pervognsen / bitwise Goto Github PK

I tried to do this

func main(argc: int, argv: (char*)[]): int {
    return 0;
}

Should type[] be just a synonym for type*? Or do you plan to support arrays with unknown size? (which are just pointers anyway)

struct A
{
  b: B;
}
// says Unresolved type name 'B'
#static_assert(sizeof(A) == 4)

struct B
{
  data: int32;
}

The workaround is to rearrange declarations so that B is defined before A.

Package b:

enum Em{
    Em1
}

Package c:

  import b{Em}

  func main() : int {
      em : Em = Em1;
      return 0;
  }

Running:

 ~/src/bitwise/ion$ git pull && gcc main.c
 Already up-to-date.

 ~/src/bitwise/ion$ IONHOME=. ./a.out c
 /mnt/tera1/devel/src/bitwise/ion/system_packages/builtin/types_win32.ion(4): warning: Unknown declaration #directive 'static_assert'
 /mnt/tera1/devel/src/bitwise/ion/system_packages/builtin/types_win32.ion(5): warning: Unknown declaration #directive 'static_assert'
 /mnt/tera1/devel/src/bitwise/ion/c/c.ion(4): error: Unresolved name 'Em1'

Using import b{...} makes everything compiling fine, but importing enum should probably import values as well, as otherwise it's not very useful, but pretty surprising

(Also I'm on linux, I thought _win32.ions are not parsed)

buf_free should set the pointer to NULL

#define buf_free(b) ((b) ? free(buf__hdr(b)) : 0, b = NULL)

int *asdf = NULL;
buf_push(asdf, i);
buf_free(asdf);
assert(asdf == NULL);

[edited]

The ion binary compiles fine with gcc main.c

However, the Ion tests don't compile unless I duplicate linux files:

cp stdio_linux.ion stdio_osx.ion
cp time_linux.ion time_osx.ion

Also, on OSX (10.13 High Sierra, XCode 9.3) clock_t is defined as unsigned long:

@foreign typedef time_t = long;
@foreign typedef clock_t = ulong;

from mac system headers

typedef unsigned long       __darwin_clock_t;   /* clock() */
typedef long                __darwin_time_t;    /* time() */
typedef __darwin_time_t     time_t;
typedef __darwin_clock_t    clock_t;

stdio_osx.ion:

@foreign const _IOFBF = 0;
@foreign const _IOLBF = 1;
@foreign const _IONBF = 2;

@foreign const BUFSIZ		= 1024;
@foreign const FILENAME_MAX	= 1024;
@foreign const FOPEN_MAX	= 20;
@foreign const TMP_MAX		= 308915776;
@foreign const L_tmpnam		= 1024;

from mac system headers

#define _IOFBF  0       /* setvbuf should set fully buffered */
#define _IOLBF  1       /* setvbuf should set line buffered */
#define _IONBF  2       /* setvbuf should set unbuffered */

#define BUFSIZ  1024    /* size of buffer used by setbuf */

                                /* must be == _POSIX_STREAM_MAX <limits.h> */
#define FOPEN_MAX   20          /* must be <= OPEN_MAX <sys/syslimits.h> */
#define FILENAME_MAX    1024    /* must be <= PATH_MAX <sys/syslimits.h> */

/* System V/ANSI C; this is the wrong way to do this, do *not* use these. */
#ifndef _ANSI_SOURCE
#define P_tmpdir    "/var/tmp/"
#endif
#define L_tmpnam    1024        /* XXX must be == PATH_MAX */
#define TMP_MAX     308915776

#24 (comment)

I'm currently a bit behind your schedule so I might have missed some information, but are there news about

• defer statement
• no ; needed
• formatter
• C output/ C single header output
• completion plugins/ output
• renaming
  .. being added to ion?

thanks

These expressions in print_test()

expr_binary('+', expr_int(1), expr_int(2)),
expr_unary('-', expr_float(3.14)),

currently print

(*= 1 2)
(%= 3.140000)

The char literals probably should be replaced by TOKEN_ADD and TOKEN_SUB.

Since you state that the structure will change based on the feedback:

I am highly interested to see this in functional styles also.

You can show and explain the different abstraction layers and how they are composed together. 🙂

Thousands are very interested to see somebody with empathy and deep understanding doing this. 😄

There is so much material out there for the imperative approach and so less for the functional one (yet). 🙂

Rust has very functional approaches and can go very low level, while its suitable for many of the tasks you describe, since its without garbage collection. 🙂

Haskell is capable to do GPU programming and highly focused on education by its design.

