o implement add1() using 32 bit ints, this makes _krb5_n_fold()
about 5% faster on an amd64 platform. 64 bit ints yield a
further improvement but we would need to test the platform
to see if they are natively supported. This should yield
better performance improvements on big endian machines as
we have to byte swap on little endian boxen.
o fix two cases where a malloc(3)d pointer may be dereferenced
before we test that it is not NULL.
All in lib/krb5/n-fold.c:
1. eliminate malloc/free from rr13() because it is always a
buffer of the same size called in a tight loop.
2. eliminate memcpy(3) from rr13() by bouncing back and forth
between two buffers buf1, buf2 instead of performing the
calculation into a tmp buffer and memcpy(3)ing the result
back into buf.
3. eliminate code cases from rr13() that I can visually determine
will never occur but I'm guessing that the compiler can't, i.e.
i. now that we're no longer using malloc(3), rr13()
cannot fail, so make it void and avoid the if in
the calling routine checking its error code. In
case you ask, yes, this made the tests run a little
faster,
ii. rr13() has code for being passed a number of bits
not divisble by 8 but _krb5_n_fold() only passes
an int * 8. So, we eliminate this conditional and
the associated code.
4. we make rr13() take 2 destination buffers and copy the results
into both of them, we use this to eliminate another memcpy(3)
from the calling routine. This appears to make it a bit faster
as well.
In both hx509_cert_init() and hx509_cert_init_data(), there is an
output parameter for the error code but there are cases where the
error is used as a return value instead of the specified hx509_cert.
We fix these issues. We also check if error is non-NULL and only
set the error in this case, allowing the functions to be called
with error == NULL without segfault.
When building with OpenSSL at a custom prefix, some test cases will fail
to compile due to missing include path compiler options. This patch adds
them, as well as defining CPPFLAGS and LDADD for test_expr.
Signed-off-by: Love Hornquist Astrand <lha@h5l.org>
To stop the errors when building concurrently, we make a number of
changes:
1. stop including generated files in *_SOURCES,
2. make *-protos.h and *-private.h depend on the *_SOURCES,
3. make all objects depend on *-{protos,private}.h,
4. in a few places change dir/header.h to $(srcdir)/dir/header.h,
This appears to work for me with make -j16 on a 4-way box.
OpenSSL_add_all_algorithms() should only be run once per application
or it will cause data structures to expand. It's not a classic
memory leak as all of the memory will be free(3)d when EVP_cleanup()
is called but as we are a library we cannot call this. We provide
a short term fix here which is using heim_base_once_f() to ensure
that we only call it once.
But the long term fix should be to stop using OpenSSL_add_all_algorithms()
entirely because it both has side effects outside our library and
the caller may destroy our OpenSSL global variables by calling
EVP_cleanup() on his own. It is suboptimal to have potential
interactions between our library and other code in this way.
