building libgcrypt with --host x86_64-w64-mingw32

[Werner Koch]

On Wed, 17 Feb 2016 23:48, dkg at fifthhorseman.net said:

> this is troubling -- can you explain more about why you think the 64-bit
> version might get worse entropy inputs than the 32-bit version?

I am simply not sure and I have not done ay test at all.  An audit of
the code with an emphasis on the 64 bit changes in Windows would be the
Right Thing to do.  The original idea was to use a new rndw64.c with all
the old Windows pre-XP code removed but Jussi meanwhile fixed a couple
of things in rndw32.c so it might now be easier to closely check the
code and compare it to Peter' s latest version of Cryptlib (which was
the original source of that code).

> right, but in that case developers probably do need a reliable 64-bit
> Windows DLL.  If we can sort out the RNG business, that would be good.
> Should i open a ticket at https://bugs.gnupg.org/ to record the concern?

Right now we have no need for libgcrypt in 64 bit DLLs.  The only 64 bit
DLL we use is GpgEX which does not do any crypt but delegates that to
GPA or Kleopatra.  However, Andre is working on a 64 bit GpgOL and there
we need a bit of crypto.  So <https://bugs.gnupg.org/gnupg/issue1810>
needs to be done by then.


Shalom-Salam,

   Werner

-- 
Die Gedanken sind frei.  Ausnahmen regelt ein Bundesgesetz.