6 ms·

Perhaps I should update the question. I'm not referring to ALL software quality. I'm referring to the quality of codebases that are 1) old, 2) large, and 3) sup

Perhaps I should update the question. I'm not referring to ALL software quality. I'm referring to the quality of codebases that are 1) old, 2) large, and 3) supported by many people. You make a good point though. Perhaps I just miss the good ol' days of working on small teams with small codebases that were pretty easily maintained.

quanticle 5y ago
Even with those caveats, it's unclear whether the premise is satisfied. The Linux kernel is old (old enough to drink!) large (and getting larger!) and supported by many people. Has its quality declined? I don't think so. It supports more hardware than ever. Kernel panics don't happen nearly as often as they used to. New features (BPF), make kernel programming easier. It's difficult to say that the Linux kernel's quality has declined over time.
- timhrothgar 5y ago
  I had this specific example in mind. How has Linux done it? Was it ultimately due to the benevolent dictator for life (BDFL) management practice?
  - xxpor 5y ago
    The kernel isn't working on commercial deadlines. Maintainers don't have to worry about their stack rank at the end of the year. They don't have to justify their head count. Some of this happens internally I'm sure, at places like IBM, Red Hat, and Intel, but none of it is coming from Linus.
  - ssdspoimdsjvv 5y ago
    Probably that, and the average Linux contributor being a lot brighter than your run of the mill programmer.
    - 6510 5y ago
      If someone was to design a good small set of opcodes we could go back to rolling machine code by hand. While at it we might as well fill the atx box with cute tiny isolated computers that talk over ip addresses. They have to be small enough as to not leave space for an OS. Some ROM is fine ofc.
- peakaboo 5y ago
  People being up the Linux kernel on the same way they bring up Oprah as an example of being successful in America. 99.99999% of software in big corporations will not have even 5% of the quality of the Linux kernel. The reason the kernel has such a high quality is because of Linux being a dictator, training everybody in not always so nice ways to write code that doesn't break anything and that is maintainable. He cares about the code and he has the status and the mandate to prevent it from becoming shit.
  - PragmaticPulp 5y ago
    > 99.99999% of software in big corporations will not have even 5% of the quality of the Linux kernel. Having done a lot of work on the kernel, I think you’d be surprised at the relatively low quality of a lot of kernel drivers. The core code is generally quite good, but it’s not true that the kernel is full of pristine code. I’ve worked at multiple companies where code quality standards exceeded a lot of the weirder stuff I’ve seen (and often fixed) in kernel driver code.
  - quanticle 5y ago
    The Linux kernel is an easy example, sure, but it's not the only one. You can say the same thing about proprietary operating systems and software as well. It's been a long time since I've had misbehaving applications blue-screen Windows. That's something that used to happen daily when I ran Windows 98. Or if you prefer MacOS, I remember MacOS Lion being incredibly unstable, with beach-balls and crashes galore. That's all been fixed. Applications have gotten better too. Browsers have gotten significantly more robust against misbehaving pages. Microsoft Office doesn't eat my work (even if I forget to hit ctrl-s). In fact, I would go so far as to say that the only software which has gotten worse is video games. Used to be you could put in a disc, install the game, and be reasonably assured that you were getting a playable final product. Today, you put in the disc, install the game, and then have to download multiple gigabytes of patches... and the game is still often buggy (Bethesda, I'm looking at you!). >99.99999% of software in big corporations will not have even 5% of the quality of the Linux kernel. That is true. But it was equally true when Linus Torvalds dropped the first version of the Linux kernel all the way back in '91. It's not clear to me that things have gotten worse since then.
