open-swe/agent/utils/adf.py

"""Minimal Atlassian Document Format (ADF) <-> markdown conversion.

Jira Cloud v3 issue descriptions and comments are ADF JSON, not markdown. These
helpers convert the node types Atlassian actually emits in descriptions/comments
so bodies read as markdown in prompts, and produce valid ADF for agent-authored
comments (prose). ``markdown_to_adf`` is intentionally minimal: agent comments
are plain prose, so each blank-line-separated block becomes one paragraph.
"""

from __future__ import annotations

from typing import Any

_MARK_WRAP = {
    "strong": "**",
    "em": "_",
    "code": "`",
    "strikethrough": "~~",
}


def _apply_marks(text: str, marks: list[dict[str, Any]]) -> str:
    for mark in marks:
        mtype = mark.get("type")
        if mtype == "link":
            href = (mark.get("attrs") or {}).get("href", "")
            text = f"[{text}]({href})"
        elif mtype in _MARK_WRAP:
            wrap = _MARK_WRAP[mtype]
            text = f"{wrap}{text}{wrap}"
    return text


def _render_nodes(nodes: list[dict[str, Any]]) -> str:
    return "".join(_render_node(n) for n in nodes)


def _render_node(node: dict[str, Any]) -> str:  # noqa: PLR0911, PLR0912
    ntype = node.get("type")
    content = node.get("content", []) or []
    attrs = node.get("attrs", {}) or {}

    if ntype == "text":
        return _apply_marks(node.get("text", ""), node.get("marks", []) or [])
    if ntype == "hardBreak":
        return "\n"
    if ntype == "paragraph":
        return _render_nodes(content) + "\n\n"
    if ntype == "heading":
        level = min(int(attrs.get("level", 1)), 6)
        return f"{'#' * level} {_render_nodes(content)}\n\n"
    if ntype == "blockquote":
        inner = _render_nodes(content).strip()
        return "".join(f"> {line}\n" for line in inner.splitlines()) + "\n"
    if ntype == "codeBlock":
        lang = attrs.get("language", "")
        return f"```{lang}\n{_render_nodes(content)}\n```\n\n"
    if ntype == "rule":
        return "---\n\n"
    if ntype == "bulletList":
        return (
            "".join(f"- {_render_nodes(li.get('content', [])).strip()}\n" for li in content) + "\n"
        )
    if ntype == "orderedList":
        out = []
        for i, li in enumerate(content, start=1):
            out.append(f"{i}. {_render_nodes(li.get('content', [])).strip()}\n")
        return "".join(out) + "\n"
    if ntype == "listItem":
        return _render_nodes(content)
    if ntype in ("mediaSingle", "mediaGroup"):
        return _render_nodes(content)
    if ntype == "media":
        alt = attrs.get("alt") or attrs.get("id", "media")
        url = attrs.get("url", "")
        return f"![{alt}]({url})\n\n" if url else f"[media: {alt}]\n\n"
    if ntype == "inlineCard":
        return (attrs.get("url", "")) or ""
    if ntype == "mention":
        return attrs.get("text", "") or ""
    if ntype == "emoji":
        return attrs.get("text", "") or attrs.get("shortName", "") or ""
    # Unknown/container node: recurse into content.
    return _render_nodes(content)


def adf_to_markdown(adf: Any) -> str:
    """Convert an ADF document (or None) to a markdown string."""
    if not adf or not isinstance(adf, dict):
        return ""
    return _render_nodes(adf.get("content", []) or []).strip()


def markdown_to_adf(text: str) -> dict[str, Any]:
    """Convert plain markdown/prose to a minimal ADF document.

    Blank-line-separated blocks become paragraphs; single newlines within a
    block become hardBreaks. Inline markdown is left as literal text.
    """
    text = text or ""
    blocks = text.split("\n\n")
    paragraphs: list[dict[str, Any]] = []
    for block in blocks:
        if not block.strip():
            continue
        lines = block.split("\n")
        para_content: list[dict[str, Any]] = []
        for idx, line in enumerate(lines):
            if idx > 0:
                para_content.append({"type": "hardBreak"})
            if line:
                para_content.append({"type": "text", "text": line})
        paragraphs.append({"type": "paragraph", "content": para_content})

    if not paragraphs:
        paragraphs = [{"type": "paragraph", "content": []}]
    return {"type": "doc", "version": 1, "content": paragraphs}