if a DSO_load(NULL,...) operation fails, it will have to call DSO_free() on
the DSO structure it created and that will filter through to this "unload"
call.

If the stack size is "< 1", then the library never actually loaded. To keep
things clean higher up, I'll treat this as a vacuous case without an error.
It makes the error stack easier to follow real world cases, and the error
this ignores was only useful for catching bugs in internal code, not
mismatched calls from applications (which should be handled in the generic
DSO layer).

with RSA_METHOD (the **_get_default_methods do set the default value if
it's not set). However, the code had some duplication and was a bit
conter-intuitive.

initialised, at which point an appropriate default was chosen. This meant a
call to RSA_get_default_method might have returned FALSE.

This change fixes that; now any called to RSA_new(), RSA_new_method(NULL), or
RSA_get_default_method() will ensure that a default is chosen if it wasn't
already.

technique used is far from perfect and alternatives are welcome.
Basically if the translation flag is set, the string is not too
long, and there appears to be no path information in the string,
then it is converted to whatever the standard should be for the
DSO_METHOD in question, eg;
    blah --> libblah.so   on *nix, and
    blah --> blah.dll     on win32.

This change also introduces the DSO_ctrl() function that is used
by the name translation stuff.

the result.

I have retained the old behavior of the CONF_* functions, and have
added a more "object oriented" interface through NCONF_* functions
(New CONF, you see :-)), working the same way as, for example, the
BIO interface.  Really, the CONF_* are rewritten so they use the
NCONF_* functions internally.

In addition to that, I've split the old conf.c code into two files,
conf_def.c and conf_api.c.  conf_def.c contains the default config
object that reads a configuration file the standard OpenSSL way, as
well as configuration file with Win32 registry file syntax (I'm not
sure I got that one right).  conf_api.c provides an API to build other
configuration file readers around (can you see a configuraion file in
XML?  I can :-)).

Finally, I've changed the name conf_lcl.h to conf_def.h, since it's
made specifically for that "class" and none others.

for consistency. Not sure though whether HAVE_DLFCN_H should be
included too? If we go the autoconf route then this probably wouldn't
be included.

* "no-dso" option available in Configure so that all DSO methods will
  return NULL, overriding any support the platform might otherwise
  have built.
* dlfcn_no_h config string now available rather than just dlfcn. This
  is for platforms that have dlfcn.h functions but do not have (or
  need) the dlfcn.h header file.

variety of platforms. A few are missing, and they will be added in
eventually, but as this is new stuff, it was better to not break lots of
platforms in one go that we can't easily test. The changes to "Configure"
should illustrate how to add support to other systems if you feel like
having a go.

NB: I'll add something shortly to allow you to add "dlfcn.h" support on
those platforms that don't have (or need) a dlfcn.h header file. (The
symbol for Configure will probably by "dlfcn_no_h").

Thanks to Richard Levitte, who is responsible for the dso_dl.c support,
understanding the trickier aspects of the build process, and giving great
feedback on everything else.

[Don't use this stuff if you're easily offended by changes to the
interface or behaviour - it's still work in progress.]

PR:

the build process (an upcoming commit no doubt), and is very much *new*
code - what that means is that it compiles ok - usually. It certainly
doesn't mean it runs well or even properly yet. Please don't muck round
with this unless you're looking to help out and hunt bugs. :-)

Currently this code doesn't have any support for controlling the "load"
behaviour (eg. paths, filename translations, etc). That'll be handled
using DSO_ctrl() and various flags, once we work out a sensible set of
flags.

as important to keep a low profile here :-))

a low profile, so we don't get discontinuity in the numbering...)

off.

variables are locked.

it's a good thing if ssleay_rand_status() would do the corresponding
lock and unlock as everyone else...

Update the "randomness" section for the upcoming 0.9.5a release.

Don't force the use of GNU make with Borland C++ Builder.  Contributed by Joon Radley <jradley@vps.co.za>

0.9.5a should not break anything that works in 0.9.5.