Will Coleda via RT wrote:
Hunt down existing use of 'new_i_s' and either replace or remove the
functionality.
The FixedIntegerArray is an example of where 'new_from_string' is used.
'new_i_s' was experimental and can die. Any PMC's using the behavior can
instead pass a hash argument to 'new'.
Given the usage in compilers/imcc/pbc.c:
_class = interp->vtables[r->pmc_type]->pmc_class;
p = VTABLE_new_from_string(interp, _class, s, PObj_constant_FLAG);
Is there a standard for passing the flags? In the old scheme, we had a specific argument to
pass in the flags... There are a lot of cases where we pass in flags here; Do we now have to
create a Hash PMC in each case? If it's a Hash, then we have to setup a convention to pass
the flags... but doing this just as a convention seems to be ripe for mistakes; Clues welcome
on how to proceed.
It was already just a convention, and various different PMCs used the
arguments passed to 'new_from_string' in different ways. To kick things
off, let's convert the few PMCs that implement the 'new_from_string'
vtable function to handle the behavior in 'init_pmc' using 'flags' and
'string_rep'. The 'flags' key in the init hash can be the integer flags,
and 'string_rep' the string value for 'new_from_string'. Though, I
suspect a bit of intelligence should be applied, with saner names for
some of the PMCs.
Creating a hash is expensive, but then new_from_string was also quite
rare (only implemented on 4 PMCs), so isn't a huge target for optimization.
Allison
