Use font-kit to implement XeTeXFontInst #188

cormacrelf · 2019-11-14T07:36:37Z

#186 It's a work in progress. Notable semi-related things included:

The Read implementation was converting -1 (error) to u64 and telling
read_to_end to start reading at the other end of the universe.
Many cleanups of the font manager/font info c2rust conversion.
fixes Merge xetex_layout_engine.rs into xetex_layout_interface.rs #142
FT_MAKE_TAG macro replaced
Some name splitting code refactored, to make find_native_font more manageable as font-kit will hopefully replace some of it.
Fixed some non-FFI-safe warnings that are new in recent nightly compilers I think?

Some things that need work:

The global name_of_file and other tex globals are way too involved in font loading. It's painfully unnecessary. set/getReqEngine also.
There's heavy mixing of the slash-code font loading options like "/AAT", with the rest of the font loading. I'm not quite sure how that works or how it could be disentangled. Also, font-kit will find a font by name, but if it has to search by name and by type... ?
PlatformFontRef has replaced XeTeXFont, which was just a type alias, but that has made it clear that not every PlatformFontRef is actually a PlatformFontRef -- many are cast willy-nilly to XeTeXFontInst and the platform-specific subclasses. This is terrible.
The C++ inheritance needs to go. I haven't come up with a neat way of doing it yet.
The harfbuzz interactions could be done with the user_data pointing to a font_kit Font. I'm not sure Font supports all the things harfbuzz is going to need, but you can always keep forcing font-kit to use the freetype loader and use Font.native_font() to access FT_Face.
I think putting things in harfbuzz user_data should use a Pin pointer?

Some notes:

loadAATFont and loadOTFont return completely different objects, each under a *mut libc::c_void pointer, but both pointers are stored in the font_layout_engine static mut. font_area 's data stored at the same offset is effectively an enum tag of layout engine types.
loadAATFont's "layout engine" is actually just a CFDictionaryRef of font attributes. Every thing the "layout engine" is required to do, xetex_ext calls into some aat:: function for, after discovering font_area is 0xffffu32. So you could think of it as a layout engine, since the resulting API looks similar to using functions (not methods) from xetex_layout_engine.
loadOTFont is poorly named. It doesn't load the font. It takes a font, matches some options, and essentially wraps createLayoutEngine. No other code uses createLayoutEngine.
loadOTFont returns a XeTeXLayoutEngine, which holds a font: XeTeXFontInst from createFont, and fontRef: PlatformFontRef, which is the font loaded with findFontByName.
I honestly have no idea why you don't see panics in measure_native_node, which doesn't handle AAT fonts but is called during, among other places, line_break.

Here's the big plan:

Make a trait for everything in xetex_ext that checks font_area/0xffffu32/0xfffeu32
impl LayoutEngine for a newtype struct AATLayoutEngine { attributes: CFDictionaryRef }
impl LayoutEngine for XeTeXLayoutEngine
Store Box<dyn LayoutEngine> in font_layout_engine, and delete font_area. (Hooray!)
See if you can use Skribo to make another impl LayoutEngine similar to XeTeXLayoutEngine
Improve the LayoutEngine API over time

cormacrelf · 2019-11-14T16:28:55Z

engine/src/xetex_font_info.rs

+
+    pub unsafe fn get_glyph_bounds(&self, gid: GlyphID) -> GlyphBBox {
+        if let Some(font) = &self.fk_font {
+            if let Ok(rect) = font.typographic_bounds(gid as u32) {


FYI, this is the only place we're actually querying font information so far. Further down, font-kit is also used to load the FT_Face.

cormacrelf · 2019-11-14T16:33:47Z

engine/src/xetex_font_manager.rs

+        let mut varString = String::new();
+        let mut slice = &variant_string[..];
+        while !slice.is_empty() {
+            if slice.starts_with("AAT") {


This is the slash-options intermingling I was talking about.

3927a7c

…_layout_engine

…#141)

ttstub_input_read actually translated into read_exact, which means that the provided methods of Read were getting messed up by the errors being returned. So it's been renamed to what it actually means: ttstub_input_read_exact, and supplemented with ttstub_input_read which does the normal Read::read. Importantly, Read::read is implemented with the latter.

…font manager

e.g. [lmroman12-regular], parsed into something? Not sure. Kept that code.

let mut sl = CStringListBuilder::new(); sl.push_non_null_terminated(&b"ot"[..]); ... let sl = sl.freeze(); hb_shape_plan_create_cached(..., sl.as_ptr());

Note especially the very clean API for layout_text.

