openal-soft assumes that aligned_alloc is available with c++17. Newer
android ndks set c++17 by default, however they do not expose
aligned_alloc without setting min sdk version to 28.
We can avoid this issue by forcing openal to be compiled with c++11.
Also note, we have HXCPP_CPP11 defined, however, hxcpp ignores this for
the android toolchain. This means we must set it explicitly
See:
f5e0eef34d/common/almalloc.cpp (L15)
Hxcpp doesn't respect HXCPP_CPP11 for android. We can avoid this issue just by adding -std=c++11 here. This resolves a conflict with openal soft 1.20.1 for android builds. This works because the library assumes c++17 and invokes a function that isn't available in android sdk below 28.
If the text is invalid, then readNextChar returns -1 and does not
progress to the next character. This previously meant that we got stuck
and looped indefinitely.
Previously, the first for loop would reach the end of the characters, so
no further characters were read in the second loop. This meant that the
array remained filled with 0 values.
This function wants the utf8 length, but text->length does not give
this. We can instead use -1 to tell harfbuzz to treat it as null-
terminating, which is the case with all hashlink strings anyway.
A HxString is not guaranteed to be utf8 on hxcpp, so we need to use
hxs_utf8 to convert it otherwise it incorrectly displays any utf16
encoded string.
Unfortunately, hxs_utf8 doesn't give us the length of the utf8 string.
The best thing we can do is to tell harfbuzz it is null terminated. This
isn't perfect because technically hxcpp strings are allowed to contain
NULL, but it's better than all utf16 strings being broken.
Hxcpp's mbedtls has MBEDTLS_NET_C disabled, which meant that older
versions of curl which use this feature cannot be linked against it.
This version of curl no longer requires this feature, which avoids the
issue.
Hxcpp 4.3.0 has an mbedtls_config.h file in ${HXCPP}/src/hx/libs/ssl,
which we have to ensure is included, but older versions do not.
To support both 4.3.0 and older versions, we can add an empty
mbedtls_config.h and add the include path at the end. This way it will
only be used if it does not exist in the previous include paths.
* Fix copyPixels crash on Hashlink
* Move nullcheck to C++
* Simplify further
* Avoid unneccessary object allocation
* Formatting
* Simplify by using an if statement
We're currently using freetype 2.9.1's algorithm for choosing font metrics, but it's not exactly the same as SWF (but it is closer than freetype 2.10's new algorithm).
This algorithm should more closely match the font metrics used by AS3 compilers when embedding fonts with [Embed] metadata in AS3.
This allows us to upgrade freetype beyond 2.9.1 (actually, 2.8.1 due to a bug in freetype 2.9) without wildly different font metrics. The reason that we don't want to change font metrics algorithms is that freetype's new algorithm is very different than the algorithm used for SWF fonts. The older freetype algorithm is closer to SWF, so we want to stick with it.
Thankfully, freetype supports accessing various metrics stored in font files in a more raw form, so we can provide our own custom algorithm in a way that is fully supported by freetype. I just copied the existing algorithm straight from 2.9.1 to restore our preferred behavior.
I confirmed that, after this change, OpenFL renders metrics like Lime 8.1.3. But we have the same upgraded freetype as Lime 8.2.0 that we had to revert for 8.2.1.
I plan to experiment with a mode that matches SWF even more closely, now that I have a better understanding of how fonts and freetype work.
We use a default of 72 for now to ensure text is formatted correctly. Not entirely sure why we are stuck on 72 but that is an investigation for the future.
This fixes some unexpected changes in text rendering for OpenFL which rely on a private function in Lime. Previously we defaulted at 72 dpi, which apparently is expected to layout the text properly. Most displays are 96 dpi or higher today, so we should probably look into this. For RenderGlyph, which was previously broken over several versions, we will now use a dpi of 96 for now. In the next version of lime, we absolutely should alter the function signature to allow for renderGlyph to accept a dpi argument.
tag VER-2-9-1
Starting with freetype 2.10.0, the sum of the ascent and descent values seem to be more likely to be less than the baseline-to-baseline measurement (called the font's height), which is the font designer's recommend distance between baselines. However, OpenFL doesn't account for the full baseline-to-baseline height at all, so with smaller ascent and descent values, lines render with smaller gaps between them. OpenFL needs to update its TextEngine algorithm to use the height of the line instead of adding ascent and descent together alone, and then we can update freetype. Updating TextEngine is not a trivial change, and may likely require time to discover bugs and stabilize, so it's better to roll back the freetype update for now, and apply it again later after OpenFL can handle it properly.
I think it should probably use 96 dpi instead of 72.... In the next minor version of Lime I think we should change the function sig to allow a dpi argument.
We change the way RenderGlyph populates the binary data here. It's a little heavier than the 8-bit method used previously but I was having issues getting that to work properly with `Image`.
This change eliminates the cleanup attempt of the current OpenAL context and associated system resources. A change in OpenAL 1.20.0 made it unlikely that we will be able to clean up things in this way moving forward. We should steer users toward using `lime.system.System.exit()` or petition a change in the haxe stdlib to allow us to hook into Sys.exit().
I am not 100% satisfied with this, so perhaps we will find another solution. In the end, I think the benefit of updating OpenAL supersedes any inconvenience here.
Closes https://github.com/openfl/lime/issues/1803
