I absolutely hate Nim for spoiling pretty much all other programming languages for me. I wish it gained more traction. Congratulations on your book and keep up the great work, Araq. You did an outstanding job with Nim and I wish you all the best!
I am getting ahead of myself since I am still learning the language but I am coming from Python and various other languages so it has not been a complex transition.
I had added nim std lib to the Techempower benchmarks, and while Nim shines in the plain text and other non db related requests, it comes at the very bottom as far as tests like Fortune go. When I did the work, there were no maintained/active Postgres drivers with async and pipelining support. And IMO, that wa the biggest reason why Nim hasn't had great benchmarking scores.
Actual work on DB drivers is out-of-scope for me though. I noticed there is an async Postgres driver available: https://github.com/treeform/pg. I don't know how well tested this driver is, and no mention of pipelining support.
It would be great to see Nim's web frameworks higher up on the Techempower benchmarks. Lack of such DB driver features usually comes down to time. The support for those features in stdlib would be a good bounty issue.
I think the lowest hanging fruit for db drivers would be to just wrap libpq, the default PG driver. Given Nim's c interop, that should be easy. I was planning to do that, but then life happened and I had to skip that project.
I would say, that if you are trying to create a very involved framework, you should look to solve this problem. Otherwise the current perf of db_postgres can be a big deterrent to a lot of potential users.
https://github.com/nim-lang/Nim/issues/13559
I had much better luck with tiny_sqlite: https://github.com/GULPF/tiny_sqlite
I'll email you regarding what/how to help.
I’d be curious to get your Nim highlights.
haskell's types are cool
vuetify makes me consider nuxt
but only elixir has otp, immutable data structures, intercluseter message passing, channels, good performance, ease of use and descent ecosystem out of the box.
rust is the only other new language Im looking at but for things outside web development
Beam really is a marvel.
copy one write and immutable data structures mean you ca have garbage collection reserved for the termination of a thread whihc would be a nontrivial task in languages that encourage mutability cough go cough
Beam processes are a blessing
its also extremely tuned for creating threads that are super lightweight. consequently, scaling strategies that are optimal in elixir would be a TERRIBLE idea in other languages. You could make a dynamic cluster of node/go instances that communicate over rabbitmq and have a supervisor process that can keep track of their state and resurrection in case of a crash. Its going to be error prone and take weeks of fiddling to get right. beam gives you that out of the box. Its not perfect but its more than good enough for your first 10000 customers.
This benefit extends all the way up to Phoenix. because threads are trivial to create and the overhead is low, its a viable strategy to generate a stateful process for every user as they interact with the system, this is what allows liveview to be so powerful.
Each channel connection gets its own thread as well. A single machine can handle thousands of threads this way with full isolation. one thread crashing won't affect your other connections. nodejs cannot provide that. if a web-socket connection crashes a node instance, it ends the connection for every customer currently connected to that node.
Every attempt I've seen outside beam to recreate this revolve around OS level process which have a higher overhead and then they still need to create a mesasge passing protcol and supervision system. since OS threads are epensive, it wouldn't be a. good idea to try creating something like liveview as you'll overload your machine.
So you might say, "ok, I'll craete a multicluster virtual machine optimized for immutable data structures with a built in intra process messaging protocol for communication as well as a higher level super vision process. at that point, I'll say "congratulations, you just reinvented the beam, lets see if we can port other languages to it!"
anyways, if you're looking for something a little smaller for now, I highly recommend really understanding ecto and how it has you interact with databases. its easily the best database library I've worked with. precisely because it isn't a orm. It maps sql rows to elixir records which has much less of an impedance mismatch. its achieves almost zero friction between your database and your api.
Pros: Extremely good C and C++ interop Significant indentation Very good performance Quick compilation Terse, but readable Easy cross platform Powerful macros and close to trivial DSL construction Very little code to do a lot of stuff Incoming future features look very promising
Cons: Hard to predict and control program performance, LTO does a lot of heavy lifting Immature ecosystem, because only few users. Many libraries have bugs, missing features or just a very clunky API. Or maybe they are deprecated or abandoned all together. Many different ways to do the same thing everyThingIsCamelCase, ambiguous variable names (subjective) Documentation and tutorials is mostly reading through the code and googling forum posts outside the core language You compile either for C or C++, some libraries simply won't work without C++ compilation
For my contribution to the ecosystem see Nexus mentioned in another comment. If you see bugs please report them. Sometimes libraries are deprecated/abandoned because of new ideas, this happens in every ecosystem.
