Fix msan issue

The fact that HB_AUTO_RETURN will return rvalue-references for rvalues
is very disturbing.

Even apart from that, I'm totally lost re any hb_move needs or
hb_forward'ing to functions/templates where the type is fixed by
explicitly specifying template parameters.

==1==ERROR: AddressSanitizer: stack-use-after-return on address 0x7f6ad65e51e0 at pc 0x0000005da240 bp 0x7ffc104ab670 sp 0x7ffc104ab668
READ of size 4 at 0x7f6ad65e51e0 thread T0
SCARINESS: 55 (4-byte-read-stack-use-after-return)
     #0 0x5da23f in bool OT::Coverage::serialize<hb_map_iter_t<hb_map_iter_t<hb_filter_iter_t<OT::Coverage::iter_t, hb_set_t const&, $_7&, (void*)0>, OT::SingleSubstFormat1::subset(hb_subset_context_t*) const::'lambda'(unsigned int), (hb_function_sortedness_t)1, (void*)0>, $_20&, (hb_function_sortedness_t)1, (void*)0>, (void*)0>(hb_serialize_context_t*, hb_map_iter_t<hb_map_iter_t<hb_filter_iter_t<OT::Coverage::iter_t, hb_set_t const&, $_7&, (void*)0>, OT::SingleSubstFormat1::subset(hb_subset_context_t*) const::'lambda'(unsigned int), (hb_function_sortedness_t)1, (void*)0>, $_20&, (hb_function_sortedness_t)1, (void*)0>) harfbuzz/src/hb-ot-layout-common.hh:1055:16
     #1 0x5d88f9 in bool OT::SingleSubstFormat1::serialize<hb_map_iter_t<hb_map_iter_t<hb_filter_iter_t<OT::Coverage::iter_t, hb_set_t const&, $_7&, (void*)0>, OT::SingleSubstFormat1::subset(hb_subset_context_t*) const::'lambda'(unsigned int), (hb_function_sortedness_t)1, (void*)0>, $_20&, (hb_function_sortedness_t)1, (void*)0>, (void*)0>(hb_serialize_context_t*, hb_map_iter_t<hb_map_iter_t<hb_filter_iter_t<OT::Coverage::iter_t, hb_set_t const&, $_7&, (void*)0>, OT::SingleSubstFormat1::subset(hb_subset_context_t*) const::'lambda'(unsigned int), (hb_function_sortedness_t)1, (void*)0>, $_20&, (hb_function_sortedness_t)1, (void*)0>, unsigned int) harfbuzz/src/hb-ot-layout-gsub-table.hh:98:9

Whitespace

[subset] Remove extra iteration

[subset] Truncate empty gids at the end in retain-gids mode.

Use hb_map(hb_add(this)) to dereference OffsetTo<>'s

[algs] Rename hb_bind to hb_partial

Since our API is the invers of what std::bind is, and closer to Python
functools.partial().

[algs] Change hb_bind parameter number to be from one

To match std:;bind, even though our interfaces are very different.

Add back symmetric OffsetTo<>::friend operator+

Finally seems to be working now.

[algs] Partialize all operators

[algs] Rewrite bind API

And add a partialization API use example to hb_add()

[algs] Add hb_bind0 and hb_bind1

Fix clang build

Ugh.

In file included from hb-ot-face.cc:41:
./hb-ot-layout-gsub-table.hh:293:7: error: template parameter redefines default argument
         hb_requires (hb_is_sorted_source_of (Iterator,
         ^
./hb-meta.hh:59:27: note: expanded from macro 'hb_requires'
 define hb_requires(Cond) hb_enable_if((Cond))
                          ^
./hb-meta.hh:57:67: note: expanded from macro 'hb_enable_if'
 define hb_enable_if(Cond) typename hb_enable_if<(Cond)>::type* = nullptr
                                                                  ^
./hb-ot-layout-gsub-table.hh:40:5: note: previous default template argument defined here
    hb_requires (hb_is_sorted_source_of (Iterator,
    ^
./hb-meta.hh:59:27: note: expanded from macro 'hb_requires'
 define hb_requires(Cond) hb_enable_if((Cond))
                          ^
./hb-meta.hh:57:67: note: expanded from macro 'hb_enable_if'
 define hb_enable_if(Cond) typename hb_enable_if<(Cond)>::type* = nullptr
                                                                  ^

