[spam][crazy][log] gdb, thread local, libgit, instruction pointet

[Undiscussed Groomed for Male Slavery, One Victim of Many]

an extant concern is this:

Loading commits for ../repos/NWScript/

commit: a7278603f354eb683296f443e49621e50d35fcf0
looping over diff: a7278603f354eb683296f443e49621e50d35fcf0
process2: /usr/include/rapidjson/internal/stack.h:129: T*
rapidjson::internal::Stack<Allocator>::PushUnsafe(std::size_t) [with T
= char;
 Allocator = rapidjson::CrtAllocator; std::size_t = long unsigned
int]: Assertion `stackTop_ + sizeof(T) * count <= stackEnd_' failed.
xargs: ./process2: terminated by signal 6

I got the error just trying to run the code on all the data some hours
ago. When I reran the code on only that repository, it enumerated all
the commits, despite the trees being very very large, and did not
crash at all.

I'm currently running it in valgrind to see if there's anything to catch.

This is my test commandline:

shuf --echo ../repos/*/ --zero-terminated | xargs -0 ./process2 | tee
test.json | cut -c1-160

I use zero termination because some of the repo paths (each one is
named after a code language, and collects many remotes together
written in that language) have spaces in them, which xargs doesn't
handle well.

It would be helpful to get shuf behaving deterministically. A simple
approach would be to cache its output in a file.