The-OpenROAD-Project_OpenROAD/docs/src/scripts/extract_utils.py

## SPDX-License-Identifier: BSD-3-Clause
## Copyright (c) 2024-2026, The OpenROAD Authors

# This code contains the necessary regex parsing functions for manpage compilation.

import re


def outside_fences(text):
    """
    Predicate: is this offset outside every fenced code block?

    Tcl comments open with `#`, so `# Either run` inside a synopsis block reads
    as a level-1 heading to a plain regex. That misplaces every heading after
    it and, in turn, the section each code block belongs to.
    """
    spans = [
        (m.start(), m.end())
        for m in re.finditer(r"^```.*?^```", text, flags=re.DOTALL | re.MULTILINE)
    ]
    return lambda pos: not any(start <= pos < end for start, end in spans)


def extract_headers(text, level=1):
    assert isinstance(level, int) and level >= 1
    outside = outside_fences(text)
    pattern = r"^#{%d}\s+(.*)$" % level
    headers = [
        m.group(1)
        for m in re.finditer(pattern, text, flags=re.MULTILINE)
        if outside(m.start())
    ]
    # TODO: Handle developer commands
    # if "Useful Developer Commands" in headers: headers.remove("Useful Developer Commands")
    return headers


def synopsis_tcl_blocks(text):
    """
    The ```tcl block holding each command's synopsis, in document order.

    One `### <command>` section describes one command, so its synopsis is the
    first tcl block inside it -- the same block `extract_description` stops at.
    Any later tcl in the section is a worked example, and tcl before the first
    `###` is introductory usage. Counting those as commands too yielded more
    names and synopses than descriptions and options, and that count mismatch
    aborts the whole man2 build. Note the synopsis does not have to sit
    directly under the `###`: a section may open with `####` prose first.
    """
    outside = outside_fences(text)
    headings = [
        (m.start(), len(m.group(1)))
        for m in re.finditer(r"^(#{1,6})\s", text, flags=re.MULTILINE)
        if outside(m.start())
    ]
    synopsis = re.compile(r"```tcl\s+(.*?)```", flags=re.DOTALL)
    blocks = []
    for i, (start, level) in enumerate(headings):
        if level != 3:
            continue
        # The section ends at the next heading of the same or higher rank.
        end = next(
            (pos for pos, lvl in headings[i + 1 :] if lvl <= 3),
            len(text),
        )
        m = synopsis.search(text, start, end)
        if m:
            blocks.append(m)
    return blocks


def extract_tcl_command(text):
    # objective is to extract tcl command from the synopsis
    return [m.group(1).split()[0] for m in synopsis_tcl_blocks(text)]


def extract_description(text):
    # this is so that it always tries to match the longer headers first, to disambiguate
    sorted_headers = sorted(extract_headers(text, 3), key=len, reverse=True)
    headers = "|".join(re.escape(x) for x in sorted_headers)
    pattern = rf"### ({headers})(.*?)```tcl"
    custom_strings = re.findall(pattern, text, flags=re.DOTALL)
    return [custom_string[1].strip() for custom_string in custom_strings]


def extract_tcl_code(text, skip_markers=True):
    # Find all ```tcl blocks along with their position to check for skip markers.
    results = []
    for m in synopsis_tcl_blocks(text):
        # Check the text immediately before this block for a skip marker.
        if skip_markers:
            preceding = text[: m.start()]
            if "<!-- checker: skip -->" in preceding.split("\n")[-3:]:
                continue
        block = m.group(1)
        if "./test/gcd.tcl" not in block:
            results.append(block)
    return results


def extract_arguments(text):
    # Goal is to extract all the text from the end of tcl code to the next ### header.
    # Returns options and arguments.
    level2 = extract_headers(text, 2)
    level3 = extract_headers(text, 3)

    # form these 2 regex styles.
    # ### Header 1 {text} ### Header2; ### Header n-2 {text} ### Header n-1
    # ### Header n {text} ## closest_level2_header
    # Headers are literal text, so escape them the way extract_description
    # does. `## C++` (odb) would otherwise read as a possessive quantifier and
    # capture the wrong span, and `## ... (min-cut partitioning)` (par) as a
    # group, so the section would not match at all and match[0] below would
    # raise IndexError.
    first = [
        rf"### ({re.escape(level3[i])})(.*?)### ({re.escape(level3[i + 1])})"
        for i in range(len(level3) - 1)
    ]