F# can do GPU programming as well and is relevant in the .Net space, which performs amazingly well today on .Net Core. 🙂

Compilers are a strenght of these languages so is the initial compiler from Rust written in OCaml 😃

I think, since you aim to give educational courses is it interesting to show how things can be done in other ways as the traditions tell us.

I see a couple of assumptions in our community, who actually slow down our productivity and the personal development.

Portable low level code is possible with other languages as C, languages who are eventually more suited for the different tasks. 😃

I appreciate what you are doing and hope you find a lot of use cases to demonstrate the beauty of functional languages.

And I am sure, thousands over thousands see it very similar. 😃

Currently functions that return a pointer generate c code that puts the * character inside brackets with the function name, like this:

int (*test)(void) {
return 0;
}

This doesn't seem to be valid syntax (cl, gcc and clang all complain).

Why does the next_token() function use such a huge switch case when it could be just a few if statements? Just curious if this is just a personal preference of @pervognsen or something else.

// This is how I would see this done.
void next_token() {
    if (*stream >= '0' && *stream <= '9') {
        uint64_t val = 0;
        while (isdigit(*stream)) {
            val *= 10;
            val += *stream++ - '0';
        }
        token.kind = TOKEN_INT;
        token.val = val;
    } else if ((*stream >= 'a' && *stream <= 'z') || (*stream >= 'A' && *stream <= 'Z') || *stream == '_') {
        const char *start = stream++;
        while (isalnum(*stream) || *stream == '_') {
            stream++;
        }
        token.kind = TOKEN_NAME;
        token.start = start;
        token.end = stream;
    } else {
        token.kind = *stream++;
    }
}

See repro at: uucidl@a96b036

ion -os osx -arch x64 bugs && \
  cc out_bugs.c -o bugs.elf && \
  ./bugs.elf ; echo "rc:$?"

Expected:

Processed 101 symbols in 2 packages
Generated out_bugs.c
rc:42

Got:

Processed 101 symbols in 2 packages
Generated out_bugs.c
<foo>/Desktop/bitwise/ion/bugs/size_bug.ion:10:25: warning:
      incompatible pointer types passing 'ullong *' (aka 'unsigned long long *')
      to parameter of type 'size_t *' (aka 'unsigned long *')
      [-Wincompatible-pointer-types]
    fetch_value(&(num), &(num_size));
                        ^~~~~~~~~~~
<foo>/Desktop/bitwise/ion/bugs/size_bug.h:3:41: note: passing argument
      to parameter 'dest_size_ptr' here
int fetch_value(void* dest_ptr, size_t* dest_size_ptr);
                                        ^
1 warning generated.
rc:42

I would have expected C apis expressed in terms of size_t to be
exposed in ion via usize. It appears that on Macos at least:

cc -E out_bugs.c | grep -E '(typedef.* uint64_t;|typedef.* size_t;)'

typedef unsigned long long uint64_t;
typedef long unsigned int size_t;

Since usize is defined as uint64, we see that these types differ.

You described being fluent in C as a prerequisite for Bitwise. Could you describe it in a little more detail? I only have a beginner's level of experience with C and with programming in general, so I'd like to know how much work there is ahead of me to be able to go with the flow.

I remember you said the recording setup is wip and will improve. So feel free to close.

Nevertheless I‘d like to point out that 4K would be a good option, makes the image/ text look much cleaner, YouTube supports it and especially you said this is a resource for future visitors and in 5 years 1080 will be a crutch and not appropriate for the quality content of the videos.

If 4K is not an option for some reason maybe at least 1440p would do for some time.

Thanks

Unless I've missed, ion currently has no conditional compilation. Tying that with some built-in consts for OS and such would be great. This requires falling out to c whenever this is needed, which is one place in the current ion compiler.

An implementation for this might possibly be tied up with some implementation of macros, if any.

(As an aside, I'm mostly against general compile time code execution, because I'm afraid of people getting carried away with it. But it's technically a potential solution for these sorts of things, too.)

I don't know what plans exist on macros, and I know there are lots of ways to do them with lots of pros and cons. That said, they are used extensively in the ion compiler in c. They might possibly be worth adding to ion, and in a way familiar to c programmers, but better. Something perhaps to consider, at least.

