zulip/zerver/lib/markdown/tabbed_sections.py

190 lines
6.6 KiB
Python

import re
from typing import Any, Dict, List, Mapping, Optional
import markdown
from markdown.extensions import Extension
from markdown.preprocessors import Preprocessor
from zerver.lib.markdown.priorities import PREPROCESSOR_PRIORITES
START_TABBED_SECTION_REGEX = re.compile(r"^\{start_tabs\}$")
END_TABBED_SECTION_REGEX = re.compile(r"^\{end_tabs\}$")
TAB_CONTENT_REGEX = re.compile(r"^\{tab\|\s*(.+?)\s*\}$")
CODE_SECTION_TEMPLATE = """
<div class="code-section {tab_class}" markdown="1">
{nav_bar}
<div class="blocks">
{blocks}
</div>
</div>
""".strip()
NAV_BAR_TEMPLATE = """
<ul class="nav">
{tabs}
</ul>
""".strip()
NAV_LIST_ITEM_TEMPLATE = """
<li data-language="{data_language}" tabindex="0">{label}</li>
""".strip()
DIV_TAB_CONTENT_TEMPLATE = """
<div data-language="{data_language}" markdown="1">
{content}
</div>
""".strip()
# If adding new entries here, also check if you need to update
# tabbed-instructions.js
TAB_SECTION_LABELS = {
"desktop-web": "Desktop/Web",
"ios": "iOS",
"android": "Android",
"mac": "macOS",
"windows": "Windows",
"linux": "Linux",
"python": "Python",
"js": "JavaScript",
"curl": "curl",
"zulip-send": "zulip-send",
"web": "Web",
"desktop": "Desktop",
"mobile": "Mobile",
"mm-default": "Default installation",
"mm-docker": "Docker",
"mm-gitlab-omnibus": "GitLab Omnibus",
"require-invitations": "Require invitations",
"allow-anyone-to-join": "Allow anyone to join",
"restrict-by-email-domain": "Restrict by email domain",
"zoom": "Zoom",
"jitsi-meet": "Jitsi Meet",
"bigbluebutton": "BigBlueButton",
"disable": "Disabled",
"chrome": "Chrome",
"firefox": "Firefox",
"desktop-app": "Desktop app",
"system-proxy-settings": "System proxy settings",
"custom-proxy-settings": "Custom proxy settings",
"stream": "From a stream view",
"not-stream": "From other views",
"via-recent-conversations": "Via recent conversations",
"via-left-sidebar": "Via left sidebar",
"instructions-for-all-platforms": "Instructions for all platforms",
"public-streams": "Public streams",
"private-streams": "Private streams",
"web-public-streams": "Web-public streams",
"via-user-profile": "Via the user's profile",
"via-organization-settings": "Via organization settings",
"via-personal-settings": "Via personal settings",
"default-subdomain": "Default subdomain",
"custom-subdomain": "Custom subdomain",
"zulip-cloud": "Zulip Cloud",
"self-hosting": "Self hosting",
"okta": "Okta",
"onelogin": "OneLogin",
"azuread": "AzureAD",
"keycloak": "Keycloak",
}
class TabbedSectionsGenerator(Extension):
def extendMarkdown(self, md: markdown.Markdown) -> None:
md.preprocessors.register(
TabbedSectionsPreprocessor(md, self.getConfigs()),
"tabbed_sections",
PREPROCESSOR_PRIORITES["tabbed_sections"],
)
class TabbedSectionsPreprocessor(Preprocessor):
def __init__(self, md: markdown.Markdown, config: Mapping[str, Any]) -> None:
super().__init__(md)
def run(self, lines: List[str]) -> List[str]:
tab_section = self.parse_tabs(lines)
while tab_section:
if "tabs" in tab_section:
tab_class = "has-tabs"
else:
tab_class = "no-tabs"
tab_section["tabs"] = [
{
"tab_name": "instructions-for-all-platforms",
"start": tab_section["start_tabs_index"],
}
]
nav_bar = self.generate_nav_bar(tab_section)
content_blocks = self.generate_content_blocks(tab_section, lines)
rendered_tabs = CODE_SECTION_TEMPLATE.format(
tab_class=tab_class, nav_bar=nav_bar, blocks=content_blocks
)
start = tab_section["start_tabs_index"]
end = tab_section["end_tabs_index"] + 1
lines = [*lines[:start], rendered_tabs, *lines[end:]]
tab_section = self.parse_tabs(lines)
return lines
def generate_content_blocks(self, tab_section: Dict[str, Any], lines: List[str]) -> str:
tab_content_blocks = []
for index, tab in enumerate(tab_section["tabs"]):
start_index = tab["start"] + 1
try:
# If there are more tabs, we can use the starting index
# of the next tab as the ending index of the previous one
end_index = tab_section["tabs"][index + 1]["start"]
except IndexError:
# Otherwise, just use the end of the entire section
end_index = tab_section["end_tabs_index"]
content = "\n".join(lines[start_index:end_index]).strip()
tab_content_block = DIV_TAB_CONTENT_TEMPLATE.format(
data_language=tab["tab_name"],
# Wrapping the content in two newlines is necessary here.
# If we don't do this, the inner Markdown does not get
# rendered properly.
content=f"\n{content}\n",
)
tab_content_blocks.append(tab_content_block)
return "\n".join(tab_content_blocks)
def generate_nav_bar(self, tab_section: Dict[str, Any]) -> str:
li_elements = []
for tab in tab_section["tabs"]:
tab_name = tab.get("tab_name")
tab_label = TAB_SECTION_LABELS.get(tab_name)
if tab_label is None:
raise ValueError(
f"Tab '{tab_name}' is not present in TAB_SECTION_LABELS in zerver/lib/markdown/tabbed_sections.py"
)
li = NAV_LIST_ITEM_TEMPLATE.format(data_language=tab_name, label=tab_label)
li_elements.append(li)
return NAV_BAR_TEMPLATE.format(tabs="\n".join(li_elements))
def parse_tabs(self, lines: List[str]) -> Optional[Dict[str, Any]]:
block: Dict[str, Any] = {}
for index, line in enumerate(lines):
start_match = START_TABBED_SECTION_REGEX.search(line)
if start_match:
block["start_tabs_index"] = index
tab_content_match = TAB_CONTENT_REGEX.search(line)
if tab_content_match:
block.setdefault("tabs", [])
tab = {"start": index, "tab_name": tab_content_match.group(1)}
block["tabs"].append(tab)
end_match = END_TABBED_SECTION_REGEX.search(line)
if end_match:
block["end_tabs_index"] = index
break
return block
def makeExtension(*args: Any, **kwargs: str) -> TabbedSectionsGenerator:
return TabbedSectionsGenerator(**kwargs)