Comment by systemvoltage | Hacker News Reader

Comment by systemvoltage | Hacker News Reader

Parent Full thread

This is a really broad brush.

Oracle DB will fail less often than a python open source utility with 10 stars on Github.

It really depends on the scale of the project and how many people are contributing to it, and its momentum going forward - commercial or open-source.

I've had MATLAB engineers fix a compiler bug in 2 hours while I was on the phone line with them. Sure, it costs my company $13k/year for the license but when you have a jet engine hoisted up for testing and its costing the company thousands of dollars of loss time, yep, commercial software or open-source software with solid support is a must.

Proprietary software sometimes is quite good if not the best in class. There is no need to pick up pitch forks and large brushes to pain everything red. If nothing else, it shows how misguided, immature and closed-minded some of the GNU folks are. Most of the open source contributors are nice though, and I don't wish to ironically generalize everyone.

I would use Oracle as a counterexample - while it's very stable, and there is a feature for everything (only some of them are in various extra $$ addon products), however, I was really surprised at the horrible quality and dev-unfriendliness of all the tooling and ecosystem compared to the equivalent open source tools; and the brittleness of their integration with any non-Oracle products, which is a necessity in many larger companies.

While there is a lot of quite good proprietary software, and Oracle is good in many aspects, I would not really call Oracle best in class; in so many situations PostgreSQL would be preferable even if you ignore Oracle licencing (and you can't really ignore Oracle licencing).

GNU Octave appears to have some commercial support vendors listed, that probably wouldn't mind receiving $13k/year in exchange for answering the phone: https://www.gnu.org/software/octave/commercial-support.html

That sounds great! Why can't both proprietary software and open source software co-exist? For startups and small companies, open source software is great. For large enterprises, the cost of the software is a vanishingly small % of the cost, and in fact they don't care as much - a lot of large enterprises are continuing to open source their own internal tools.

MATLAB has over 2 dozen toolboxes, with insanely detailed documentation. We are actually very happy to have MATLAB and our engineers are trained to use it for decades. We gladly pay for it. It would be a massive undertaking to go GNU Octave even if it did everything MATLAB did. I think we are getting stuck in the weeds, the point is:

We are engineers. Software is a tool to get the job done.

There are many other packages that cannot compete with anything open source - ANSYS Mechanical for example.

>Why can't both proprietary software and open source software co-exist?

Probably because it's impossible to have code that is both proprietary and open source at the same time. Unfortunately the code is either public or it isn't, because of copyright law there's not really an in-between.

Don't get me wrong, I'm not saying your company should use any particular product, but you do have options of where you can put that money that you can exercise when it's most appropriate. None of those consultants seem to be as big as Mathworks though, which probably matters.

>There are many other packages that cannot compete with anything open source - ANSYS Mechanical for example.

It only takes one entrepreneur to get the ball rolling by releasing a serious open source competitor. The whole industry can build on top of it after that.

Would it really be “a massive undertaking to go GNU Octave” when its syntax is a superset of MATLAB’s?

Either way, in the longer term I think saving a bit of money wouldn’t hurt, spending some sponsoring Free software could eventually catch it up to competitors, and having Free industry standard tools more easily accessible to future engineers to begin decades of training earlier could benefit enterprises.

> Would it really be “a massive undertaking to go GNU Octave” when its syntax is a superset of MATLAB’s?

Does anyone actually use MATLAB just for the "MATLAB" part of it? Where I work, it's the Simulink part that's locked us in... I'm still not sure how MATRIXx withered away and gave Mathworks free reign in the control systems design space...

I don’t mind paying for software I guess.

and the FOSSjobs wiki lists similar vendors for coreboot, Debian, LVFS and Samba:

https://github.com/fossjobs/fossjobs/wiki/resources#freelanc...

If you were willing to pay 13k/year as a support fee, I suspect there isn’t a single piece of FOSS you couldn’t get immediate support turnaround for.

Case in point, Red Hat built a $32B business off of mostly selling support contracts for FOSS. Notionally some amount of that was licensing deals, but the core reason to buy RHEL instead of using Debian or something was because RHEL comes with support.

You aren't wrong, but I do think it's worth pointing out that Red Hat writes a metric ton of FOSS code also which it contributes upstream (or the upstream is entirely RH owned/controlled).

For example, Red Hat is the number two contributor to Kubernetes (behind Google). Red Hat extends K8s with OpenShift which makes K8s a PaaS, and that is also open source and freely available as OKD[1]. RH doesn't get paid directly for that work but funds it with support contracts (which in turn enable landing more support contracts).

I know that doesn't contradict what you said, I just don't want to leave a reader with the impression that RH just tacks on a support contract and makes some money. RH does a huge amount of dev on it too.

[1]: https://www.okd.io/

So then what difference does it make? I am not going to fix the code myself anyways and visibility of the source code is irrelavant.

It is like buying a hammer to nail some nails, I don't care the metallurgy of the hammer as far as it gets the job done. If it fails, I buy a better one.

It's about the relative freedom of the market.

A hammer has well known metallurgical properties, so anyone can make one. You don't care whom you buy from, but you buy a cheaper, hopefully better made hammer because nobody tries to fleece you for a 100$ hammer.

In the world with only commercial knowledge of hammers, they cost 10G, people spend many years and invest much capital into going to hammer school, and maybe even you just rent a hammer at a 13G a year rate with support. Nevermind the cost for say, a screwdriver.

Speaking of MathWorks, I had the misfortune of having to use one of their less famous offerings (Polyspace) and it was borderline unusable with basic functionality failing out of the box. I believe our license was $50k/year and most of the bugs simply were simply left unfixed after we reported them.

This seems to be par for the course for commercial static analysis tools. The Clang family of tools, on the other hand, has served us quite well but unfortunately they don't perform some of the checks that we need to do for regulatory reasons.

Would you mind telling some of the kinds of checks the Clang tools do not perform? I wonder if they are mostly syntactic or if they need dataflow analyses. (That said, Polyspace has 2 tools now, Bug Finder and Code Prover, whose applications are somewhat different.)

Clang does not check MISRA rules or calculate other metrics that are considered important in embedded industries.

The vast majority of MISRA rules are either purely syntactic (governing things as trivial as variable names) or obvious to the point of uselessness (like "don't violate the One Definition Rule"). The few actually useful ones require dataflow analysis but frankly Polyspace doesn't appear to do a very good job at it. My impression was that it achieved a 0% false negative rate by way of a 99.99% false positive rate. However, we never reported any of these false positives as bugs as the reports likely would have been rejected outright. The bugs we reported were all far more clear, such as the tool refusing to compile perfectly valid code, or glitches in the user interface.

I'm honestly not privy to most of the politics and no longer work in this area, but my impression was that even if Clang were to implement these checks, it would not be taken seriously by auditors as it doesn't come with a price tag or tick other boxes like generating nice PDF reports with violation counts and so on.

> This is a really broad brush.

Unless the person you're replying to edited their original comment, they did the opposite of a broad brush.

The comment clearly specified that what was said was based on personal experience and reiterated that milage may vary.

---

It's difficult to gauge overall quality and satisfaction for any product, even more so when talking: software + open source V.S. proprietary + people with different backgrounds and expectations + possibly different versions.

In my opinion, this is especially true when a good part of people's experiences come from software being used in their jobs. Meaning, they don't have much control over it so they're subject to whatever has been given to them and that highly affects their judgment.

Probably depends on the field. Yeah, engineering software might be lacking, but for machine learning tools, open-source runs laps around other proprietary tools (including MATLAB, which would be considered a joke).

Reply on news.ycombinator.com