dcdpr / jp / 22244247340

58%

push

github

feat(md, config): Implement streaming markdown rendering (#399)

Introduce the `jp_md` crate to handle progressive rendering of
CommonMark content from LLM streams. This implementation solves the
latency issues associated with waiting for a full response before
displaying formatted output.

The core of this change is the `Buffer` utility, which segments incoming
text chunks into complete markdown blocks based on CommonMark grammar.
This allows for immediate rendering of paragraphs, headers, and lists as
they are completed. Fenced code blocks are handled with line-by-line
granularity to enable real-time syntax highlighting.

A custom `TerminalFormatter` replaces standard rendering to ensure
ANSI-aware line wrapping. Unlike generic formatters, it correctly
calculates visual column widths by excluding escape sequences and
manages SGR state across line breaks to prevent background color bleed.

Users can now configure their rendering preferences via the new
`style.markdown` configuration section:

```toml
[style.markdown]
wrap_width = 80
table_max_column_width = 40
theme = "gruvbox-dark"
hr_style = "line"
```

The architectural decisions and trade-offs for this implementation are
documented in RFD 004.

NOTE: This change has NOT been wired up to the CLI yet. This will happen
in a follow-up PR.

---------

Signed-off-by: Jean Mertz <git@jeanmertz.com>

2359 of 2902 new or added lines in 14 files covered. (81.29%)

13965 of 23894 relevant lines covered (58.45%)

378.79 hits per line