Boost program_options zero_tokens broken

[Andrew Hutchings]

Boost's program_options zero_tokens has a know bug (https://svn.boost.org/trac/boost/ticket/1132)

Basically it means for boolean you cannot give any parameters. Which is fine if the boolean default is false, but if the default is true you have no way to set it to false when zero_tokens is used.

This has effectively broken many options.

[Andrew Hutchings (linuxjedi)]

I'll work on an urgent fix for this later today.

[Vijay Samuel (vjsamuel)]

Hi,
  I ve already fixed some boolean options in embedded innodb which had this
problem. I could probably fix all the remaining boolean type options as well
if this bug is assigned to me.

Cheers,
-Vijay

On Sep 12, 2010 2:05 PM, "Andrew Hutchings" <email address hidden>
wrote:

I'll work on an urgent fix for this later today.

** Changed in: drizzle/elliott
      Status: New => Confirmed

--
Boost program_options zero_tokens broken
https://bugs.launchpad.net/bugs/636230
You received th...

[Andrew Hutchings (linuxjedi)]

its ok Vijay, I have already pushed a branch with it which is running through param-build now. The urgency is mostly down to the release due tomorrow.

[Vijay Samuel (vjsamuel)]

Nice fix. Did you go through the entire code base? There are instances in most of the plugins that fail with respect to boolean type options with zero tokens.

[Andrew Hutchings (linuxjedi)]

Explanation of fix:

OK, so, first I used vj's method of zero_tokens()->multitoken(). This allows 0 or 1 token. The problem with this is if you do something like:

drizzle --boolean-option=true test

Where test is the test database. It would think this is a second token to the boolean option or fail.

The fix I have committed instead uses:

implicit_value(true)

So when --boolean-option is given without a =true or =false it will set it to true without the downside of zero_tokens()->multitoken().

[Andrew Hutchings (linuxjedi)]

"It would think this is a second token to the boolean option or fail." <- I of course mean 'and fail'

[Andrew Hutchings (linuxjedi)]

unfortunately new fix is broken with exactly the same problem

[Andrew Hutchings (linuxjedi)]

going to re-write and merge as part of Monty's PO work.