Those toolkits are only useful for edge cases. SDK performance is adequate for 95% of real world applications.
> From matlab to gnuplot, oscilloscope and spectrum analyser software, data analysis applications are frequently developed using C++.
I.e tools that don't need to be re-implemented, some of which had they been built today could easily have been written with a bytecode / JVM / CLR language.
Another tool I'm not re-implementing. You conveniently left out that all of their models are developed with Python - which is what 99.99% of developers interested in ML are trying to do.
My problem with your line of argument in this whole subthread is that no matter how many examples people give you of important categories where C++ predominates, you've already decided that the sum total of all that is "niche".
I'm interesting in your background: which languages do you use, and what kind of programs do you write? Perhaps it'll become clearer whether C++ is any use to you or not.
From the discussion I've read in this thread so far, it seems like you're claiming "what's the point of the entire field of mechanical engineering" when you, personally, just want to drive a car (not build one), fly on a plane (not build one) and so on.
> I'm interesting in your background: which languages do you use
Scala, Java, C#, Obj-C / Swift, Python, JS.
> what kind of programs do you write?
At work : line of business, database / search / indexing systems, mobile apps, web apps
At home : mobile apps, indie games
> it seems like you're claiming "what's the point of the entire field of mechanical engineering" when you, personally, just want to drive a car (not build one), fly on a plane (not build one) and so on.
You've misunderstood. What I'm claiming is that there are very few tasks at hobby-project scale that C++ is ideally suited for. C++ is a language that's used at team-scale, and is no longer an approachable language for the individual.
In other words, with Java you can build Android apps, with C# you can build Unity games, with Python / JavaScript / Whatever you can build web apps.
What can the individual do with C++ they couldn't do easier faster or better with some other language? Maybe embedded systems - but C seems just as effective for that domain. Likewise for kernel development for obvious reasons.
I actually want a reason to use C++, but "well you can work on Photoshop or high performance computing or HFT plumbing software" is not a helpful answer - both for me and anyone else hoping to pick up Cpp in their spare time.
Well if you're looking for anecdotal evidence: my last hobby project (OCR for a specific book+script), I used Mathematica to prototype (fast iteration) and C++ with OpenCV to implement it (way faster execution).
I started the project in Python but switched because it wasn't particularly good at anything vs. the combination of Mathematica and C++. Actually I prototyped it in C++ too before doing some extra analysis in Mathematica only because it had some algorithms that OpenCV didn't.
Those toolkits are only useful for edge cases. SDK performance is adequate for 95% of real world applications.
> From matlab to gnuplot, oscilloscope and spectrum analyser software, data analysis applications are frequently developed using C++.
I.e tools that don't need to be re-implemented, some of which had they been built today could easily have been written with a bytecode / JVM / CLR language.
> On many repositories I can rather clearly see a broad pink bar, e.g. https://github.com/search?utf8=%E2%9C%93&q=topic%3Amachine-l
In your own link C++ ends up in a distant 6th place. What is that supposed to prove? That it's more popular than JavaScript in ML development?
> https://github.com/tensorflow/tensorflow
Another tool I'm not re-implementing. You conveniently left out that all of their models are developed with Python - which is what 99.99% of developers interested in ML are trying to do.