Here's maybe some syntax for syntactic macros:

macro DO_SOMETHING(x, y) {{{
    // Replacement syntax between braces.
    // Semi heredoc-ish, allow arbitrary number of opening braces,
    // so we can ignore braces in the middle.
    // And yes, this particular macro is useless.
    }
}}}

This requires a prepass and presumably these macros are non-hygienic, like c macros. Some syntax for concatenation or string formatting would also be good, like c.

I'd also recommend they be automatically namespaced like everything else in ion.

One other more innovative idea follows, but it wouldn't cover all use cases of standard syntactic macros. For cases it does cover, it would be more embedded in the language. First start with any old function:

// ** here because this isn't a macro, and we might need to change the pointer.
func buf_push(b: char**, elem: char) {
    buf_fit(b, 1 + buf_len(*b));
    (*b)[buf__hdr(*b).len++] = elem;
}

This function can then be called normally, or else it can be called macro-expanded:

// Type explicit here just to emphasize macro-ized call.
field: CompoundField = parse_expr_compound_field();
// Here, I use a trailing bang to tell the compiler I want it to create an ad hoc
// variation of the original function.
buf_push!(&fields, field);

In this case, I'd still recommend it create a new name-mangled function to match the call site.

As a final possibility of my own, I could also imagine importing packages, including the current package, with redefinitions of types or constants. As for the ad hoc function type replacement, it wouldn't cover all use cases of syntactic macros.

Of course, one could go all out with custom parsers like in Rust or any number of other ideas, but I think all that goes beyond the scope of a c-like language.

This doesn't work at present:

token.pos.name = name ? name : "<string>";

It gives the following error:

error: Left and right operands of ternary expression must have arithmetic types or identical types

Instead, I have to cast:

token.pos.name = name ? name : (:char const*)"<string>";

It would be nice if the compiler could auto-coerce string literals to char const*.

There might be other cases where this applies, too, beyond ternaries. It's just the case where I remember I've run across it.

The following program illustrates the issue:

`import libc {...}

func main() {
x := 1.0d;
y := x + 1e-10d;
printf("x=%.12f y=%.12f\n", x, y);
printf("y-x = %.12f\n", y-x);

#assert(fabsd(y-x) < 2e-10d);

}`

The printf's work as expected, but the assertion fails. This is the output:
`x=1.000000000000 y=1.000000000000

y-x = 0.000000000000

ionprobl: /home/amoura/ion-probl/tst.ion:9: main: Assertion `(fabs((y) - (x))) < (0.000000)' failed.

Aborted (core dumped)
`
It looks like while scanning the assertion, the literal 2e-10d is scanned as 0.000000.

See f7e447b

Is there a way to see the code after the changes occurred on day 62 and 63 yet?

(I apologize if this is the wrong place to ask)

Some identifiers used in the ion c compiler are reserved words in ion, such as var or import. When controlling a whole api, this can be worked around by changing the identifier name. However, if someone needed to use a third-party api with naming conflicts, they'd have to make adapters in c before they could use it.

It might be worth making some escape mechanism (like a prefix such as $identifier) for such cases.

An added complexity is that this would allow expression of any identifier name in ion, which could then break a c back end, so I guess the c back end would then need to have a way to change those identifiers to some safe word.

This one's a little pedantic.
I was reading the lexer code (I've fallen a bit behind) and it seems there's nothing preventing "1.2.3.4" from parsing without error.

diff --git a/ion/lex.c b/ion/lex.c
index 56087ea..eb5e534 100644
--- a/ion/lex.c
+++ b/ion/lex.c
@@ -611,11 +611,14 @@ void lex_test(void) {
     assert_token_eof();
 
     // Float literal tests
-    init_stream("3.14 .123 42. 3e10");
+    init_stream("3.14 .123 42. 3e10 1.2.3.4");
     assert_token_float(3.14);
     assert_token_float(.123);
     assert_token_float(42.);
     assert_token_float(3e10);
+    assert_token_float(1.2);
+    assert_token_float(.3);
+    assert_token_float(.4);
     assert_token_eof();

In ion, any pointer type seems to auto cast to void*, but they don't auto-cast to void**. For example, I have to cast from char** to void** here:

str: char* = NULL;
buf_push((:void**)&str, &val, 1);

Where I elsewhere have the following function:

func buf_push(b: void**, elem: void*, elem_size: usize) {
    buf_fit(b, 1 + buf_len(*b), elem_size);
    memcpy((:char*)*b + elem_size * buf__hdr(*b).len++, elem, elem_size);
}

If I leave out the cast, it says the following:

error: Invalid type in function call argument. Expected void**, got char**

I understand that there are bizarre edge cases for pointers on various ancient or strange systems. So maybe this is by design because of edge cases that I haven't studied well. But in my mind, this ought to work. If it should work, then it would be nice to automate it.

