diff options
-rw-r--r-- | README.html | 65 | ||||
-rw-r--r-- | README.md | 59 | ||||
-rwxr-xr-x | docs.sh | 2 | ||||
-rw-r--r-- | main.c | 2 | ||||
-rw-r--r-- | parse.c | 2 |
5 files changed, 125 insertions, 5 deletions
diff --git a/README.html b/README.html index fe1fd75..e34ab71 100644 --- a/README.html +++ b/README.html @@ -28,4 +28,67 @@ it is nearly as fast in theory.</p> <p><code>tests</code> has some test programs written in <code>toc</code>.</p> -<p>To compile the compiler on a Unix-y system, use</p> +<p>To compile the compiler on a Unix-y system, just run <code>build.sh</code>. You can supply a compiler by running <code>CC=tcc build.sh</code>, or built it in release mode with <code>./build.sh release</code> (which will help speed up compiling large programs). </p> + +<p>On other systems, you can just compile main.c with a C compiler. toc uses several C99 and a couple of C11 features, so it might not work on all compilers. But it does compile on quite a few, including <code>clang</code>, <code>gcc</code>, and <code>tcc</code>. It can also be compiled as if it were C++, but it does break the standard in a few places*. So, MSVC can also compile it. The <em>outputted</em> code should be C99-compliant.</p> + +<hr /> + +<h3><code>toc</code> Source Code</h3> + +<p><code>toc</code> is written in C, for speed and portability. It has no dependencies, other than the C runtime library.</p> + +<h4>Build system</h4> + +<p><code>toc</code> is set up as a unity build, meaning that there is only one translation unit. So, <code>main.c</code> <code>#include</code>s <code>toc.c</code>, which <code>#include</code>s all of <code>toc</code>'s files. This improves (from scratch) compilation speeds, since you don't have to include headers a bunch of times for each translation unit. This is more of a problem in C++, where, for example, doing <code>#include <map></code> ends up turning into 25,000 lines after preprocessing. All of toc's source code, which includes most of the C standard library, at the time of this writing (Dec 2019) is only 22,000 lines after preprocessing; imagine including all of that once for each translation unit which includes <code>map</code>. It also obviates the need for fancy build systems like CMake.</p> + +<h4>New features</h4> + +<p>Here are all the C99 features which <code>toc</code> depends on (I might have forgotten some...):</p> + +<ul> +<li>Declare anywhere</li> +<li><code>stdint.h</code></li> +<li>Non-constant struct literal initializers (e.g. <code>int x[2] = {y, z};</code>)</li> +<li>Variadic macros and <code>__VA_ARGS__</code></li> +<li>Flexible array members</li> +</ul> + +<p>The last three of those could all be removed fairly easily.</p> + +<p>And here are all of its C11 features:</p> + +<ul> +<li>Anonymous structures/unions</li> +<li><code>max_align_t</code> and <code>alignof</code> - It can still compile without these but it won't technically be standard-compliant</li> +</ul> + +<h4>More</h4> + +<p>See <code>main.c</code> for a bit more information.</p> + +<hr /> + +<p>* for those curious, it has to do with <code>goto</code>. In C, this program:</p> + +<pre><code> +int main() { + goto label; + int x = 5; + label: + return 0; +} +</code></pre> + +<p>Is completely fine. <code>x</code> will hold an unspecified value after the jump (but it isn't used so it doesn't really matter). Apparently, in C++, this is an ill-formed program. This is a bit ridiculous since</p> + +<pre><code> +int main() { + goto label; + int x; x = 5; + label: + return 0; +} +</code></pre> + +<p>is fine. So that's an interesting little "fun fact": <code>int x = 5;</code> isn't always the same as <code>int x; x = 5;</code> in C++.</p> @@ -23,9 +23,64 @@ x : int; x = 5; // Declare x as an integer, then set it to 5. `toc` is statically typed and has many of C's features, but it is nearly as fast in theory. - + See `docs` for more information (in progress). `tests` has some test programs written in `toc`. -To compile the compiler on a Unix-y system, use +To compile the compiler on a Unix-y system, just run `build.sh`. You can supply a compiler by running `CC=tcc build.sh`, or built it in release mode with `./build.sh release` (which will help speed up compiling large programs). + +On other systems, you can just compile main.c with a C compiler. toc uses several C99 and a couple of C11 features, so it might not work on all compilers. But it does compile on quite a few, including `clang`, `gcc`, and `tcc`. It can also be compiled as if it were C++, but it does break the standard in a few places\*. So, MSVC can also compile it. The *outputted* code should be C99-compliant. + +--- + +### `toc` Source Code + +`toc` is written in C, for speed and portability. It has no dependencies, other than the C runtime library. + +#### Build system +`toc` is set up as a unity build, meaning that there is only one translation unit. So, `main.c` `#include`s `toc.c`, which `#include`s all of `toc`'s files. This improves (from scratch) compilation speeds, since you don't have to include headers a bunch of times for each translation unit. This is more of a problem in C++, where, for example, doing `#include <map>` ends up turning into 25,000 lines after preprocessing. All of toc's source code, which includes most of the C standard library, at the time of this writing (Dec 2019) is only 22,000 lines after preprocessing; imagine including all of that once for each translation unit which includes `map`. It also obviates the need for fancy build systems like CMake. + +#### New features + +Here are all the C99 features which `toc` depends on (I might have forgotten some...): + +- Declare anywhere +- `stdint.h` +- Non-constant struct literal initializers (e.g. `int x[2] = {y, z};`) +- Variadic macros and `__VA_ARGS__` +- Flexible array members + +The last three of those could all be removed fairly easily. + +And here are all of its C11 features: + +- Anonymous structures/unions +- `max_align_t` and `alignof` - It can still compile without these but it won't technically be standard-compliant + +#### More + +See `main.c` for a bit more information. + +--- + + +\* for those curious, it has to do with `goto`. In C, this program: +<pre><code> +int main() { + goto label; + int x = 5; + label: + return 0; +} +</code></pre> +Is completely fine. `x` will hold an unspecified value after the jump (but it isn't used so it doesn't really matter). Apparently, in C++, this is an ill-formed program. This is a bit ridiculous since +<pre><code> +int main() { + goto label; + int x; x = 5; + label: + return 0; +} +</code></pre> +is fine. So that's an interesting little "fun fact": `int x = 5;` isn't always the same as `int x; x = 5;` in C++. @@ -1,4 +1,4 @@ -#!/bin/bash +#!/bin/sh markdown README.md > README.html for x in docs/*.md; do echo $x @@ -31,6 +31,8 @@ allow omission of trailing ; in foo ::= fn() {}? #ifdef __cplusplus #define new new_ #define this this_ +#elif __STDC_VERSION__ < 199901 +#define inline #endif #include "toc.c" @@ -1931,7 +1931,7 @@ static bool parse_file(Parser *p, ParsedFile *f) { return ret; } -#define PARSE_PRINT_LOCATION(l) //fprintf(out, "[l%lu]", (unsigned long)(l).line); +#define PARSE_PRINT_LOCATION(l) /* fprintf(out, "[l%lu]", (unsigned long)(l).line); */ /* in theory, this shouldn't be global, but these functions are mostly for debugging anyways */ static bool parse_printing_after_types; |