Re: [HACKERS] pgbench: Skipping the creating primary keys afterinitialization

Hello Masahiko-san,

> [...] Personally I prefer "t" for table creation because "c" for create 
> is a generic word. We might want to have another initialization command 
> that creates something.

Ok, good point.


About the patch: applies, compiles, works for me. A few minor comments.

While re-reading the documentation, I think that it should be "Set custom 
initialization steps". It could be "Require ..." when -I implied -i, but 
since -i is still required the sentence does not seem to apply as such.

"Destroying any existing tables: ..." -> "Destroy existing pgbench tables: 
...".

I would suggest to add short expanded explanations in the term definition, 
next to the triggering letter, to underline the mnemonic. Something like:
   c (cleanup)   t (table creation)   g (generate data)   v (vacuum)   p (primary key)   f (foreign key)

Also update the error message in checkCustomCommands to "ctgvpf".

Cleanup should have a message when it is executed. I suggest "cleaning 
up...".

Maybe add a comment in front of the array tables to say that the order is 
important, something like "tables in reverse foreign key dependencies 
order"?

case 'I': ISTM that initialize_cmds is necessarily already allocated, thus 
I would not bother to test before pg_free.

-- 
Fabien.