----------------------------------------
> To: d...@perl.it; libwin32@perl.org; bul...@hotmail.com
> Subject: Re: 3 Win32::API bugs
> Date: Thu, 10 May 2012 10:53:45 +0200
> From: cos...@streppone.it
> > Or should I forget about callback and try and fix the 3 other bugs (func
> > not found dll handle leak, no unsigned return values, no shorts) which
> > are easier targets to kill than Callback and then try Callback if I have
> > the time?
>
> I would tend to agree with this.
>
> --
> Cosimo
I still don't know what letter to use for shorts. $ has no lower case version
for signed short. H is too closed to hex numbers and pack's existing H unpack
feature letter (H="Half"), someone might want to put in a 'H' type code one day
to get strings back in ASCII HEX from Win32::API rather than binary gibberish.
T (T="shorT", also "strucT", but sturct is S already), isn't used by pack, W
(W="Word") is used for BER numbers by pack, and I've never seen those anywhere,
so W or T I think are safe choices.
The unsigned features I'm not sure how to implement, I dont think it makes any
sense for in parameters, since on a C level, a function can't tell if it got a
signed or unsigned. The only kind of pointer fill supported by Win32::API is
fill a string pointer, so for a pointer to int or pointer to handle you still
have to unpack it yourself (I think, I've only used Win32::API for a few days).
For the return parameter currently I thinking of a bit flag field for T_NUMBER
and T_INTEGER that gets masked off for the switch, and then the flag is check
on whether to make an IV or a UV. If the return unsigned bug is fixed, small
but I think irrelevant risk of breakage complaints.
Implementing the short feature is the most difficult I think. I made a test for
the Dll handle leak and fixed that already. *I think* I fixed Callback. A
couple problems with Callback was the ILT jump table, all the functions are 1
opcode long with the ILT jump table, shannons law has been broken by MS. Fixed
that. The biggest problem I think was, assuming that the compiler will always
copy the fake global vars which are replaced with EBP stack reads, into EDX.
Any, slight, tiny, change of the compiler by MS can, and probably has in the
past, broken it. I rewrote it that any mov from fake global pointer to any
register will be caught and replaced. Also, under -Od, VC didn't use the
traditional mov opcode, it used one specific for EAX (ask intel about that
one), fixed that. I also reworked the optimization pragma to be more specific
what ::Callback wants (dont toss the 0XC0DEs) to CallbackTemplate, and
CallbackTemplate can be compiled with O1 O2 and Od now.
In 00_API.t, there are a bunch of cdecl tests that are commented out. They
refer to functions like c_call_sum_int c_call_sum_int_dbl? Does anyone have
these, they aren't in API_test.cpp.
