Re: [Tinycc-devel] VFS support?

[grischka]

On 14.08.2023 14:34, Steve Fan wrote:
> I want to implement a compressed, read-only header manifest embedded directly 
> into the compiler itself. This way I can have one single static binary to run 
> everywhere without headers.
>
> I want to see if we can wrap the C file manipulation API with a custom TCC flavor, 
> i.e. open to be tcc_open_impl, read to be tcc_read_impl, lseek to be tcc_lseek_impl, 
> and basically replace (fopen|fread|fwrite|fclose|fseek|fdopen|fprint|ftell|) to 
> tcc_<$1>_impl. And then let the user how it is implemented using a new config 
> def like CONFIG_TCC_FS_USER (and by default substitute to the equivalent C library 
> call)
>
> I need this because mere header substitution still has edge cases. I speculate 
> that Clang will replace those symbols with an inline direct system call so my C 
> preprocessor definition hack didn't work on optimized code.

The recommended order of things is to include system headers first, then
redefine and declare.  Like

   #include <unistd.h / io.h> // supposed to declare "int open();" etc.

   #define open vfs_open
   int vfs_open(const char*, int, ...);
   ...

   #include "tcc.c"
   #include "vfs_impl.c"

-- gr

> On debug-mode, this works surprisingly well.
>
> Having said that, I emulated the C file manipulation functions using Rust with 
> a lightweight VFS layer, embedding the headers with compressed source text and 
> deflated them in runtime on demand (via some macro magic in Rust), and it is 
> working amazing well on Windows for both debug and release mode, although I 
> just came across the weird problem in the previous paragraph on Linux for 
> release mode.
>
> Stefan
>
>
> _______________________________________________
> Tinycc-devel mailing list
> Tinycc-devel@nongnu.org
> https://lists.nongnu.org/mailman/listinfo/tinycc-devel