Hacker News Reader

Comment by mtanski | Hacker News Reader

legacy numeric applications / routines?

Except for some obscure edge case, it wouldn't really be a great idea to run them on a JS vm... FORTRAN code still in use is computation-heavy stuff for use cases where speed makes a big difference, the performance hit of a VM wouldn't be acceptable.

Well, for example, we could run 1980-era climate model written in Fortran on desktop web browser; it would run with reasonable speed.

Ah modern technology: using 1000x faster hardware to run 30 year old code at the same speed, but in a browser!

That is what I think every time I see WebGL demos that most of the time remind me of Voodoo or Amiga demos, but hey it is on the browser!

You have accelerated graphics on any modern browser, on any OS, on any CPU architecture. I'll rag on JS as much as the next person but you have to admit 1) that's a pretty big feat 2) it's not an apples to apples comparison.

This is not true, because I have a couple of devices with pefectly fine working OpenGL ES 3.x GPUs where the browsers either don't support any form of WebGL or the FPS are a single digit number.

Really, the 2nd coming of VRML.

Thirty years ago it would have been run on supercomputers.

That's what I would define as a "some obscure edge case", see my comment above.

The thing is that pretty much all Fortran is "some obscure edge case". Otherwise it'd be written in something that's not Fortran.

Another example, I know the guy who maintains NIST's fluid properties solver (back in the day he worked at a subcontractor for NASA characterizing superfluids in fuel tanks). It's ancient Fortran compiled to a windows .dll and wrapped in VB in an excel file. It's super battle tested (being used for nearly 30 years), and it'd be awesome to be exposed in the browser.

I use quite frequently LAPACK (http://www.netlib.org/lapack/) which I would hardly define as "edge case".

Everyone who does any stats uses LAPACK, it's just that most of them aren't aware of it.

There is a lot of battle-tested Fortran out there. If you're for one reason or another in a situation where you have to deploy to the web platform and you need some numerical algorithms, I can imagine it could make sense to pick some routines from netlib instead of reinventing the wheel.

Why not keep the numerical code on the sever?

Hmm, I guess it depends on the application whether that's a good idea or not.

Or rather, before flang + emscripten/wasm/whatever, there was no reasonable way to run Fortran in a browser; now there is. Application developers can then choose whatever is most important for their application.

Disclaimer: my web dev experience predates the "modern web" with SPA's, fast javascript runtimes, html5 etc.

Never thought I'd live to see the day people would be talking about compiling fortran to javascript to run in a web browser. Just mind blowing.

If you can run the processing on the client side, then you can scale up much, much easier and provide whatever service much cheaper, or even free. Eg, you can provision fewer servers, no need to build a processing queue or load balancing setup, if the input is provided by the client then you use less bandwidth by avoiding a round trip on that, etc.

Not sure it makes sense for computationally expensive code to be running in the user's browser.

Cluster-scale calculations? Probably not. If we could /easily/ port FORTRAN code to JavsScript then you could do this [1] kind of thing with a lot less manual re-implementing.

[1] http://physics.weber.edu/schroeder/fluids/

Using python would possibly make more sense here

python numeric code using numpy and the like ultimately depends on some fortran libs underneath

Numpy has no Fortran whatsoever

The numpy.linalg module depends on LAPACK, which is Fortran.

Core numpy depends on BLAS, for which both Fortran and non-Fortran implementations are available.

Go to the source, it's called lapacklite and it's C code translated from Fortran

https://github.com/numpy/numpy/tree/master/numpy/linalg

Hmm, seems you're right. Though that seems to be some fallback thing which is used if the real thing isn't found during the build. At least on Ubuntu 16.04 the lapack_lite module links against the real (Fortran) lapack library:

  % ldd /usr/lib/python3/dist-packages/numpy/linalg/lapack_lite.cpython-35m-x86_64-linux-gnu.so|grep -E 'lapack|fortran'
	liblapack.so.3 => /usr/lib/liblapack.so.3 (0x00007fb619ce7000)
	libgfortran.so.3 => /usr/lib/x86_64-linux-gnu/libgfortran.so.3 (0x00007f5ae0911000)

The style of netlib is isolated standalone routines, easy to recompile to js. Numpy would be far more complicated.

It would be useful if you specifically wanted to run it in JS environment, rather than for the speed

Seems like most numeric/math code have some Fortran libraries underneath.

It'd be handy to have a route to just compile them to JS rather than having to write from scratch

This was a serious question? Oh geez. What kind of world has the web wrought?

...and then your Javascript code will run on node.js, and the instance will be Dockerized for ease of deployment to a VPS. How many layers of abstractions are there?

It's turtles all the way down.

And they all live in a VM anyway.

and also nginx reverse proxy m8, with apache compiling instances to be deployed automatically to scale using latest gcc7 with 03 optimization for robustness against DDOS

Yes as some one who used to hack Fortran for a living I was having a hard time to find a use case for converting Fortran to js

Reply on news.ycombinator.com