The pen is designed to determine the bitmap size when omitted, which
helps users to see an image somehow even when they have no idea how the
arguments should be passed. And I realised that I didn't give enough
thought to rotate/skew transforms in former PRs. This commit fixes the
calculation of the bbox after transformation. Also tries to clear up how
the autosizing options work in the docstring.
Some minor fixes will follow.
this test currently fails; currently the second, third, etc. base glyphs with the same layers gets a unique PaintColrLayers with NumLayers:1 that in turn contains the shared PaintColrLayers. They should instead all share the same PaintColrLayers. Fix in the next commit
Surprisingly I found Preview.app can still display PGM images. While
somewhat legacy, it's a super straightforward format to (de)serialize.
The images are scaled to 50x50 px and only consume 5KB in total. Makes
more sense to human being than the previous base64-encoded zlib
compressed data, plus we don't have to add pillow as a dependency.
Eliminates the assumption of any specific metrics from the pen. It still
gives some image without giving any parameters, thus it should be a good
starting point for new users.
There are possibilities that the rendering results may change among
FreeType versions. I've already used the PSNR comparison for cubic
vs quadratic testing, so I applied the same technique and threshold to
all rendering tests to relax assertions. Also handles the case that
MSE becomes zero. Optional dependencies are not needed for the tests.
After experimenting with uharfbuzz for a while, I found out it was hard
to handle top-to-bottom texts, so I gave up an idea to put an ascender
or a descender value in the arguments. Instead, I simply expose 'offset',
'width' and 'height', which is way more straightforward than the
previous design.
In addition, 'contain' option is added to easily compensate and render
glyphs such as combining accents or excessively tall letters.
Part of fixing https://github.com/fonttools/fonttools/issues/2475
Old code was mistakenly only adding the synthetic-generated feature
to the first language-system. In the case of merging Noto Grantha
and Noto Tamil fonts, Noto Tamil had two script systems: 'taml' and
'tml2', and we were adding the synthetic 'locl' feature only to the
DefaultLangSys of the 'taml' script, not the 'tml2' script. That was
the main bug.
A second issue was that the lookup was being added twice to the same
(synthetic) feature. Check before adding lookup, to avoid that.
