This is the only regex matching the symbols that are built.

This includes the removal of INSTALL and NEWS files that were forced
by the non-foreign automake flavour.

Rely on pkg-config only, as that's what we're going to do for
libdvdread next.

Rely solely on pkg-config to find the library, as that makes it much simpler.

This new buildsystem is non-recursive (so it uses higher parallel
build), but also drops libdvdnavmini (which is just an alias for
libdvdnav itself), and a few more useless sophistications.

Fabian Keil authored 12 years ago and Jean-Baptiste Kempf committed 12 years ago

Fixes segfaults with a rip of Ghost Protocol.

Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Fabian Keil authored 12 years ago and Jean-Baptiste Kempf committed 12 years ago

Fixes crashes with non-compliant DVDs after applying
the duplicate detection patch for libdvdread.

It might make more sense to do those checks in libdvdread
instead and zero out structures that don't check out.

Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Erik Hovland authored 12 years ago and Jean-Baptiste Kempf committed 12 years ago

There was a time where dvdnav examples failed to build on mingw.  Rather
than fix them, we just disabled examples since we had no need for them.
Although the examples seem to build fine now, this option may be useful
in general.

Patch provided by John Stebbins (stebbins AT jetheaddev DOT come),
thanks!

Mirakels authored 12 years ago and Jean-Baptiste Kempf committed 12 years ago

As offset was an uint64_t offset could not be negative. Seeking
backwards is then only possible by get_position(), substract an
offset and do a SEEK_SET.
Now with a signed offset SEEK_CUR can be used to go backwards
without getting the current position first.

Mirakels authored 12 years ago and Jean-Baptiste Kempf committed 12 years ago

This also fixes wrong positioning in dvdnav_sector_search() for
multi-angle DVDS.
BLOCK_TYPE_ANGLE_BLOCK cells seem to overlap sectors in the previous
cell hence they should be skipped when adding cell lengths up to get
to the correct sector.

dvdnav_sector_search() already checks for BLOCK_TYPE_ANGLE_BLOCK
cells in calculating the new position but dvdnav_get_position()
did not. As dvdnav_sector_search() starts by calling
dvdnav_get_position() the end result for the new position could
be wrong.

Erik Hovland authored 12 years ago and Jean-Baptiste Kempf committed 12 years ago

If the cell number is 0 dvdnav will crash w/ a SIGBUS. Check the cell
number before this can happen and just continue.

Erik Hovland authored 12 years ago and Jean-Baptiste Kempf committed 12 years ago

If the start byte is in the wrong place then dvdnav can crash because it
is out of bounds.

Erik Hovland authored 12 years ago and Jean-Baptiste Kempf committed 12 years ago

The pgc pointer might be null when it is retrieved from the IFO. This
can cause a segfault when pgc is dereferenced a few lines later.

Erik Hovland authored 12 years ago and Jean-Baptiste Kempf committed 12 years ago

If a VOB is not open we will dereference a null pointer in
DVDReadBlocks. That will produce a segfault. If we check for an open VOB
when we check to see if we need to change the currently open VOB we can
ensure that a VOB is open when DVDReadBlocks dereferences it.

This patch supplied by John Stebbins <stebbins AT jetheaddev DOT com>.

Erik Hovland authored 12 years ago and Jean-Baptiste Kempf committed 12 years ago

This adds a function to create a duplicate of a dvdnav handle and it's
state. This allows you to recursively move through the handle. HandBrake
uses these functions to do recursive automated search of the menus for
something that looks like the main feature.

Erik Hovland authored 12 years ago and Jean-Baptiste Kempf committed 12 years ago

When vm_start is called it should check to see if the vm is stopped. If
it is, reset it and marked the vm as started (set vm->stopped to zero).

Erik Hovland authored 12 years ago and Jean-Baptiste Kempf committed 12 years ago

Replace vm_stop() w/ vm_close() where necessary.

Erik Hovland authored 12 years ago and Jean-Baptiste Kempf committed 12 years ago

It is convenient to have vm_stop() have different processing when called
by other parts of the library. When vm_stop is changed in this way, it
means that the vm needs to be able to do the same processing internally.
The introduction of the static function vm_close() provides that
processing.

