Re: [Monotone-devel] bug in option.cc getopt

[Stephen Leake]

Stephen Leake <address@hidden> writes:

> revision 774b2d041536fd47832d7f844688061791847295 Timothy Brownawell
> <address@hidden> 2010-08-11T13:07:52
>
> introduced a bug in option.cc getopt; it can't handle an empty string
> for 'name.
>
> This only happens in MinGW, in tests/automate_identify.
>
> I'm not clear yet why it is getting an empty string.
>
> I also don't see what the proper result should be.

Figured it out. The test is this:

-- check what happens if our filename is the stdin marker (hint: it should stop)
check(mtn("automate", "identify", "-"), 1, false, false)

so the option name is the empty string. 

This line:

  string err = (F("option '%s' has multiple ambiguous expansions:")
                % name).str();

throws an exception:

mtn: fatal: boost::io::bad_format_string: boost::bad_format_string: 
format-string is ill-formed

Apparently boost 1.34.1 (on MinGW) doesn't accept an empty string in
this case, while boost 1.42 (on Debian) does? I didn't try to step into
the boost code.


mtn 0.48 says this for 'mtn automate identify -':

mtn: option error: unknown option ''

So I just added a check for an empty name:

  if (name.size() == 0)
    throw unknown_option(name);

Now all tests are passing on MinGW

Hmm. That may conflict with this change:

   075540bcf4d90afcc170e6f19b1e930501206fd6
   2010-08-22T12:13:12  address@hidden
     * option.cc (getopt): when a user tries to complete "-", we list
       all available options, but of course all short-named options match
       then as well, so we need to be more careful and cannot simply set
       "--" in front of every option name

-- 
-- Stephe