20.【Marp】Why Japanese “Tofu” (□) Cannot Be Fixed by Settings

The Structural Relationship Between PPTX and Fonts

tags:

Markdown
Marp
PowerPoint
Fonts
GitHubActions

🎯 What This Article Is About

In the previous article,
I described how Japanese text turns into □ (tofu) when generating PPTX files with
Marp × GitHub Actions.

👉 This is not a solution article.

This article has only one purpose:

To explain, structurally, why Japanese tofu cannot be fixed by “tweaking settings.”

Conclusion (Up Front)

Japanese tofu is not caused by missing Marp settings.
It is a structural consequence of how PPTX and fonts work.

Therefore:

Changing the viewer
Replacing fonts in PowerPoint
Temporarily “fixing” it locally

may appear to work, but:

the problem will reappear in CI.

🧠 What “Tofu” (□) Really Is

First, the basics.

□ is not character encoding corruption
It is not a UTF-8 issue

It simply means:

The font does not contain the glyph for that character.

In PPTX:

A specific font is assigned
If that font lacks a glyph

PowerPoint may render □ instead of falling back.

💣 Why This Happens So Easily with Marp + PPTX

The Core Problem: `code` / `pre`

In Marp:

Inline code
Code blocks

are rendered using monospace fonts.

However:

Most monospace fonts
do not contain Japanese glyphs

As a result:

HTML: browsers automatically apply font fallback
PPTX: font is fixed → □ appears

👉
This is the main reason why
“it looks fine in HTML but breaks in PPTX.”

❌ Why It Feels Like “Settings Fixed It”

Common attempts:

Replacing fonts in PowerPoint
Changing the viewer
Installing Japanese fonts locally

These may indeed:

Fix it temporarily
Fix it on that machine

But in CI:

The environment is not the same
The viewer does not exist

👉
Fixes without reproducibility are meaningless in CI.

🔁 Why the Problem Reappears in CI

In CI environments:

OS images are clean
Fonts are minimal
No viewer behavior is involved

In other words:

“It worked locally” guarantees nothing.

The reappearance of tofu in CI
is expected behavior, not an anomaly.

🧩 Where the Only Correct Fix Actually Is

The conclusion is clear.

Not PowerPoint
Not the viewer
Not GitHub Actions

👉
The only reproducible fix point is font specification
in Marp front matter (CSS) from the start.

That is the only place where behavior can be controlled consistently.

⚠️ Why This Still Isn’t a “Perfect Solution”

To be honest:

Few monospace fonts fully support Japanese
Font fallback differs by environment
PPTX itself is a black box

Which means:

There is currently no 100% safe configuration.

What you can do is:

Minimize recurrence probability
Avoid known landmines

That’s the realistic limit.

🧠 Practical Judgment Rule

When in doubt, ask this one question:

Is this actually “code”?

YES → use code / pre
NO → use normal text

Do not put Japanese explanations inside code blocks.
This alone eliminates most incidents.

Summary

Japanese tofu is not a bug
It is not a misconfiguration
It is a structural issue of PPTX and fonts