Post · bonfire.cafe

what i dislike about c++ is header files should have the interface under the header, while code files should only have the code to implement the interface

separating this way makes it reasonably easy to go "oh, this part of the codebase is in this interface, so the implementation is in this code file"

but where do c++ templates fall ? they're meant to be consume, they're generating code, they're basically fancy macros, they need to be in the header

but they're also code

congrats! now code has two places where it may be, and the model of the interface and the implementation being separate is broken.

d@nny disc@ mc²

@hipsterelectron@circumstances.run · 5 days ago

@SRAZKVT rust handles this by just dumping the internal representation in an unstable binary format into the intermediate layer but for various reasons fails to then use that to actually solve this lol

sarah tonin :wlfBlep:

@SRAZKVT@tech.lgbt · 5 days ago

@hipsterelectron rust has everything in the same files, like most other languages

it's not good, but it's not actively bad either

even c isn't perfect, macros also fit in headers, and some people write absolutely fucked up things in headers (single header libraries for example)

witch_t *navi

@navi@social.vlhl.dev · 5 days ago

@SRAZKVT @hipsterelectron

macros become so more manageable when you do as little as possible on them and then defer to a function

honestly people would hate them so much less if that was the only usage

---

honestly the only annoying part of that for c is, having to manually declare function signatures in two places (codegen tooling could solve this, by e.g., generating a header with all non-static functions in a given .c file)

i was gonna say having to do internal headers was also annoying but honestly, that's for the best because it means internal interfaces are also nudged to be consistent

Federation Bot

@Federation_Bot · 5 days ago

@navi @hipsterelectron to me interface headers are the ones who should be written by hand, this way you could have a tool that checks if there is any function that isn't marked as static that also isn't in header, and gets flagged, and then, either it's forgotten from header, so you know to add it, or it should be static, and you fix it

not many people check generated files, so while you COULD check it, nearly nobody would

1 more replies

d@nny disc@ mc²

@hipsterelectron@circumstances.run · 5 days ago

@navi @SRAZKVT i'm using automake now and it's unfortunately very comfy and allows me to not build a whole build system before writing some got damn c code. hoping that getting more familiar with this (and maybe doing automake extensions, it's written in perl) will be a good stepping stone towards a Good Build Tool for C

sarah tonin :wlfBlep:

@SRAZKVT@tech.lgbt · 5 days ago

not many people check generated files, so while you COULD check it, nearly nobody would

d@nny disc@ mc²

@hipsterelectron@circumstances.run · 5 days ago

@SRAZKVT c++ could have solved this with modules but they made sure to make it completely fucked up. i'm going to write a module mapper at some point but sticking with c for now

Luzzy :verified_trans:

@meluzzy@woof.tech · 5 days ago

@hipsterelectron @SRAZKVT sure modules introduced in C++20 are already implemented in the most popular compilers right? RIGHT?
Hahahaha (dies)

1+ more replies

d@nny disc@ mc²

@hipsterelectron@circumstances.run · 5 days ago

@SRAZKVT the point i was making with rust is that it employs an implementation technique that could be used to solve exactly this problem. what you've identified here is imo an interface issue and not an implementation problem—it's a problem because of how the interface fails to model the appropriate program structure, and we can solve it by cheating

1+ more replies