Mirror of forum.nim-lang.org

1709 :: Could c2nim use this technique to parse C headers?

• andrea (orginal) [2015-10-14T12:03:29+02:00] view original

  I recently ran into this tweet : "The way Pixie gets type information from C header file is just genius".

  According to the comments to the tweet by the author of Pixie:

  • "we compile a program that invokes these C++ templates: https://github.com/pixie-lang/pixie/blob/master/pixie/PixieChecker.hpp … Those templates generate data about types."
  • "We then run the resulting program to dump all the type information we need to do complex imports: https://github.com/pixie-lang/pixie/blob/master/pixie/uv.pxi#L58-L91 …"

  I was wondering whether this technique - that seems pretty smart - could be used to simplify c2nim and gain better compliance with respect to C macros and so on.

---

• Araq (orginal) [2015-10-14T12:53:30+02:00] view original

  The biggest issue that c2nim is facing is that it needs better preprocessor support and Pixie's solution is to run the C preprocessor and compiler over the header file. You can run an external preprocessor over the source before invoking c2nim already but then you lose all the #define information. If you want to simplify c2nim's inner workings, replace its parser with clang's.

---

• judofyr (orginal) [2015-10-14T16:28:35+02:00] view original

  I started playing with parsing the header files with libclang, and constructing the Nim code using compiler/ast. It's relatively easy to implement, although there's a lot of cases to handle.

---