https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120013
Bug ID: 120013
Summary: SIGSEGV ICE with modules related to instantiation of
templates across partition units
Product: gcc
Version: 15.1.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c++
Assignee: unassigned at gcc dot gnu.org
Reporter: miuna.oshino at gmail dot com
Target Milestone: ---
Issue originally discussed in a C++ Discord community.
I am not familiar with g++'s internals at all so take my observations with a
grain of salt.
> In all cases g++ crashes because of including certain std headers in `hewwo`
> in the partition chain `hewwo -> uwu -> owo`.
> Many headers cause this, but we will use <chrono> as a scapegoat because it
> is my sworn enemy
>
> CRASH 1: `export std::array<char, 5> owo;`, the original scenario:
> <https://godbolt.org/z/c7cMP3roE>
> - `hewwo`: Variable with <chrono>
> - `uwu`: Alias to DIFFERENT array (no chrono)
> - `owo`: Variable of DIFFERENT array WITH <chrono>
> - Including <chrono> in all 3 fixes it.
> - Referencing `hewwo`'s array instantiation in `uwu` is ok
> - `export` or not doesn't change anything at all, on any of the 3 symbols
>
> CRASH 2: `export import :uwu;` <https://godbolt.org/z/h3Te6G6cx>
> - `hewwo`: Variable with <chrono>
> - `uwu`: Alias to DIFFERENT array **(no chrono)**
> - `owo`: Variable of DIFFERENT array **(no chrono)**
> - Happens when `owo` does not have <chrono>.
> - Including <chrono> in `uwu` but NOT in `owo` changes the crash to `export
> import :owo;` (passes through uwu without an issue)
> - Removing the instantiation of a different array symbol in `uwu` changes the
> crash to `export import :owo;`. (passes through uwu without an issue)
> - Including <chrono> in all 3 fixes it.
> - `export` or not doesn't change anything at all, on any of the 3 symbols
> - Maybe same as crash 1 but different context for diagnostics?
>
> CRASH 3: some iterator stuff <https://godbolt.org/z/oa1T1MfG5>
> - `hewwo`: Variable with <chrono>
> - `uwu`: Alias to DIFFERENT array **(no chrono)**
> - `owo`: **NOTHING**, but WITH <chrono>
> - Including <chrono> in all 3 fixes it.
> - Eerily similar again, wondering if it's just the same issue
> - **This one does give some information**
>
> I feel like crash 1 and 2 are red herrings, and crash 3 gives real information
> **Again in all 3 cases the crash happens because we include <chrono> (or
> other specific headers) in `hewwo` with those array definitions, but NOT in
> importing modules.**
>
> Crash #3 makes me think a chain of 3 module units is required for this bug
