Stream: platform development

Topic: ✔ Zig: General protection exception (no address available)

Oskar Hahn (Dec 04 2024 at 17:24):

I ran into an memory error in a zig platform.

The roc code that is running is with lukes roc-parser. The smallest example I could build is:

import parser.Parser exposing [many, map, sepBy]
import parser.String exposing [parseStr, codeunit, codeunitSatisfies]

part1 : Str -> Result Str _
part1 = \_rawInput ->
    input = "XXX"
    _ = parseStr puzzleParser input
    Err TODO

puzzleParser =
    lineParser |> sepBy (codeunit 'X')

lineParser =
    many symbolParser

symbolParser =
    codeunitSatisfies (\c -> c != 'X') |> map \_ -> Other

When I run this, the platform crashes with the message General protection exception (no address available). I don't know what that means.

I also get a traceback. The top most functions are:

./roc_app:0:0: 0x5be3cba551d4 in #Attr_#inc_3 (roc_app)
./roc_app:0:0: 0x5be3cba54793 in List_prepend_2cc6e6d3c5a48a76ea218c439d44b6318e7bd267419a22dcb25b258a2c06a (roc_app)
./roc_app:0:0: 0x5be3cba531ff in Parser_82_d0954aeb42c3a999750fa5b4068c6679ed2537c3257fc7e6c4e91bdc4133ae0 (roc_app)
./roc_app:0:0: 0x5be3cba540db in Result_map_8acb95ddb9a746c2bf4dc0f4f96ce3b3e1f1e4f2559e7641b193db1f161d1c41 (roc_app)

I think, it is this line from the parser package: https://github.com/lukewilliamboswell/roc-parser/blob/327ae1c3f59fc998390e1ce201e1f79ac53ba98f/package/Parser.roc#L360

What does #Attr_#inc_3 mean? Can I assume, that this is from inside the builtins?

It seems to have something to do with my allocator. I am using a zig allocator and not C.alloc. If I switch to C.alloc, everything works. But I don't understand what I am doing differently. I tried to set the alignment to always 16, but this did not help.

I have no idea, how to debug this further. Do you have any idea?

Oskar Hahn (Dec 04 2024 at 17:37):

One other think. I am using zig 0.13. For the builtins in my host code, I am using the builtins from this PR. But I don't think that that this is relevant, since all the data is allocated inside roc and not from the platform.

Oskar Hahn (Dec 04 2024 at 17:55):

I think I found the bug. But I have to test it. I am am away for some hours. I report later

Anton (Dec 04 2024 at 18:09):

running it with valgrind can help provide additional info

Brendan Hansknecht (Dec 04 2024 at 19:10):

It's trying to increment a refcount

Brendan Hansknecht (Dec 04 2024 at 19:10):

Either of the list itself or an element

Oskar Hahn (Dec 04 2024 at 22:23):

Ok. I found the bug. And after I saw it, it is painful, that I even started this thread...

For the zig allocator, I have to save the size of each allocation, because allocator.free() functions requires it, and roc does not provide it.

Therefore I allocate a bit more memory and safe the size before the actual memory.

This works for roc_alloc. But for roc_realloc, I returned the pointer to my size-value instead of the pointer to the actual memory. The fix was as simple as changing return new_slice.ptr to return new_slice.ptr + alignment.

Notification Bot (Dec 04 2024 at 22:24):

Oskar Hahn has marked this topic as resolved.

Luke Boswell (Dec 05 2024 at 00:04):

Nice. I'm glad you found it. That zig allocator is really helpful for zig platforms. :grinning:

Last updated: Jul 06 2025 at 12:14 UTC