If it shouldn't work, then I clearly need to design the example function differently ...

GCC I believe supports case ranges. It helps a lot.

void next_token()
{
        switch (*stream) {
        case '0'...'9':
                token.kind = TOKEN_INT;
                token.val = 0;
                while (isdigit(*stream)) {
                        token.val *= 10;
                        token.val += (*stream - '0');
                        stream++;
                }
                break;
        case 'a'...'z':
        case 'A'...'Z':
        case '_':
                token.kind = TOKEN_NAME;
                token.start = stream++;
                while (isalnum(*stream)
                        || *stream == '_') stream++;
                token.length = stream - token.start;
                break;
        default:
                token.kind = *stream++;
        }
}

It is 3 dots, not 2.

If the last character of a source file/string is a dot (.), it will likely crash, since you assume that there's a character after it

Please create tags/ releases/ branches for videos.
I think tags are appropriate but anything would do.

When revisiting the videos it will make it WAY easier to look at the code from that time.

Thanks for your work, much appreciated!

 path : char[256];   // in C: char (path[256]);
 p := path;          // in C: char (p[256]) = path;

gcc -std=c11 test.c returns test.ion:4:21: error: invalid initializer, as initializing arrays such way is no-no.

Which leads to more general question - should p be an array(as now), pointer to char(as c++), pointer to array of chars or forbidden at all.

Steps to reproduce:

1. Create a package libfoo with one file with the following contents:

struct Foo {
    x, y: int;
}

func bar() {
    foo: Foo;
    foo_p := &foo;
    foo_pp := &foo_p; // This triggers the type info generation for 'libfoo_Foo'
}

2. Create a second package with a main method and import libfoo {...}

Ion compilation works, but the generated C will error on compilation with error C2065: 'libfoo_Foo': undeclared identifier.

You'll get a line similar to:
[190] = &(TypeInfo){TYPE_PTR, .size = sizeof(void *), .align = alignof(void *), .base = TYPEID(53, TYPE_PTR, libfoo_Foo *)},

If the struct Foo isn't directly used in the main package, the C struct def for Foo will never be generated.

For kicks, I've been porting the ion compiler to ion. Mostly trying to retain things unchanged.

As far as I can tell, ion doesn't currently support anonymous and/or ad hoc aggregates, such as the following:

struct Typespec {
    TypespecKind kind;
    SrcPos pos;
    Typespec *base;
    union {
        const char *name;
        struct {
            Typespec **args;
            size_t num_args;
            bool has_varargs;
            Typespec *ret;
        } func;
        Expr *num_elems;
    };
};

I could imagine expressing that like so in ion:

struct Typespec {
    kind: TypespecKind;
    pos: SrcPos;
    base: Typespec*;
    union {
        name: char const*;
        function: struct {
            args: Typespec**;
            num_args: usize;
            has_varargs: bool;
            ret: Typespec*;
        }
        num_elems: Expr*;
    }
}

Meanwhile, I've been pulling out the nested types and giving the fields names where missing. That's doable, but some things, like Expr, get pretty intense about it. Do you have the intention to add these features to ion?

(And macros are another matter, but I've been working around those for now, too.)

In porting the ion compiler to ion, I had conflicts with ion builtins, specifically the TYPE_... constants. I ended up renaming the TYPE_... constants in the compiler to CMPL_TYPE_..., though it's easy to mix things up, still.

Maybe this isn't likely to trip people up, but any ongoing additions to builtins can break existing code.

I'm not sure the right way to address this matter. Some possibilities:

• Just ignore it. Maybe won't cause much trouble.
• Promise never to add new builtins (including new constants on existing enums). Could freeze builtins and add a new std library or whatnot for anything new, and it needs explicitly imported.
• Never auto-import anything. This would clearly be a breaking change and would make uses of the very basics more awkward.
• Allow per-file opt-out of (particular?) builtins.

Maybe other options, too.

When I compile Ion with GCC on Linux, it prints a bunch of warnings from builtin/types.ion. It's been happening for a while now. Here is a shell session reproducing the issue.

david@davidvm5 ~/bitwise (per_master)
$ git log -1
commit b5c2d0ad7132e666a69e402019e8731d204e2445 (HEAD -> per_master, origin/master, origin/HEAD)
Author: Per Vognsen
Date:   Wed Jun 6 16:15:23 2018 +0700

    #define __USE_MINGW_ANSI_STDIO 1 to support mingw-gcc
david@davidvm5 ~/bitwise (per_master)
$ cd ion
david@davidvm5 ~/bitwise/ion (per_master)
$ gcc main.c -o ion
david@davidvm5 ~/bitwise/ion (per_master)
$ cd ..
david@davidvm5 ~/bitwise (per_master)
$ echo $IONHOME
/home/david/bitwise/ion
david@davidvm5 ~/bitwise (per_master)
$ ./ion/ion testpkg
/home/david/bitwise/ion/system_packages/builtin/types.ion(1): warning: Unknown declaration #directive 'foreign'
/home/david/bitwise/ion/system_packages/builtin/types.ion(2): warning: Unknown declaration #directive 'foreign'
/home/david/bitwise/ion/system_packages/builtin/types.ion(4): warning: Unknown declaration #directive 'static_assert'
/home/david/bitwise/ion/system_packages/builtin/types.ion(5): warning: Unknown declaration #directive 'static_assert'
/home/david/bitwise/ion/system_packages/builtin/types.ion(6): warning: Unknown declaration #directive 'static_assert'
/home/david/bitwise/ion/system_packages/builtin/types.ion(7): warning: Unknown declaration #directive 'static_assert'
/home/david/bitwise/ion/system_packages/builtin/types.ion(8): warning: Unknown declaration #directive 'static_assert'
/home/david/bitwise/ion/system_packages/builtin/types.ion(9): warning: Unknown declaration #directive 'static_assert'
/home/david/bitwise/ion/system_packages/builtin/types.ion(10): warning: Unknown declaration #directive 'static_assert'
/home/david/bitwise/ion/system_packages/builtin/types.ion(11): warning: Unknown declaration #directive 'static_assert'
/home/david/bitwise/ion/system_packages/builtin/types.ion(12): warning: Unknown declaration #directive 'static_assert'
/home/david/bitwise/ion/system_packages/builtin/types.ion(13): warning: Unknown declaration #directive 'static_assert'
/home/david/bitwise/ion/system_packages/builtin/types.ion(14): warning: Unknown declaration #directive 'static_assert'
/home/david/bitwise/ion/system_packages/builtin/types.ion(15): warning: Unknown declaration #directive 'static_assert'
/home/david/bitwise/ion/system_packages/builtin/types.ion(17): warning: Unknown declaration #directive 'static_assert'
/home/david/bitwise/ion/system_packages/builtin/types.ion(18): warning: Unknown declaration #directive 'static_assert'
/home/david/bitwise/ion/system_packages/builtin/types.ion(19): warning: Unknown declaration #directive 'static_assert'
/home/david/bitwise/ion/system_packages/builtin/types.ion(20): warning: Unknown declaration #directive 'static_assert'
/home/david/bitwise/ion/system_packages/builtin/types.ion(21): warning: Unknown declaration #directive 'static_assert'
/home/david/bitwise/ion/system_packages/builtin/types.ion(22): warning: Unknown declaration #directive 'static_assert'
/home/david/bitwise/ion/system_packages/builtin/types.ion(23): warning: Unknown declaration #directive 'static_assert'
/home/david/bitwise/ion/system_packages/builtin/types.ion(24): warning: Unknown declaration #directive 'static_assert'
/home/david/bitwise/ion/system_packages/builtin/types.ion(25): warning: Unknown declaration #directive 'static_assert'
/home/david/bitwise/ion/system_packages/builtin/types.ion(26): warning: Unknown declaration #directive 'static_assert'
/home/david/bitwise/ion/system_packages/builtin/types.ion(27): warning: Unknown declaration #directive 'static_assert'
/home/david/bitwise/ion/system_packages/builtin/types.ion(28): warning: Unknown declaration #directive 'static_assert'
/home/david/bitwise/ion/system_packages/builtin/types.ion(29): warning: Unknown declaration #directive 'static_assert'
/home/david/bitwise/ion/system_packages/builtin/types.ion(30): warning: Unknown declaration #directive 'static_assert'
/home/david/bitwise/ion/system_packages/builtin/types_win32.ion(4): warning: Unknown declaration #directive 'static_assert'
/home/david/bitwise/ion/system_packages/builtin/types_win32.ion(5): warning: Unknown declaration #directive 'static_assert'
/home/david/bitwise/ion/system_packages/builtin/types_x64.ion(1): warning: Unknown declaration #directive 'static_assert'
Processed 100 symbols in 2 packages
Generated out_testpkg.c
david@davidvm5 ~/bitwise (per_master)
$ ls testpkg/
test.ion
david@davidvm5 ~/bitwise (per_master)
$ cat testpkg/test.ion 
func main() : int {
  return 1;
}

