The multi-header-set work adds explicit notification for incoming HTTP/3 header sets via on_hset_in, allowing applications to call lsquic_stream_get_hset() when a header set arrives.
One related behavior question is whether an unclaimed HTTP/3 header set should block DATA delivery.
Current behavior to preserve for now:
read_data_frames() may parse a HEADERS frame and continue processing subsequent DATA in the same read path.- If the application does not claim a non-HTTP/1.x header set, existing
stream_readf() behavior remains the primary guard when the queued hset is observed at the stream read layer.
- The multi-header-set implementation should not add a new filter-level rule that stops DATA solely because
stream->uh contains an unclaimed HTTP/3 hset.
Question:
Should the public API define that DATA must not be delivered while an earlier HTTP/3 header set remains unclaimed, or should applications be responsible for claiming hsets promptly when on_hset_in fires?
If we choose the stricter behavior, it should be added deliberately with tests and API documentation because it can change existing header-bypass users that currently do not immediately drain 1xx/early-hints header sets.
The multi-header-set work adds explicit notification for incoming HTTP/3 header sets via
on_hset_in, allowing applications to calllsquic_stream_get_hset()when a header set arrives.One related behavior question is whether an unclaimed HTTP/3 header set should block DATA delivery.
Current behavior to preserve for now:
read_data_frames()may parse a HEADERS frame and continue processing subsequent DATA in the same read path.stream_readf()behavior remains the primary guard when the queued hset is observed at the stream read layer.stream->uhcontains an unclaimed HTTP/3 hset.Question:
Should the public API define that DATA must not be delivered while an earlier HTTP/3 header set remains unclaimed, or should applications be responsible for claiming hsets promptly when
on_hset_infires?If we choose the stricter behavior, it should be added deliberately with tests and API documentation because it can change existing header-bypass users that currently do not immediately drain 1xx/early-hints header sets.