- sllabres 5y ago
  >> Kernel panics don't happen nearly as often as they used to My earliest kernel I have used in something one might call production are 1.2.12 around 1995. I must say even then, with this early kernels I had no panics at all and much higher uptimes (patching for security wasn't as much of an issue at that time ;-) )
  - not2b 5y ago
    There were many exploitable remote root holes then, but no one was attacking them because Windows was a much juicier target.
peakaboo 5y ago
You listed the reason it's a mess yourself - many people worked on it and it's old. For big companies there is no incentive at all for a developer to personally care about and do battle over things like code quality and reduce complexity. If they are use any kind of agile (lol) system, it will just be about polishing the turd so it doesn't break and add features in some way that doesn't require big rewrites.
wrs 5y ago
Old large codebases are mostly maintained by people who weren’t around when the code was originally coming into existence. They don’t know the implicit design assumptions and decisions, or even the history of requirements. One thing you’ll find in nearly any software project of any age is a lack of good documentation of those things, so as you lose the community folklore, people will start making myopic changes, cargo-culting, violating future-looking design principles, and so forth. Pretty soon you just have a pile of incoherent features and making systematic improvements is hard because the code is no longer systematic.
mbesto 5y ago
> 1) old, 2) large, and 3) supported by many people. Software entropy implies that software that doesn't change will always suffer entropy if the environment changes. In a vacuum, we would never need to change software once it was "feature complete". But that's not how the world works. Environments change and so that inevitably means software will corrode. Candidly, I think you're just looking at the world through rose tinted glasses.
com2kid 5y ago
> Perhaps I should update the question. I'm not referring to ALL software quality. I'm referring to the quality of codebases that are 1) old, 2) large, and 3) supported by many people. Because Google[1] and Facebook came along and scared everyone by iterating at a LOLWTF pace and companies in surrounding spaces looked at what was taking up time in their release schedules and the answer was "test passes" so they fired all their testers, and told devs to add unit tests but unit tests don't cut it. Companies that used to have immaculate software quality had dedicated test automation engineers who had the job of abusing software in crazy bizarre ways. Then they hired armies of manual testers to go over anything that hadn't been automated. Lots of problems existing with this system, one of which was career advancement for software engineers in test was limited because it is hard to get recognized for the two primary jobs of an SDET: Signing off on code Blocking a release on quality grounds So you had a gradual rot of SDET and test orgs at companies, with pools of brilliance that slowly got drained as the best engineers got tired of being undervalued. Start from that base, and then around ~2010 everything needs to start "moving fast". Apple and MS both get rid of their test teams, and with two of the largest employers of dedicated software engineers in test getting out of the field, the entire field itself falls apart. Now it is career suicide, an ever shrinking career path that pays far less than doing "real" development work. That leaves us at where we are today. Everything sucks and breaks all the time. [1] Everyone forgets how bad the first 5 major versions of Android were.
- mbrodersen 5y ago
  I have no QA testing and the code I maintain haven’t had a customer production bug for 5+ years. And we are taking complex software written in C++. The reason why is 90000 end-to-end use case tests. I routinely implement new features or refactor major parts of the application and the tests will tell me if it is ready for production or not.