Fix clang build

In file included from hb-ot-face.cc:41:
./hb-ot-layout-gsub-table.hh:293:7: error: template parameter redefines default argument
         hb_requires (hb_is_sorted_source_of (Iterator,
         ^
./hb-meta.hh:59:27: note: expanded from macro 'hb_requires'
 define hb_requires(Cond) hb_enable_if((Cond))
                          ^
./hb-meta.hh:57:67: note: expanded from macro 'hb_enable_if'
 define hb_enable_if(Cond) typename hb_enable_if<(Cond)>::type* = nullptr
                                                                  ^
./hb-ot-layout-gsub-table.hh:40:5: note: previous default template argument defined here
    hb_requires (hb_is_sorted_source_of (Iterator,
    ^
./hb-meta.hh:59:27: note: expanded from macro 'hb_requires'
 define hb_requires(Cond) hb_enable_if((Cond))
                          ^
./hb-meta.hh:57:67: note: expanded from macro 'hb_enable_if'
 define hb_enable_if(Cond) typename hb_enable_if<(Cond)>::type* = nullptr
                                                                  ^
1 error generated.

[algs] Remove pair copy constructor

Use default.

Whitespace

[subset] Switch SingleSubst to use iterators in serialize.

Reduce captures of lambdas

Fix tests

Oops.

Remove variadic form of hb_min/hb_max

Unused, and why here and not in other functions...

Add arithmetic operators

Add hb_bitwise_* ops

[array] Adjust operator!=

See comments.

More adjustment to OffsetTo<>::friend opeator+

Let's see if I break any bots.  But yeah, it wasn't accepting a
non-const pointer.  It just happens that we don't use that in the
code it seems.

[iter] Simplify operator!= of iterator filters

Both to save ops, and also because lambdas don't implement operator!=,
so this was failing in range-based for loop if a lambda was passed to
hb_map() or hb_filter().  Just check end-condition assuming that we
are comparing to .end() or iterators that are otherwise derived from
current iterator.  Ie. don't compare things that are expected to be
in common.

Add all pair_t comparison operators

[iter] Use default operators instead of redefining empty ones

Convert Consonant_Initial_Postfixed to CONS_MED

Fix record-test.sh on machines without sha1sum

[ci][test] Ignore other gcov symbols also

To fix https://travis-ci.org/harfbuzz/harfbuzz/jobs/532693197

[ci][travis] Update its distribution

It may break things, lets see

Another try

Completely revert the thing back

Another try at fix

Fails locally.  Trying to understand.  Sigh

Ugh.  How was the Travis bot happy before, but isn't now?! :(

Move OffsetTo operator+ back out of class

Apparently there's different overload resolution rules that apply, at
least with some (older?) version of gcc.

hb-ot-name-table.hh: In member function ‘void OT::name::accelerator_t::init(hb_face_t*)’:
hb-ot-name-table.hh:244:62: error: ambiguous overload for ‘operator+’ (operand types are ‘hb_blob_ptr_t<OT::name>’ and ‘OT::NNOffsetTo<OT::UnsizedArrayOf<OT::IntType<unsigned char, 1u> > > {aka const OT::OffsetTo<OT::UnsizedArrayOf<OT::IntType<unsigned char, 1u> >, OT::IntType<short unsigned int, 2u>, false>}’)
       this->pool = (const char *) (const void *) (this->table+this->table->stringOffset);
                                                              ^
hb-ot-name-table.hh:244:62: note: candidates are:
hb-ot-name-table.hh:244:62: note: operator+(const C*, long int) <built-in>
hb-ot-name-table.hh:244:62: note: operator+(const char*, long int) <built-in>

Revert "Revert symmetric OffsetTo overloads"

This reverts commit 01912efb74fc554a81c8cfe572145ce45b8fa58b.

Actually this didn't break things.  Fixing

Revert symmetric OffsetTo overloads

Reverts 57f65ae9355004044325dd6441cde761bca5e0a3

Caused ambiguous-overload on some gcc...

Use implicit lambda return type

Add symmetric friend operator+ for OffsetTo

Move OffsetTo<> deref operators in-class as friends

Whitespace

[iter] Remove sort categorization

See comments.

Start of gen-ucd.py, to replace UCDN

[subset] retian nameids from STAT and fvar tables

Hopefully last warning fix

[ci] add HB_TINY to asmjs builder

Fix -Wunused-function on HB_NO_SHAPE_AAT

We should add a bot for it
Part of #1652

Fix building with HB_NO_SUBSET_LAYOUT

Fixes an unused function warning when building with HB_NO_SUBSET_LAYOUT
as part of the Chrome build.

[ucdn] Fix Hangul composition

https://github.com/grigorig/ucdn/issues/23

Fix more warnings

[config] Add NDEBUG and HB_NDEBUG

Part of https://github.com/harfbuzz/harfbuzz/issues/1652

One more warning fix

No idea where these appear from...

[icu] Comment

Fix moreeeeeeeeeee

Fix more semi-colon issues

[iter] Fix zip iterator sortedness classification logic

Fix more excess semi-colon errors

[iter] Track strictly-sorted iterators

This make output of hb_enumerate() sorted regardless of input iterator.

