Mirror of forum.nim-lang.org

3798 :: Few mixed up questions

• ErikCampobadal (orginal) [2018-05-03T01:23:59+02:00] view original

  Hello!

  Been using nim for a few days, been reading a lot and feel really happy right now!

  Few questions (I myself come from high level languages although I've used few c)

  1. What is best to use: ref object or object and why. If using ref object, is "new" always needed?
  2. What's the difference between method and proc. I can't find much information about it. I do not understand what dynamic dispatch means.
  3. What are templates and why / when are they used. They seem like procs to me with rare types "untyped".
  4. Although I've tried basic importc pragma tests. Is there anywhere where I can fully find how to use it? Importing headers seems not to work giving me a not found (I tried adding the path to the compiler and yet not working. Local "header.h", not <header.h>. I've been watching the nim math library to see how it works. It shoud be easy to understand if it was written correctly...

  Thanks a lot for your time!

---

• mashingan (orginal) [2018-05-03T08:15:14+02:00] view original

  I'll answer to best what I know

  • 1. object is preferable if you want value-type, while ref object is for reference-type (it's obvious from the name tho, :P ). I usually use reference-type when it's costly to create it such as IO, Files, memory etc. When using ref object, new is required although the compiler can infer it and you can use the syntax like object.
  • 3. View template as a block labeled snip of code. So when using template you put "the block of code" in where you invoke the template
  • 4. For local header, you put it "\"header.h\"" instead of "<header.h>"

---

• mratsim (orginal) [2018-05-03T10:06:53+02:00] view original

  My own spin:

  1. Object are allocated on the stack while ref objects are allocated on the heap by the garbage collector. let a = foo will copy the object in the first case but will copy the reference to the object in the second case, so a and foo are completely distinct for object but the same in the second case (i.e. a refrence to the same memory location stored in 2 variables)
  2. Nim is a statically typed language, meaning which proc to call must be known at compile-time. However in some cases (object-oriented) you might want to run on specialized type only known at runtime, for example you have something called an animal but you don't know if it will be a cat or a dog at compile-time. You want functions with a common name, but a different behaviour at runtime --> use inheritance + methods. If everything is known at compile-time, just use Generics.
  3. Template = copy-pasting a block of code, a proc actually exists in the final binary, a template only exist at Nim level but not in the C/Javascript or Assembly produced.
  4. Importc for header only is tricky: have a look in the following libraries that wrap header only C and C++:

  • Using a C intermediate file that include the header: https://github.com/define-private-public/stb_image-Nim
  • Using {.passC -I"fooPath".} to include the header path: https://github.com/status-im/nim-ttmath/blob/master/src/ttmath.nim#L5
  • Or the poor man solution, copy the headers in ./nimcache, Nim is actually looking there by default ;)

---

• DTxplorer (orginal) [2018-05-03T12:55:39+02:00] view original

  To use C code in Nim I compile the C files with -c option (on GCC), then I use {.link: "your_binary.o".} pragma on top of the Nim file, then simple {.importc.} for each function or type. Maybe there is uses cases where importing headers is necessary but I haven't encountered this case for the moment. (maybe it is headers only libraries ?)

---

• DeletedUser (orginal) [2018-05-04T17:43:50+02:00] view original

  About templates, they're like C macros (but hygienic), a simple way to do metaprogramming without manually building syntax like in Nim macros. They're not like procs in that they aren't called at runtime, they simply exist during compile time and transform AST given to its parameters. An untyped parameter is any raw AST that hasn't been type checked yet, and a typed parameter has been type checked and gone through AST transformation. If you name a template parameter like x: int, x will be type checked and have the int type. If you want a value to use in a template inside when expressions or other static checks instead of an expression like x: int would give, you can use static[int], but it's not entirely stable yet.

---

• ErikCampobadal (orginal) [2018-05-07T20:45:34+02:00] view original

  Thanks a lot for the answers in here, made the questions clear! Still without understanding the "untyped" one tho.

  Thanks!

---