tree 76bfc9865bc9ee26e4f73999bcecf424dd649598
parent 41de629c1fe88f9ea95d0400af6896f3c84a0dff
author Paul Jakma <paul.jakma@hpe.com> 1466088806 +0100
committer Paul Jakma <paul.jakma@hpe.com> 1473759396 +0100

lib: keep hash of node's commands to detect duplicate installs

* command.h: (struct cmd_node) Add a hash, so duplicate installs of
  a cmd_element to a command node can be detected.  To help catch
  strays from the VIEW/ENABLE node consolidation particularly
  (installs to VIEW automatically install to ENABLE too now).
* command.c: (cmd_hash_{key,cmp}) helpers for the hash - just directly
  on the pointer value is sufficient to catch the main problem.
  (install_node) setup the hash for the command node.
  (install_element) check for duplicate installs.
  The assert on the cmd_parse_format seems misplaced.
  (install_default_basic) separate the basic, VIEW, node default commands
  to here.
  (cmd_init) get rid of dupes, given consolidation.
  (cmd_terminate) clean up the node command hash.

Not done: The (struct cmd_node)'s vector could be replaced with the
cmd hash, however much of the command parser depends heavily on the
vector and it's a lot of work to change.  A vector_lookup_value could
also work, particularly if vector could be backed by a hash.

The duplicate check could be disabled in releases - but useful in
development.  It's a little extra overhead at startup.  The command
initialisation overhead is already something that bites in
micro-benchmarks - makes it easy for other implementations to show
how much faster they are with benchmarks where other load is low
enough that startup time is a factor.