[iter] Minor

[subset] Fix shadowed 'groups' param in cmap.

Adjust uniscribe_bug_compatible mode

More correct behavior.  We were commenting out some legit conditions
before.

Part of https://github.com/harfbuzz/harfbuzz/issues/1652

[config] Better compile away morx/kerx/trak

Part of https://github.com/harfbuzz/harfbuzz/issues/1652

[indic] Don't constrain how many C, M, ... occur

Fixes https://github.com/harfbuzz/harfbuzz/issues/1709

Part of https://github.com/harfbuzz/harfbuzz/issues/1652

[test] Don't call deprecated API

Another extra semi-colon

Revert "[ragel] Regenerate ragel-generated files using ragel 7.0.0.11 May 2018"

This reverts commit 9b05db33b54e6e5f0b4658f4c06e7fe563f8923b.

Fixes https://github.com/harfbuzz/harfbuzz/issues/1708

Revert "[ragel] Switch to -T1 output instead of -F1"

This reverts commit ae8719eb596485ebff07dd5016256015cd0cf86b.

Part of https://github.com/harfbuzz/harfbuzz/issues/1708

Fix extra semi-colon

[ragel] Switch to -T1 output instead of -F1

Fedora upgraded to ragel 7, which is buggy if char is signed.
Switching to -G2 output fails with sign-compare error:

../../src/hb-buffer-deserialize-json.hh:107:12: error: comparison of integer expressions of different signedness: ‘unsigned int’ and ‘const char’ [-Werror=sign-compare]
    if ( 9u <= ( (*( p))) && ( (*( p))) <= 13u ) {
         ~~~^~~~~~~~~~~~~

Switching to -T1 for now.  It actually results in smaller code,
at the expense of some binary searching instead of flat tables.
In the not distant future, we might actually generate two different
outputs and choose between depending on size-optimize options.

Fixes https://github.com/harfbuzz/harfbuzz/issues/1708

Minor

Fix MSVC build

[config] Define HB_NO_SUBSET_LAYOUT in HB_LEAN

Assumning subsetter would be used for printing-like uses in that case,
which don't need GSUB/GPOS.

[config] Add HB_NO_UNISCRIBE_BUG_COMPATIBLE

Part of https://github.com/harfbuzz/harfbuzz/issues/1652

[config] Disable buffer serialize routines in HB_TINY

Part of https://github.com/harfbuzz/harfbuzz/issues/1652

Merge pull request #1665 from n8willis/docs-gtkdoc-colormath

[Docs] Add gtk-doc comments for OT color and OT math

Fix builds

[config] Enable HB_NO_MT in HB_TINY

Now that user can override it if needed...

Part of https://github.com/harfbuzz/harfbuzz/issues/1652

Force-disable CFF code under disabling conditions

Subsetter size goes down from 190kb to 119kb.  Main library about 7kb.

Part of https://github.com/harfbuzz/harfbuzz/issues/1652

[config] Allow overriding chosen config

[test] minor, fix -Weverything bot

[test] minor

style fix and add return statement

Merge branch 'master' into docs-gtkdoc-colormath

Corrections to OT Color gtk-doc comments.

Corrections to OT Math gtk-doc comments.

Fix build

Disable sbix if no-color or no-ot-font-bitmap

Part of https://github.com/harfbuzz/harfbuzz/issues/1652

[iter] Check for more before forwarding/rewinding past ends

[iter] Fix filter rewinding

[array] Fix rewinding

Update src/hb-ot-color.cc

Co-Authored-By: Khaled Hosny <khaledhosny@eglug.org>

[iter] Add hb_map_sorted() and hb_map_retains_sorting()

Update src/hb-ot-math.cc

Co-Authored-By: Khaled Hosny <khaledhosny@eglug.org>

Update src/hb-ot-math.cc

Co-Authored-By: Khaled Hosny <khaledhosny@eglug.org>

Update src/hb-ot-color.cc

Co-Authored-By: Khaled Hosny <khaledhosny@eglug.org>

Update src/hb-ot-color.cc

Co-Authored-By: Khaled Hosny <khaledhosny@eglug.org>

Update src/hb-ot-color.cc

Co-Authored-By: Khaled Hosny <khaledhosny@eglug.org>

Update src/hb-ot-color.cc

Co-Authored-By: Khaled Hosny <khaledhosny@eglug.org>

Update src/hb-ot-color.cc

Co-Authored-By: Khaled Hosny <khaledhosny@eglug.org>

[config] Rename

[config] Add HB_NO_NAME

Part of https://github.com/harfbuzz/harfbuzz/issues/1652