GitLab

This reverts commit 77dac848.

I thought my use of 'select' was causing bugs, it wasn't. The problem I
found was in logstash's lumberjack input.

Conflicts:
	proto.c

Conflicts:
	Makefile

polling/select; seems more reliable? I've had selects timeout when I'm
certain there's data to read.

  failure/timeouts) any pending messages will be discarded.

  send/recv failure.

Don't make assumptions about make

use MAKE variable, set it to a sensible default.

- fix off-by-one on offset counts
- support stdin if the path  is simply "-"

  this will let harvesters continue to read 'old files' until they stop
  being written to, which will solve some race conditions and especially
  solve 'bad rotations' where logrotate renames, but the application
  writing to the log file isn't notified and keeps writing to the old
  one.

  duration-since-last-flush-or-start; I think it makes the code more
  readable to say "is the current time after next_flush_time?"

  Same functional goals, different implementation and wire protocol.

- remove newlines from log calls

- remove some newlines in logs

This reverts commit e348e113.