Erik Hovland authored 12 years ago and Jean-Baptiste Kempf committed 12 years ago

libdvdnav's configure2 lacked endian detection. This problem was pointed
out by Ganael Laplanche <ganael DOT laplanche AT martymac DOT org>
through a patch that he submitted to fix this issue (and others) for the
BSDs. Thanks!

Erik Hovland authored 13 years ago and Jean-Baptiste Kempf committed 12 years ago

It's easier to manage pkg-config when cross-compiling, and libdvdread
installs .pc files, so check those first.

Patch provided by Mike Frysinger <vapier AT gentoo DOT org>, Thanks!

Erik Hovland authored 13 years ago and Jean-Baptiste Kempf committed 12 years ago

This patch fixes the following error messages one might get when
building libdvdnav on W32 using MinGW:

libtool: link:  gcc -shared .libs/libdvdnavmini-4.dll.def
.libs/dvdnav.o .libs/read_cache.o .libs/navigation.o .libs/highlight.o
.libs/searching.o .libs/settings.o .libs/remap.o  -Wl,--whole-archive
../src/vm/.libs/libdvdvm.a -Wl,--no-whole-archive   -O3 -march=i686
- -o .libs/libdvdnavmini-4.dll -Wl,--enable-auto-image-base -Xlinker
- --out-implib -Xlinker .libs/libdvdnavmini.dll.a
Creating library file:
.libs/libdvdnavmini.dll.a.libs/dvdnav.o:dvdnav.c:(.text+0x27):
undefined reference to `DVDCloseFile'

Patch provided by LRN <lrn1986 AT gmail DOT com>, Thanks!

Erik Hovland authored 13 years ago and Jean-Baptiste Kempf committed 12 years ago

cellnr is used to index into cell_playback after subtracting
one from it. If cellnr is 0, then it will index -1 in cell_playback
which will seek out of boundary of cell_playback. This manifested into a
segfault for some users as reported by this launchpad bug:
https://bugs.launchpad.net/ubuntu/+source/libdvdnav/+bug/934471

By checking cellnr and skipping the indexing if cellnr is equal to zero
then we avoid the segfault. There might be a bigger issue w/ regard to
retrieving a value of zero for cell new row, but this fix works for the
reporter.

Thanks goes to Sylvain Henry (hsyl20 AT gmail DOT com for both
reporting the bug to launchpad and submitting a potential patch (even
though we went w/ a different fix).

Erik Hovland authored 13 years ago and Jean-Baptiste Kempf committed 12 years ago

When converting to ticks it is possible for the first value to overflow.
We can prevent this overflow if we cast the first value to int64_t
before multiplying. This will cause the compiler to use int64_t as the
base type for the operations.

Erik Hovland authored 13 years ago and Jean-Baptiste Kempf committed 12 years ago

Now that dvd_read_name returns a status code we can check it before
calling remap_loadmap.

Erik Hovland authored 13 years ago and Jean-Baptiste Kempf committed 12 years ago

dvd_read_name has several places where it can fail. Since it doesn't
return a value it can't communicate that to any callers. This commit
changes the function so it does all of the same work but it checks every
possible failure point.

Erik Hovland authored 13 years ago and Jean-Baptiste Kempf committed 12 years ago

Roger Pack reported that when playing the DVD "Tangled"
w/ mplayer and recent libdvdnav, mplayer crashes w/ the error message:
Assertion failed: (vm->state).pgc->cell_playback[(vm->state).cellN - 1].block_mode != 0, file libdvdnav/vm/vm.c, line 1141

This line should only be triggered when a developer wants a "strict"
build of the library. Most likely using the very general term
STRICT to #ifdef this clause is what is causing this issue. This
patch adds a DVDNAV_ in front of the strict to prevent the
macro from being enabled accidentally.

Erik Hovland authored 13 years ago and Jean-Baptiste Kempf committed 12 years ago

The CLUT that is copied in from struct variable palette is an
array of 16 values. So the &(palette) is unnecessary. The size
argument was also changed to use the palette array instead of
the type (just in case palette ever gets changed).

Thanks goes to Lee Essen who originally submitted the patch
way back in 2008 and to Roger Pack who brought the patch to
our attention recently.

Erik Hovland authored 13 years ago and Jean-Baptiste Kempf committed 12 years ago

It is possible (although unlikely) that the while loop will
never execute and next_vobu will be uninitialized at this point.