resolves #386 install CJK font

[ggrossetie]

With this change, CJK characters are now correctly rendered when converting to PNG:

For reference, here's the result using http://www.plantuml.com/

The package font-noto-cjk (and all its dependencies) is quite large, approximately 200MB.
Here's the Docker image size before (0.10.0) and after (latest):

yuzutech/kroki     0.10.0     434MB
yuzutech/kroki     latest     650MB

resolves #386

[zengxs]

The package font-noto-cjk-extra includes all font weights, but most of them are unnecessary.

I think source-han-sans might be a better choice:
(noto-cjk is same with source-han-sans, ref this)

Weight	Url	Size
Regular	https://github.com/adobe-fonts/source-han-sans/blob/release/OTC/SourceHanSans-Regular.ttc	19.1 MB
Normal	https://github.com/adobe-fonts/source-han-sans/blob/release/OTC/SourceHanSans-Normal.ttc	17.8 MB
Medium	https://github.com/adobe-fonts/source-han-sans/blob/release/OTC/SourceHanSans-Medium.ttc	18 MB

I think the weight Regular is enough for kroki.

---

Can also use serif cjk font source-han-serif (like the result using http://www.plantuml.com/).

---

And if this font is installed, the font wqy-zenhei (Chinese font) can be removed:

kroki/server/ops/docker/jdk11-alpine/Dockerfile

Lines 24 to 27 in ba0913c

	# Chinese font (available only in edge/testing)
	&& apk add --update --no-cache \
	--repository https://dl-cdn.alpinelinux.org/alpine/edge/testing/ \
	wqy-zenhei

[ggrossetie]

Thanks for your input @zengxs
That's exactly the kind of information I was looking for! 👍🏻

source-han-sans is not available as an apk package but I will try to manually install the font using TrueType Collection (ttc) files.

[ggrossetie]

I took a closer look and Noto has exactly the same size:

Weight	Url	Size
Regular	https://github.com/googlefonts/noto-cjk/blob/main/NotoSansCJK-Regular.ttc	19.1 MB
Medium	https://github.com/googlefonts/noto-cjk/blob/main/NotoSansCJK-Medium.ttc	18 MB

The package font-noto-cjk-extra is quite large because it contains several variations:

• NotoSansCJK-Bold.ttc
• NotoSansCJK-Regular.ttc
• NotoSerifCJK-Bold.ttc
• NotoSerifCJK-Regular.ttc
• NotoSansCJK-Black.ttc
• NotoSansCJK-DemiLight.ttc
• NotoSansCJK-Light.ttc
• NotoSansCJK-Medium.ttc
• NotoSansCJK-Thin.ttc
• NotoSerifCJK-Black.ttc
• NotoSerifCJK-ExtraLight.ttc
• NotoSerifCJK-Light.ttc
• NotoSerifCJK-Medium.ttc
• NotoSerifCJK-SemiBold.ttc

As you've mentioned, we probably don't want to include all font weights.
I just noticed that we are using the -extra package but instead we could use: font-noto-cjk which only contains:

• NotoSansCJK-Bold.ttc
• NotoSansCJK-Regular.ttc
• NotoSerifCJK-Bold.ttc
• NotoSerifCJK-Regular.ttc

I think source-han-sans might be a better choice:

I don't have a strong opinion since I can't tell if source-han-sans is objectively better than noto-sans or not.

[ggrossetie]

Here's the Docker image size before (0.11.0) and after (latest) with font-noto-cjk package:

yuzutech/kroki     0.11.0     455MB
yuzutech/kroki     latest     530MB

That's a 75MB increase which is fine and I'm willing to move forward with this change.

[zengxs]

noto-sans-cjk is also good, it is default cjk font in most linux desktop distros, but source-han-sans is more famous in China.

In fact, they are almost the same font (ref this post), just with different names.

---

If you want to install font from alpine package repository, noto-sans is the best choice.
If you want to install font manually, source-han-sans's font-name has better compatibility than noto-sans (many programs/websites optimized for source-han-sans but noto-sans not), although this improvement is negligible. 😃

[ggrossetie]

Thanks again for your feedback and links to resources, this is greatly appreciated 👍🏻

noto-sans-cjk is also good, it is default cjk font in most linux desktop distros, but source-han-sans is more famous in China.
In fact, they are almost the same font (ref this post), just with different names.

OK, I see.
It seems that Google and Adobe worked hand in hand to produce a CJK font.

If you want to install font from alpine package repository, noto-sans is the best choice.
If you want to install font manually, source-han-sans's font-name has better compatibility than noto-sans (many programs/websites optimized for source-han-sans but noto-sans not), although this improvement is negligible.

I quite like the fact that noto-sans is available as an apk package, it keeps the Dockerfile simple.
Let's start with Noto and see if this option is good enough or not (i.e., we can always reevaluate later).

[ggrossetie]

For reference, here's the result using Noto: