making var comptime yields SIGBUS (Misaligned address error) during runtime

[benmkw]

I am attempting to create a trie during compile time and use it for string matching (similar to the perfekt hashing approach from Andrews blog post) but ran into this problem.

source

on line 158 changing the var to be comptime results in runtime crash, while omitting the comptime works fine (whole program just executed at runtime). comptime var nodes = [_]TrieNode{TrieNode{}} ** size;

The crash happens on line 54 child.*.?.value = .{ .char = character };

I need nodes to be comptime in order to make const trie = create_trie_from(&input, &trie_node_pool); comptime and the idea is that LLVM can do something useful with that information.

zig version 0.5.0+bda355d97 (current master)

[benmkw]

I played a bit with creduce and created this minimal example (some manual editing was done but creduce did some really cool things as well)

const s = struct {
    p: ?*void = null,
};

pub fn main() void {
    comptime var a = s{}; // remove this comptime and it works
    var c = &a; // change this to const and it works
    c.p = null; // a.p = null works
}
}

'zig run 5reduce.zig' terminated by signal SIGBUS (Misaligned address error)

I also tried the ir-clean-up-vars branch at 01c2eba where this issue remains

creduce without help

for reference, this is what creduce achieves all on its own. After this, some manual inlining helps it to make some more progress.

const a = struct {
    b: [128]?*a = [_]?*a{null} ** 128,
};
const c = struct {
    d: []a,
    e: usize,
    f: usize,
    fn allocate(g: c) *a {
        return &g.d[g.e];
    }
};
fn h(i: *a, j: []const u8, k: *c) void {
    var l = i;
    for (l.b) |*child| child.* = k.allocate();
}
fn m(inputs: [][]u8, k: *c) *a {
    var i = k.allocate();
    for (inputs) |j| h(i, j, k);
    return i;
}
pub fn main() void {
    var input = [_][]u8{""};
    const o = 0;
    comptime var d = [_]a{a{}};
    var k = c{
        .d = &d,
        .f = o,
        .e = 0,
    };
    const p = m(&input, &k);
}

[andrewrk]

That's good to know that creduce can create zig test cases! Do you think you would be willing to write a tutorial on how to use creduce with zig?

[benmkw]

I don't have a website or such, feel free to use/ distriubute as you want, I basically just did the following after looking at this, (the official website would somehow not load for me):

creduce  --not-c ./reduce.rb main.zig # yeah really, don't like bash, could use anything here, e.g. zig

where reduce.rb is

#! /usr/bin/env ruby

ret = `zig run main.zig 2>&1`

# or 
# ret = `zig fmt main.zig && zig run main.zig 2>&1`


status =  $?.exitstatus

if status == 138 # SIGBUS
    exit 0 # zero means, continue reducing 
else
    exit 1 # means, this is a branch which lost the bug we wanted to find
end

It's really fun to watch main.zig in your editor while creduce works on it.

I had some issues where creduce would reduce to an empty file after just a few seconds but my return value checks were wrong in these cases, writing the right reduce.rb script was actually a bit tricky although it seems easy now.

for the--not-c see here

I think if creduce would understand functions and could inline them that would help a lot (I presume the c version would be able to do this), or if zig could preprocess code and inline functions (not recursive ones) that may also be an (easier) option. The reason for this is that it seems as if it can't really delete function arguments at the moment so these are a barrier which stops propagation of deletions.

[andrewrk]

Thank you so much! This is really useful. I'll compile this information and put it into either the wiki or CONTRIBUTING.md.

[benmkw]

Your welcome!

Maybe you could add to it by converting the ruby script to zig and adding the option/ an example to search stdout/ stderr such that its really easy to combine textual things and return value. I find myself spending a bit of time figuring these things out every time they reoccur.

[Vexu]

Duplicate of #1487