…extLayoutEngine calls

cormacrelf · 2019-11-23T08:57:23Z

Update:

Swapped out the void*-and-font_area-based dynamic typing for enums, using enum_dispatch to allow dropping down to non-trait methods.
Had an annoying time initializing a static mut hashmap, so it's in a lazy RefCell, which is suboptimal. Rust really doesn't make this bad pattern easy.
There's a bad measurement going on somewhere.

burrbull · 2019-11-23T11:32:34Z

Maybe if you split this PR on several and rebase it will be much easier to understand what's going on here.

burrbull · 2020-11-07T08:59:20Z

@cormacrelf
Could you make another try?

retry #188 (part 1 Enum Dispatch)

retry #188 (part 2)

burrbull · 2021-02-08T16:12:35Z

Closing this as all main changes are already in oxidize

cormacrelf commented Nov 14, 2019

View reviewed changes

cormacrelf force-pushed the font-kit branch from 8fd6caa to a5aa2b2 Compare November 14, 2019 16:43

cormacrelf added 27 commits November 18, 2019 17:29

Revert "Use freetype APIs from dependency crates as much as possible."

1054902

3927a7c

Clean up c2rust font manager/info

fc3f63e

XeTexFont -> PlatformFontRef; xetex_layout_interface moved into xetex…

42c3505

…_layout_engine

Fiddle with the translated C++ constructors

878e36e

Begin using font-kit, first for replacing freetype-rs (fixes crlf0710…

a993157

…#141)

Attempt to get some glygh bbox using font-kit

a148fbc

check errors in the Read implementation !!!

8b7216a

XeTeXFontInst backing with Vecs & read_to_end

3fa8831

Replace a lot of freetype imports

1070c1c

make it build on linux again

d479de8

No need to check ttstub's result in Seek

a4a7464

Remove more freetype, core text, definitions; remove extern "C" from …

2851bfb

…font manager

Remove extern "C" from the bridge API, because of ffi-safety warnings

c59e792

Misc added documentation notes & some renaming of french variable names

b67014b

replace FT_MAKE_TAG macro expansions with a const fn

f1932b3

refactor packed font name splitting

82ece06

e.g. [lmroman12-regular], parsed into something? Not sure. Kept that code.

Change some font pathnames to use CStr

50e1a89

remove all FreeType definitions from xetex_layout_engine.rs

37b02e2

Fix overloading of PlatformFontRef as *mut XeTeXFontInst

ef95793

Remove useless get/setReqEngine static

f948e4d

'reqEngine' turned into an enum ShaperRequest

e1d8884

Handle shaper list creation in its own struct

601fba1

let mut sl = CStringListBuilder::new(); sl.push_non_null_terminated(&b"ot"[..]); ... let sl = sl.freeze(); hb_shape_plan_create_cached(..., sl.as_ptr());

Fix a missing AAt/0xffffu32 match in xetex_ext

a4dd4b7

Allocate XeTeXLayoutEngine_rec with Box

abad6ea

TextLayoutEngine trait + impl for XetexLayoutEngine

0910dc8

Note especially the very clean API for layout_text.

Remove last bits of freetype importing

be9be65

cormacrelf force-pushed the font-kit branch from a5aa2b2 to be9be65 Compare November 18, 2019 07:03

cormacrelf added 14 commits November 18, 2019 18:50

Implement TextLayoutEngine for aat as well

a9fbc99

rename font_area and font_layout_engine for refactoring

9dd0297

Use enum_dispatch for TextLayoutEngine

a823c83

WIP: converted a lot of FONT_AREA shenanigans into enum-dispatched- T…

19d5721

…extLayoutEngine calls

Down to 77 errors

0a48579

.. 59 errors

021577f

Down to 5

c89962b

And now borrowck

cf9ad2c

Compiles

f9bf7cf

Fix remaining font_area linking

0e16656

Revert dump change

02b0a42

Prevent recursive borrowing with measure_native_node

28443eb

Correct the isOpenTypeMathFont translations

c065e17

Fix linux build

e3f3dd7

burrbull added a commit that referenced this pull request Feb 7, 2021

Merge pull request #353 from burrbull/cormc

d0a40ce

retry #188 (part 1 Enum Dispatch)

burrbull added a commit that referenced this pull request Feb 8, 2021

Merge pull request #354 from burrbull/cormc

d93606d

retry #188 (part 2)

burrbull closed this Feb 8, 2021

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Use font-kit to implement XeTeXFontInst #188

Use font-kit to implement XeTeXFontInst #188

cormacrelf commented Nov 14, 2019 •

edited

Loading

cormacrelf Nov 14, 2019

cormacrelf Nov 14, 2019

cormacrelf commented Nov 23, 2019

burrbull commented Nov 23, 2019

burrbull commented Nov 7, 2020

burrbull commented Feb 8, 2021

Use font-kit to implement XeTeXFontInst #188

Use font-kit to implement XeTeXFontInst #188

Conversation

cormacrelf commented Nov 14, 2019 • edited Loading

cormacrelf Nov 14, 2019

Choose a reason for hiding this comment

cormacrelf Nov 14, 2019

Choose a reason for hiding this comment

cormacrelf commented Nov 23, 2019

burrbull commented Nov 23, 2019

burrbull commented Nov 7, 2020

burrbull commented Feb 8, 2021

cormacrelf commented Nov 14, 2019 •

edited

Loading