GitLab

When we decode using libopus's fixed-point APIs, libopus internally
 applies soft clipping prevention.
When we decode using libopus's floating-point APIs, this behavior
 is disabled.
If we're ultimately planning to output the data to the user in
 fixed-point, we need to apply the clipping prevention ourselves.

The caller can switch between the fixed/float APIs and the
 stereo/multichannel APIs on every call.
Detect this and reset the dither state to avoid potentially
 injecting noise from a very out-of-date state when switching from
 fixed to float back to fixed, or from the wrong channels when
 switching between stereo and multichannel.
Normal applications won't do this stuff, but we should be
 well-behaved if they do.

For multichannel files, this potentially caused writes beyond the
 end of the output buffer provided by the user.

Reduces cruft on 'make clean'. I've rearranged some of the
non-functional self-test targets, but not really fixed them.

The old ones gave a different warning on Apple's clang 4.2.

This is required by the xmingw build.

Just normalizing coding style.

OpenSSL on Windows does not pull certificates from any well-known
 location (in fact most binaries continue to use the default Unix
 path, which usually doesn't even exist).
We could ship our own set of certificates (e.g., cloned from the
 Mozilla root list), but I don't want to be responsible for
 releasing libopusfile updates when things like DigiNotar
 fiasco [1] happen.
That approach also means that we would need to load, parse, and
 keep a copy of every certificate in the system for every SSL
 session.

OpenSSL has had patches sitting in their bugtracker which load
 certificates from the Crypto API's system certificate store.
However, those patches have been sitting around for several years,
 so movement on that front in the near future seems unlikely.
We don't care about using OpenSSL's builtin CAPI engine, though, so
 we can do the same thing with less than 200 lines of code.
This puts the maintenance burden on Windows Update, which will be
 far more timely and effective than getting people to upgrade
 libopusfile, and gets us on-demand loading of just the
 certificates we need.

[1] <https://bugzilla.mozilla.org/show_bug.cgi?id=682927>

Apparently Vista includes more things in its Winsock implementation
 and errno.h than earlier versions of Windows.

This keeps differences which can be cleanly abstracted away clean
 (closesocket, ioctlsocket, getsockopt, setsockopt), and makes
 differences which cannot be cleanly abstracted explicit (SOCKET,
 INVALID_SOCKET, WSAGetLastError/WSASetLastError).
It also gets rid of wsockwrapper.[ch], since it contained just a
 single function.

This can successfully pass the seeking_example tests on
 big.chained.blob over https when built with i686-w64-mingw32 and
 run under wine.
It does not solve the certificate distribution problems with using
 OpenSSL on a real Windows system.

It turns out i686-pc-mingw32 does define these functions, and they
 are always available (unlike _fseeki64/_ftelli64).
This means we can build and link without requiring
 i686-w64-mingw32.
The resulting binary still doesn't run in wine for me, but that may
 be a personal problem.

This allows an i686-pc-mingw32 build to use getaddrinfo/
 freeaddrinfo correctly.
It's not needed for an i686-w64-mingw32 build, but it doesn't hurt
 to leave it in here.
i686-pc-mingw32 still won't work correctly, because it does not
 support fseeko/ftello, nor (unless using MSVCRT 8.0 or later)
 _fseeki64/_ftelli64.
We could define __MSVCRT_VERSION__ to 0x800 to get the latter, but
 this would require different MSVCRT versions than libogg and
 libopus are built with by default, which makes a bit of a mess.

Properly check for HTTP support and handle the case where it's
 disabled.
Also fixes the include paths broken by 3e7f0ddc.

The former is too generic.

Some tweak might be still needed to take care of OPENSSL_AppLink
to get https support working. In win32, user application of openssl
is required to include openssl/applink.c or something, when openssl
is compiled with OPENSSL_USE_APPLINK.

I don't know how it should be taken care of, from the library point of
view (it must be done by user of libopusfile, since openssl always
searches that function in executable module).

Posted to the hydrogenaudio format 2012 November 19.
http://www.hydrogenaudio.org/forums/index.php?s=&showtopic=97856&view=findpost&p=814582

Fixes Coverity #980717, #980718, #980719.

This lets us seek forward by small amounts (currently less than
 90 ms) by decoding forward instead of actually seeking.
This is often a good idea, since we would have to decode at least
 80 ms of pre-roll anyway.
This optimization also handles the case of seeking to what is
 already the current position cheaply.

This became relatively easy after we dropped op_pcm_seek_page()
 from the public API.
However, because others may look to libopusfile's seeking code as a
 model, we've added an OP_SMALL_FOOTPRINT #define to cordon off
 some of these complex sections of code that are deeply specific to
 libopusfile's design, ancillary to the main seeking algorithm,
 and relatively unimportant to overall seeking performance.

We should copy the minimum of the buffer size and the remaining
 bytes, not the maximum.
Fixes #1927.

This makes it more likely the WAV file header matches the amount of
 data actually written if we run out of space.
This is easier than making the accounting better.

Thanks to e98 for the report.

Add the step to the release procedure.

The assert might also be helpful for static analysis.

These are useful alternatives to the autoconf build,
since we're maintaining them we should include them
in release packages.

This was needed in the opus Makefile.am to pass the results
to submakefiles. Since we build everything from the top-level
Makefile, we don't need this for opusfile.

I don't think it actually breaks anything if we don't (seeking will
 fail, but it will fail cleanly).
However, it _is_ an indication that the file data changed out from
 under us (or of another library bug), so we should fail fast.

Using op_get_prev_page_serial() meant if there were valid pages
 from another stream, we would re-scan them all repeatedly until we
 managed to back up, a page at a time, to a chunk with a page from
 the preferred stream with a valid granule position.
Breaking this case into its own function means we can guarantee we
 only scan each page once, and that we can stop as soon as we hit a
 page from a different link (in the unlikely event that the stream
 has no pages with valid timestamps).

Be more specific about which RFC we're quoting in the SSL hostname
 verification.

Adding AM_MAINTAINER_MODE([enable]) preserves the current behaviour of
rebuilding the autotools support by default if anything changed to make
it out of date -- which is desirable for people building from git to
avoid "Why is the build broken? Oh I forgot to rerun autogen.sh", which
is an all too frequent FAQ.

But it also allows it to be disabled, which is mostly useful for things
like distro package builds where the files should be preserved in their
original form, and configure shouldn't be run twice on the first build
if the timestamps were not strictly preserved for these files.
Signed-off-by: Ralph Giles <giles@mozilla.com>