-var-create testsuite updates

[Vladimir Prus]

Nick Roberts wrote:

>  > > (Groan) I picked up the tests which were designed to fail but changed
>  > > the
>  > > others by autopilot.  I'll work my way through the changes to work
>  > > out how many others we should have kept and submit a new patch.
>  > 
>  > Thanks a lot!  I didn't want to actually ask you to do this, since it's
>  > a pile of grunt work, but it would certainly be nice to have them back.
>  > 
>  > I only noticed because I had a few patches which changed long to
>  > long( int)? to support another compiler, and they no longer applied.
> 
> Some tests are explicitly testing -var-create e.g mi-var-cmd.exp, others
> might not be but still provide a handy check on the output e.g
> 
> *** gdb.mi/mi-var-child.exp
>   
>   # Test: c_variable-4.2
>   # Desc: create variable "struct_declarations"
>   mi_gdb_test "-var-create struct_declarations * struct_declarations" \
>   "\\^done,name=\"struct_declarations\",numchild=\"11\",type=\"struct
>   _struct_decl\"" \
>  "create local variable struct_declarations"
> 
> I don't really like mi_create_varobj as it removes checks on type, number
> of
> children etc.  

This is intentional. Unless you're testing -var-create itself, you don't
need to check every single output field from -var-create, because:

        1. Tests for -var-create itself should test everything already,
        you don't get anything by extra tests.
        2. Checking output of -var-create everywhere means any
        change in that output requires to update expected output,
        which is pointless.

I think there should be "mi_var_create_varobj_test" that tests entire output.
Tests for -var-create itself should use that.

> How about if I revert all my changes which use 
> mi_create_varobj and just add value=\".*\" to the regexp in each
> mi_gdb_test for -var-create?

Why introducing new "mi_var_create_varobj_test" won't work? gdb testsuite
has too much of literal expected output as it is.

- Volodya