Wow, thanks a lot for hunting this down!

i discovered ...

Our of curiosity: how? 🙂

If I read this correctly, there are two bugfixes here:

1. CAF re-allocated memory for the type ID list too early (the off-by-one part)
2. CAF allocated too much memory for deserialized messages because it always added the padded size for whatever type has type ID 0

These are very subtle bugs, so how did you find them? Memory profiling?

Codecov Report

Attention: 1 lines in your changes are missing coverage. Please review.

Comparison is base (3fa129d) 66.65% compared to head (4886323) 67.27%.
Report is 78 commits behind head on master.

@@            Coverage Diff             @@
##           master    #1614      +/-   ##
==========================================
+ Coverage   66.65%   67.27%   +0.61%     
==========================================
  Files         644      652       +8     
  Lines       27538    27774     +236     
  Branches     2925     3013      +88     
==========================================
+ Hits        18356    18684     +328     
+ Misses       7686     7541     -145     
- Partials     1496     1549      +53     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

I explored how the std::set<std::string> (from caf::actor_system::mpi) for message_types is used and how void is encoded as type_id_t. Then changed the 0 in CAF_BEGIN_TYPE_ID_BLOCK(core_module, 0) to 1 and got a segfault, in some unrelated function but pinned it down to the load_data function

The segfault occured at