I personally use camelCase/PascalCase, but Nim is style insensitive, meaning you can call procs written in camelCase using snake case. However this is controversial and under discussion, my personal opinion is to make style sensitivity optional, with insensitivity the default.
Ambiguous variable names I hate, you won't see that in Nexus, if anyone does they should file an issue. That's really developer dependent.
For my part Nexus has some docs and a basic tutorial. But if people want more they need to ask for it. Why write docs nobody wants or will read? Anyway, this book (Mastering Nim) is one more definite step forward for learning Nim.
I think compiling to C is what mostly happens, C++ is probably an outlier. Even then, a C++ compiler isn't difficult to install if needed.
I just compiled my toy project (detecting anomalous parity in integers) in nim 1.6.6 (4.23s), go 1.17.7 (2.43s to create both aarch64 and x86_64 binaries), zig 0.8.0-dev.1140 (2.04s to create aarch64 and x86_64 binaries) and C via clang 13.1.6 (0.14s).
nim's compilation is 175% of Go, 207% of zig, and 3021% of clang.
$ nim -v
Nim Compiler Version 1.6.6 [MacOSX: arm64]
Compiled at 2022-05-05
Copyright (c) 2006-2021 by Andreas Rumpf
active boot switches: -d:release -d:nimUseLinenoise nim c -d:release --passC:"-flto" --gc:markAndSweep --out:ap ap.nim
Nim's compiler is definitely not speedy, and this is why so much effort has been spent on incremental compilation, which tmk, still isn't working - https://github.com/nim-lang/Nim/issues/19757.
$ touch foo.nim
$ /usr/bin/time nim c foo.nim
CC: ../usr/lib/nim/lib/std/private/digitsutils.nim
CC: ../usr/lib/nim/lib/system/dollars.nim
CC: ../usr/lib/nim/lib/system.nim
CC: foo.nim
Hint: gc: refc; opt: none (DEBUG BUILD, `-d:release` generates faster code)
36891 lines; 0.154s; 39.434MiB peakmem; proj: /tmp/foo.nim; out: /tmp/foo [SuccessX]
0.24user 0.02system 0:00.27elapsed 101%CPU (0avgtext+0avgdata 44360maxresident)k
0inputs+0outputs (0major+13676minor)pagefaults 0swaps
Unless your standards are "single digit milliseconds per file like tcc itself, please!", 240 ms is not so bad. (EDIT: And this is just on some 2016-era Intel i7-6700k box, not some liquid helium cooled 10 GHz whatever.)
For me missing libraries never stopped me, I could always port/wrap whatever I needed. However with the advent of AI assisted coding, working on libraries is even faster and at some point I hope I wouldn't spend much time rolling decent libraries that are ports of existing github projects.
- Python-style syntax with significant indentation
- Uniform Function Call Syntax: a.len() == a.len == len(a)
- Fully unqualified imports by default: which might seem scary to Python programmers, but works great in practice because of static typing
- All of the above makes code readable and succinct
And from a language features side:
- Compiles to C with all the architectural targets that come with it
- Compilation to C also allows for easy C interop: wrap function signatures and types and you're done
- This, in turn, means that Nim libraries can bootstrap off of the massive C ecosystem, while adding nicer APIs on top
- Extremely performant GC by default: optional Rust-style annotations can further improve performance, and you can remove all overhead and manually manage memory with C-style pointers if you'd like
- Useful compile-time templates and macros that can directly change the AST
The community is also active and helpful on IRC/Matrix/Discord/Gitter (bridged together)
a.len() == len(a) seems like a good idea, but I'm not so sure about a.len... Isn't that a bit ambiguous?
So you'll see alternative (or new) implementations of core language features like async, threading, DSLs, typeclasses, traits, etc available as external packages, with just about the same user experience as if they were built into the stdlib or compiler.
So it's just hard to justify using Rust, for me. Unless you're writing a DB or something where a GC is a no go. But Nim's memory management seems pretty good, and many things are stack allocated.
1. https://ramanlabs.in/static/blog/raw_memory_management_patte...