quanticle 5y ago

Even with those caveats, it's unclear whether the premise is satisfied. The Linux kernel is old (old enough to drink!) large (and getting larger!) and supported by many people. Has its quality declined? I don't think so. It supports more hardware than ever. Kernel panics don't happen nearly as often as they used to. New features (BPF), make kernel programming easier. It's difficult to say that the Linux kernel's quality has declined over time.

timhrothgar 5y ago
I had this specific example in mind. How has Linux done it? Was it ultimately due to the benevolent dictator for life (BDFL) management practice?
- xxpor 5y ago
  The kernel isn't working on commercial deadlines. Maintainers don't have to worry about their stack rank at the end of the year. They don't have to justify their head count. Some of this happens internally I'm sure, at places like IBM, Red Hat, and Intel, but none of it is coming from Linus.
- ssdspoimdsjvv 5y ago
  Probably that, and the average Linux contributor being a lot brighter than your run of the mill programmer.
  - 6510 5y ago
    If someone was to design a good small set of opcodes we could go back to rolling machine code by hand. While at it we might as well fill the atx box with cute tiny isolated computers that talk over ip addresses. They have to be small enough as to not leave space for an OS. Some ROM is fine ofc.
peakaboo 5y ago
People being up the Linux kernel on the same way they bring up Oprah as an example of being successful in America. 99.99999% of software in big corporations will not have even 5% of the quality of the Linux kernel. The reason the kernel has such a high quality is because of Linux being a dictator, training everybody in not always so nice ways to write code that doesn't break anything and that is maintainable. He cares about the code and he has the status and the mandate to prevent it from becoming shit.
- PragmaticPulp 5y ago
  > 99.99999% of software in big corporations will not have even 5% of the quality of the Linux kernel. Having done a lot of work on the kernel, I think you’d be surprised at the relatively low quality of a lot of kernel drivers. The core code is generally quite good, but it’s not true that the kernel is full of pristine code. I’ve worked at multiple companies where code quality standards exceeded a lot of the weirder stuff I’ve seen (and often fixed) in kernel driver code.
- quanticle 5y ago
  The Linux kernel is an easy example, sure, but it's not the only one. You can say the same thing about proprietary operating systems and software as well. It's been a long time since I've had misbehaving applications blue-screen Windows. That's something that used to happen daily when I ran Windows 98. Or if you prefer MacOS, I remember MacOS Lion being incredibly unstable, with beach-balls and crashes galore. That's all been fixed. Applications have gotten better too. Browsers have gotten significantly more robust against misbehaving pages. Microsoft Office doesn't eat my work (even if I forget to hit ctrl-s). In fact, I would go so far as to say that the only software which has gotten worse is video games. Used to be you could put in a disc, install the game, and be reasonably assured that you were getting a playable final product. Today, you put in the disc, install the game, and then have to download multiple gigabytes of patches... and the game is still often buggy (Bethesda, I'm looking at you!). >99.99999% of software in big corporations will not have even 5% of the quality of the Linux kernel. That is true. But it was equally true when Linus Torvalds dropped the first version of the Linux kernel all the way back in '91. It's not clear to me that things have gotten worse since then.
sllabres 5y ago
>> Kernel panics don't happen nearly as often as they used to My earliest kernel I have used in something one might call production are 1.2.12 around 1995. I must say even then, with this early kernels I had no panics at all and much higher uptimes (patching for security wasn't as much of an issue at that time ;-) )
- not2b 5y ago
  There were many exploitable remote root holes then, but no one was attacking them because Windows was a much juicier target.

peakaboo 5y ago

You listed the reason it's a mess yourself - many people worked on it and it's old. For big companies there is no incentive at all for a developer to personally care about and do battle over things like code quality and reduce complexity. If they are use any kind of agile (lol) system, it will just be about polishing the turd so it doesn't break and add features in some way that doesn't require big rewrites.

wrs 5y ago

Old large codebases are mostly maintained by people who weren’t around when the code was originally coming into existence. They don’t know the implicit design assumptions and decisions, or even the history of requirements. One thing you’ll find in nearly any software project of any age is a lack of good documentation of those things, so as you lose the community folklore, people will start making myopic changes, cargo-culting, violating future-looking design principles, and so forth. Pretty soon you just have a pile of incoherent features and making systematic improvements is hard because the code is no longer systematic.

mbesto 5y ago

> 1) old, 2) large, and 3) supported by many people. Software entropy implies that software that doesn't change will always suffer entropy if the environment changes. In a vacuum, we would never need to change software once it was "feature complete". But that's not how the world works. Environments change and so that inevitably means software will corrode. Candidly, I think you're just looking at the world through rose tinted glasses.

com2kid 5y ago

> Perhaps I should update the question. I'm not referring to ALL software quality. I'm referring to the quality of codebases that are 1) old, 2) large, and 3) supported by many people. Because Google[1] and Facebook came along and scared everyone by iterating at a LOLWTF pace and companies in surrounding spaces looked at what was taking up time in their release schedules and the answer was "test passes" so they fired all their testers, and told devs to add unit tests but unit tests don't cut it. Companies that used to have immaculate software quality had dedicated test automation engineers who had the job of abusing software in crazy bizarre ways. Then they hired armies of manual testers to go over anything that hadn't been automated. Lots of problems existing with this system, one of which was career advancement for software engineers in test was limited because it is hard to get recognized for the two primary jobs of an SDET: Signing off on code Blocking a release on quality grounds So you had a gradual rot of SDET and test orgs at companies, with pools of brilliance that slowly got drained as the best engineers got tired of being undervalued. Start from that base, and then around ~2010 everything needs to start "moving fast". Apple and MS both get rid of their test teams, and with two of the largest employers of dedicated software engineers in test getting out of the field, the entire field itself falls apart. Now it is career suicide, an ever shrinking career path that pays far less than doing "real" development work. That leaves us at where we are today. Everything sucks and breaks all the time. [1] Everyone forgets how bad the first 5 major versions of Android were.

mbrodersen 5y ago
I have no QA testing and the code I maintain haven’t had a customer production bug for 5+ years. And we are taking complex software written in C++. The reason why is 90000 end-to-end use case tests. I routinely implement new features or refactor major parts of the application and the tests will tell me if it is ready for production or not.