Connection actors were still spawned with a bare smarm::spawn(), which
gets the runtime's fixed 64 KiB default stack regardless of smarm
v0.6.0's RFC 019 SpawnOpts/stack_reserve work landing one crate down.
Any handler that leans on app code with real stack needs (DB drivers,
(de)compression, ...) blows the guard page and the connection just
dies with no response - reproduced with a CCC handler that decompresses
gzip on the identity-encoding path.
Add Config::conn_stack_reserve (default DEFAULT_CONN_STACK_RESERVE =
256 KiB) and thread it through listener_loop into a
smarm::spawn_with(SpawnOpts { stack_reserve: Some(_), .. }, ...) call
for every accepted connection. Existing Config { ..Config::new(addr) }
call sites (tests/integration.rs) pick up the new field automatically
via struct-update syntax; no call-site churn beyond that.
Bump to 0.2.2.
Markk116
535f7bcc68