Mirror of forum.nim-lang.org

4092 :: [best practice] should `importc` procs use native argument names instead of cryptic a1, a2, a3 ?

• timothee (orginal) [2018-07-31T02:25:17+02:00] view original

  I like self documenting code.

  The import procs are nothing but self-documenting, see for eg lib/posix/posix.nim:

  proc fmtmsg*(a1: int, a2: cstring, a3: cint,
                a4, a5, a6: cstring): cint {.importc, header: "<fmtmsg.h>".}
  Run

  (many such examples)

  why not use instead the native argument names where the proc is defined, eg: from https://linux.die.net/man/3/fmtmsg

  
  int fmtmsg(long classification, const char *label,
             int severity, const char *text,
             const char *action, const char *tag);
  Run

  so we'd have:

  proc fmtmsg*(classification: int, label: cstring, severity: cint,
                text, action, tag: cstring): cint {.importc, header: "<fmtmsg.h>".}
  Run

  this is especially important when named arguments are being used.

  How about:

  • adding to contributing.md (see https://github.com/nim-lang/Nim/pull/8271) an note on that for new code
  • accepting PR's to fix these for old code

---

• DeletedUser (orginal) [2018-07-31T02:42:02+02:00] view original

  I don't know if there's anything to oppose here? Was a PR for this denied?

---

• timothee (orginal) [2018-07-31T03:01:22+02:00] view original

  no, but better to have a documented policy to follow for such things instead of debating it for each PR. If there's any objection to this policy, here's a good place to debate this.

---

• mratsim (orginal) [2018-07-31T08:35:12+02:00] view original

  I'm all for this but to be fair, C2nim just uses the names present in the translated header file if any.

  In this case, there was no named arguments

  int fmtmsg(long, const char *, int, const char *, const char *, const char *);
  Run

  There are also several libraries where the names are poor single letter variables. I.e. if upstream doesn't properly name their own parameters, c2nim can do anything.

---

• Bulat_Ziganshin (orginal) [2018-08-03T10:19:55+02:00] view original

  I would rather propose to support nameless parameters support in Nim. Usecases:

  1. function prototypes as in your example
  2. declaring type of procedural parameter:

  proc f (p: proc(int):int): void = ...

  3. declaring single-value parameter:

  proc f(typedesc[int]): void = ....

---