaustively tested at occasionally a cheaper cost where it can provide value to the enterprise that will one day out-balance its rebuild cost.
But in my humble opinion, those projects shouldn’t really exist.
In my humble opinion, everybody should just write perfect software first try.
In my humble opinion, you shouldn't be allowed to participate in big projects if you don't fulfill some minimal requirements.
Reminds me of the crack i had to explain the concept of technical debt. He was the main maintainer of a years old Xamarin project, he took over from his precursor.
For small projects, rewriting is often superb. It allows us to reorganize a mess, apply new knowledge, add neat features and doodads, etc.
This. I'm coding to contribute to a open-source software with very small amount of coders, and with a non-mainstream Domain-Specific Language. A lot of the code I did before has been proven to work from times to time, but they all could benefit from better outputs and better GUI. So, I end up reengineering the entire and that'll take a really long time, however, I do a lot of tests to ensure it works.
Netscape
You may know it as Mozilla, the namesake project that Mozilla (the company) felt was too hard a raison-d'etre to maintain and, yet, still exists. If you ever hear of Mozilla the org refered to as MINO, it's because they couldn't even continue their you-have-one-purpose task.
Unleash the downvotes, ye dissonants.
I only left the comment because of my pfp being the old Netscape logo haha
Except if your initial code was first written for something slightly related to the current usecase and with a tenth of the current feature set.
Then you better start from scratch and take some bits and pieces from your old code.
never put salt in your eyes
Where was this post when Elon decided to rewrite Twitter lmao
Shipping JavaScript in one week
laughs/cries in embedded we are currently changing from mbed to zephyr rtos. at the beginning it was decided we will never replace mbed, the thing i am currently replacing, so all the code is using parts of mbed everywhere. there's not enough space on the chip to have both mbed and zephyr in parallel. yay. you desktop/server/web/whatever people don't how good you have it.
Except when your rewriting it in rust, ofc 😎