Fix the check conditions in config-msvc.mak and info-msvc.mak so that
the ICU items does indeed get built into the HarfBuzz main DLL, and that
the correct configuration info is displayed.

Also update the checks in detectenv-msvc.mak so that we can detect that
we are using Visual Studio 2017 (although the 2015-built binaries use
the same CRT DLL as the 2017 ones).

See https://bugzilla.mozilla.org/show_bug.cgi?id=1336600

Current fonttools (3.9.1) generate subset-file called font.subset.ttf instead of older font.ttf.subset

Related to https://github.com/behdad/harfbuzz/pull/445

Patch from Phil Race.

For whatever reason the env variables need to be the Windows way or they
end up being empty.

Visual Studio only supported strtof() from Visual Studio 2013 onwards, so
use strtod() instead to do the operation, which should do the same thing,
sans going to a double, not a float.

The sources in src/hb-ucdn and not included correctly into the NMake
Makefiles, as we need their explicit relative location as we pass all the
sources we used into the introspection scanner.  This was not an issue
before as we excluded the UCDN sources in the build when we enabled
introspection (meaning GLib is enabled), but since we default on using
UCDN on all builds unless explicitly disabled, we need to deal with this.

This did not affect builds using UCDN without introspection due to the use
of NMake batch rules.

Fix this by creating a NMake Makefile module on-the-fly with the correct
subdir info, and using that list in there instead.

We now use UCDN by default, so make it so in the build files; however
don't hardcode HAVE_UCDN as one may still opt not to use it (but pass
it in as a CFLAG unless one explicitly disables UCDN by using NO_UCDN=1
on the NMake command line).

Note that we are not blocking builds where UCDN is disabled along with
GLib and ICU, as that will trigger a build error anyways which will tell
the user what needs to be done to remedy this.

This reverts commit b9b005f3.

This introduced invalid access cases. Revert until I fix correctly.

Ouch!

This was broken forever, since days that we did not allow moving
tape backwards. Works now. Reported by Doug Felt.

This has no security implications whatsoever since we always keep
and extra element at the end of buffer, just in case.

Discovered by oss-fuzz
CC https://github.com/behdad/harfbuzz/issues/139
Fixes https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=660

In https://crbug.com/681813 another instance of Padauk was identified
triggering collapsed glyphs. Blacklist this version by patching
hb-ot-layout.cc to print out gdef, gsub, and gpos table length, then
adding those to the list of blacklisted versions.

* Guard against underflow when adjusting length

With the fuzz-testcase in mozilla bug 1295299, we end up with a recursed lookup that removes 3 items, when `match_positions[idx]` is 0, which results in (unsigned) `end` wrapping to a huge value.

Making `end` a signed int is probably the simplest route to a fix.

Fixes https://bugzilla.mozilla.org/show_bug.cgi?id=1295299.

* Add testcase for #421.

* [indic] Add support for Grantha marks that may be used in Tamil to the Indic table.

See https://bugzilla.mozilla.org/show_bug.cgi?id=1331339.

Testcase: U+0BA4,U+0BC6,U+1133c,U+0BAA,U+1133c,U+0BC6,U+1133c

* [indic] Add test for Grantha nukta that is allowed in Tamil by ScriptExtensions.txt

[graphite] Fix shaping with varying font sizes

Hopefully fixes Travis faliure?
https://github.com/behdad/harfbuzz/pull/413#issuecomment-278481882
https://travis-ci.org/khaledhosny/harfbuzz/jobs/199785332#L1065

While it’s fine to call memcmp(x, 0, 0) in practice, the C99 standard
explicitly says that this is not allowed: even if the length is zero,
the pointer arguments must be valid.

http://stackoverflow.com/a/16363034

Coverity ID: 141178
Signed-off-by: NPhilip Withnall <withnall@endlessm.com>

Might help with https://github.com/behdad/harfbuzz/issues/399

Coverity IDs: 141039, 141040, 141041

Coverity ID: 141042

Fixes https://github.com/behdad/harfbuzz/issues/408