this is just me trying out the compiler cause I was exited to see it working, if this kind if report is not yet appropriate feel free to close!

what I noticed:

• The compiler does seem to forward declare everything. In that case you'd not need to do any reordering algorithm so I think this needs to be corrected/ enabled correctly.

This was just a bit strange to me cause after all this effort of reordering and traversing the effect does not really show through.

• var declarations are reordered even if they are not used and thus there would be no reason to reorder them
• line numbers in syntax errors do not always get propagated/ reported
• the following produces an assertion error line 577 resolve.c which disappears if one does a := f();

func f(){

}

func f2(){
    f();
}

• redefinition of a variable does not produce an error, the following compiles:

var c = 4
var c = "abd"

• var declarations do not require a semicolon, the rest does

On the other hand I'm impressed how type inference is actually working!

I hope this was not disturbing, I was just trying to get an overview and give some basic feedback, not doing anything fancy or trying to brake ion.

I'm exited to see this working 🎉 🎈

One thing you mentioned today was editor support for sublime/ vim? I could not yet find something like that yet but if someone could give a link to something (WIP) that would be nice as well!

struct Thing { a: int; }
func returns_ptr(): Thing*;

func set_value(): void
{
	returns_ptr().a = 5;
}

produces

error: Cannot assign to non-lvalue

Is this intended? Currently resorted to doing

func set_value(): void
{
    	thing := returns_ptr();
    	thing.a = 5;
}

This might be a dumb question, but how are you supposed to build this? There is no makefiles, shell scripts or cmake files. Sorry if that's an obvious question.

@foreign
func memcpy(dst: void*, src: void const *, size: usize): void*;

compiles ok, but outputs

[1] = &(TypeInfo){TYPE_VOID, .name = "void"},
  ...
[35] = &(TypeInfo){TYPE_CONST, .size = sizeof(void const ), .align = alignof(void const ), .base = 1},

(note the missing *), leading to the C compiler to report:

main.ion(292): warning C4034: sizeof returns 0
main.ion(292): error C2714: alignof(void) is not allowed

The line number is also off (past the end of the ion file), but this might be a knock-on effect.

Hi author,

I subscribed your youtube channel, and this repository. However, I haven't seen any activities on this project since last 4 months.

Can you let me know: will you continue on this project?

I'm not sure if this will lend itself to the idea, but if the video streams are not heavily reliant on graphics, it would be nice to have an audio only version available that can be augmented with the articles and reviewing the code in the repo later on. This would be beneficial for those of us that don't have a lot of time to watch videos while at a screen, but spend several hours a day stuck in traffic or doing other tasks that don't require a lot of our attention.

At: https://github.com/pervognsen/bitwise/blob/master/ion/ion.c#L79

On my machine, the bool at startup was set to true despite not having passed the corresponding check flag. This lead to ion.exe not producing any out_foo.c file

I believe other flags like target_os and target_arch are also uninitialized. My understanding of C is that only static variables are initialized to zero

With the following:

struct FooArray
{
    x : int[3];
}

struct FooWithConstArray
{
    array : FooArray const*;
}


func test_const_members() {
    foo := FooWithConstArray{};
    #assert(foo.array.x[0] == 0);
}

Then when TypeInfo gets written, ion generates:

    [53] = &(TypeInfo){TYPE_CONST, .size = sizeof(int (const [3])), .align = alignof(int (const [3])), .base = TYPEID(50, TYPE_ARRAY, int [3])},

the int (const [3]) is not parsed as a valid type expression

Somehow the type entry seems invalid to me, as all arrays in C are const. So the type info should probably not be generated at all (TYPE_CONST with base TYPE_ARRAY)

type_to_cdecl could be instrumented to assert that a TYPE_CONST should not be an array