mirror of
https://github.com/fish-shell/fish-shell.git
synced 2024-12-24 02:13:39 +08:00
159 lines
5.5 KiB
Python
159 lines
5.5 KiB
Python
# Pygments lexer for a fish command synopsis.
|
|
#
|
|
# Example usage:
|
|
# echo 'string match [OPTIONS] [STRING]' | pygmentize -f terminal256 -l doc_src/fish_synopsis.py:FishSynopsisLexer -x
|
|
|
|
from docutils import nodes
|
|
from pygments.lexer import Lexer
|
|
from pygments.token import (
|
|
Generic,
|
|
Name,
|
|
Operator,
|
|
Punctuation,
|
|
Text,
|
|
)
|
|
import re
|
|
from sphinx.directives.code import CodeBlock
|
|
|
|
|
|
class FishSynopsisDirective(CodeBlock):
|
|
"""A custom directive that describes a command's grammar."""
|
|
|
|
has_content = True
|
|
required_arguments = 0
|
|
|
|
def run(self):
|
|
if self.env.app.builder.name != "man":
|
|
self.arguments = ["fish-synopsis"]
|
|
return CodeBlock.run(self)
|
|
lexer = FishSynopsisLexer()
|
|
result = nodes.line_block()
|
|
for start, tok, text in lexer.get_tokens_unprocessed("\n".join(self.content)):
|
|
if ( # Literal text.
|
|
(tok in (Name.Function, Name.Constant) and not text.isupper())
|
|
or text.startswith("-") # Literal option, even if it's uppercase.
|
|
or tok in (Operator, Punctuation)
|
|
or text
|
|
== " ]" # Tiny hack: the closing bracket of the test(1) alias is a literal.
|
|
):
|
|
node = nodes.strong(text=text)
|
|
elif (
|
|
tok in (Name.Constant, Name.Function) and text.isupper()
|
|
): # Placeholder parameter.
|
|
node = nodes.emphasis(text=text)
|
|
else: # Grammar metacharacter or whitespace.
|
|
node = nodes.inline(text=text)
|
|
result.append(node)
|
|
return [result]
|
|
|
|
|
|
lexer_rules = [
|
|
(re.compile(pattern), token)
|
|
for pattern, token in (
|
|
# Hack: treat the "[ expr ]" alias of builtin test as command token (not as grammar
|
|
# metacharacter). This works because we write it without spaces in the grammar (like
|
|
# "[OPTIONS]").
|
|
(r"\. |! |\[ | \]", Name.Constant),
|
|
# Statement separators.
|
|
(r"\n", Text.Whitespace),
|
|
(r";", Punctuation),
|
|
(r" +", Text.Whitespace),
|
|
# Operators have different highlighting than commands or parameters.
|
|
(r"\b(and|not|or|time)\b", Operator),
|
|
# Keywords that are not in command position.
|
|
(r"\b(if|in)\b", Name.Function),
|
|
# Grammar metacharacters.
|
|
(r"[()[\]|]", Generic.Other),
|
|
(r"\.\.\.", Generic.Other),
|
|
# Parameters.
|
|
(r"[\w-]+", Name.Constant),
|
|
(r"[=%]", Name.Constant),
|
|
(
|
|
r"[<>]",
|
|
Name.Constant,
|
|
), # Redirection are highlighted like parameters by default.
|
|
)
|
|
]
|
|
|
|
|
|
class FishSynopsisLexer(Lexer):
|
|
name = "FishSynopsisLexer"
|
|
aliases = ["fish-synopsis"]
|
|
|
|
is_before_command_token = None
|
|
|
|
def next_token(self, rule: str, offset: int, has_continuation_line: bool):
|
|
for pattern, token_kind in lexer_rules:
|
|
m = pattern.match(rule, pos=offset)
|
|
if m is None:
|
|
continue
|
|
if token_kind is Name.Constant and self.is_before_command_token:
|
|
token_kind = Name.Function
|
|
|
|
if has_continuation_line:
|
|
# Traditional case: rules with continuation lines only have a single command.
|
|
self.is_before_command_token = False
|
|
else:
|
|
if m.group() in ("\n", ";") or token_kind is Operator:
|
|
self.is_before_command_token = True
|
|
elif token_kind in (Name.Constant, Name.Function):
|
|
self.is_before_command_token = False
|
|
|
|
return m, token_kind, m.end()
|
|
return None, None, offset
|
|
|
|
def get_tokens_unprocessed(self, input_text):
|
|
"""Return a list of (start, tok, value) tuples.
|
|
|
|
start is the index into the string
|
|
tok is the token type (as above)
|
|
value is the string contents of the token
|
|
"""
|
|
"""
|
|
A synopsis consists of multiple rules. Each rule can have continuation lines, which
|
|
are expected to be indented:
|
|
|
|
cmd foo [--quux]
|
|
[ARGUMENT] ...
|
|
cmd bar
|
|
|
|
We'll split the input into rules. This is easy for a traditional synopsis because each
|
|
non-indented line starts a new rule. However, we also want to support code blocks:
|
|
|
|
switch VALUE
|
|
[case [GLOB ...]
|
|
[COMMAND ...]]
|
|
end
|
|
|
|
which makes this format ambiguous. Hack around this by always adding "end" to the
|
|
current rule, which is enough in practice.
|
|
"""
|
|
rules = []
|
|
rule = []
|
|
for line in list(input_text.splitlines()) + [""]:
|
|
if rule and not line.startswith(" "):
|
|
rules.append(rule)
|
|
rule = []
|
|
if line == "end":
|
|
rules[-1].append(line)
|
|
continue
|
|
rule.append(line)
|
|
result = []
|
|
for rule in rules:
|
|
offset = 0
|
|
self.is_before_command_token = True
|
|
has_continuation_line = rule[-1].startswith(" ")
|
|
rule = "\n".join(rule) + "\n"
|
|
while True:
|
|
match, token_kind, offset = self.next_token(
|
|
rule, offset, has_continuation_line
|
|
)
|
|
if match is None:
|
|
break
|
|
text = match.group()
|
|
result.append((match.start(), token_kind, text))
|
|
assert offset == len(rule), "cannot tokenize leftover text: '{}'".format(
|
|
rule[offset:]
|
|
)
|
|
return result
|