    # find the next closest level2 header to the last level3 header.
    closest_level2 = [
        text.find(f"## {x}") - text.find(f"### {level3[-1]}") for x in level2
    ]
    trailing_level2 = [idx for idx, x in enumerate(closest_level2) if x > 0]

    # This will disambiguate cases where different level headers share the same name.
    if trailing_level2:
        last3, next2 = level3[-1], level2[trailing_level2[0]]
        second = [rf"### ({re.escape(last3)})(.*?)## ({re.escape(next2)})"]
    else:
        # No level2 header follows, so the last command section runs to the end
        # of the file. Most READMEs close with `## Authors`/`## License`, which
        # is what bounded this section; without that the lookup used to raise
        # IndexError and drop the module's man pages entirely.
        second = [rf"### ({re.escape(level3[-1])})(.*?)$"]
    final_options, final_args = [], []
    for idx, regex in enumerate(first + second):
        match = re.findall(regex, text, flags=re.DOTALL)
        # print(regex)
        # get text until the next header
        a = match[0][1]
        a = a[a.find("#") :]
        options = a.split("####")[1:]
        if not options:
            final_options.append([])
            final_args.append([])
            continue
        options, args = options[0], options[1:]
        final_options.append(extract_tables(options))
        tmp_arg = []
        for arg in args:
            tmp_arg.extend(extract_tables(arg))
        final_args.append(tmp_arg)
    return final_options, final_args


def extract_tables(text):
    # Find all lines that start with "|"
    table_pattern = r"^\s*\|.*$"
    table_matches = re.findall(table_pattern, text, flags=re.MULTILINE)

    # Exclude matches containing HTML tags
    table_matches = [table for table in table_matches if not re.search(r"<.*?>", table)]

    # Remove text containing switch
    table_matches = [table for table in table_matches if "Switch Name" not in table]

    # Remove text containing "---"
    table_matches = [table for table in table_matches if "---" not in table]

    return table_matches


def extract_help(text):
    # Match each sta::define_cmd_args block independently by tracking
    # brace depth, instead of spanning to the next "proc" keyword which
    # breaks when sta::proc_redirect is used between commands.
    matches = []
    for m in re.finditer(r'sta::define_cmd_args\s+"([^"]+)"\s*', text):
        name = m.group(1)
        after = text[m.end() :]
        # Walk through the args block tracking brace depth.
        depth = 0
        end = 0
        for j, ch in enumerate(after):
            if ch == "{":
                depth += 1
            elif ch == "}":
                depth -= 1
                if depth == 0:
                    end = j
                    break
        rest_of_line = after[end + 1 :].split("\n")[0]
        args_block = after[: end + 1] + rest_of_line
        if ";#checkeroff" in args_block.replace(" ", ""):
            continue
        # Only count commands that have a matching proc or proc_redirect.
        proc_pat = rf"(?:proc|sta::proc_redirect)\s+{re.escape(name)}\s"
        if re.search(proc_pat, text):
            matches.append((name, args_block))
    return matches


def extract_proc(text):
    proc_pattern = re.compile(
        r"""
                sta::parse_key_args\s+
                "(.*?)"\s*
                args\s*
                (.*?keys.*?})
                (.*?flags.*?})
                (\s*;\s*\#\s*checker\s*off)?
                """,
        re.VERBOSE | re.DOTALL,
    )

    matches = re.findall(proc_pattern, text)

    # remove nodocs (usually dev commands)
    matches = [tup for tup in matches if not tup[3].replace(" ", "") == ";#checkeroff"]
    return matches


def parse_switch(text):
    # Find the index of the 1nd and last occurrence of "|". Since some content might contain "|"
    switch_name = text.split("|")[1]
    switch_name = switch_name.replace("`", "").strip()
    second_pipe_index = text.find("|", text.find("|") + 1)
    last_pipe_index = text.rfind("|")
    switch_description = text[second_pipe_index + 1 : last_pipe_index - 1]
    return switch_name, switch_description


def clean_whitespaces(text):
    tmp = text.strip().replace("\\", "").replace("\n", "")
    return " ".join(tmp.split())


def clean_parse_syntax(text):
    tmp = (
        text.replace("keys", "").replace("flags", "").replace("{", "").replace("}", "")
    )
    return " ".join([f"[{option}]" for option in tmp.split()])


def check_function_signatures(text1, text2):
    set1 = set(re.findall(r"-\w+", text1))
    set2 = set(re.findall(r"-\w+", text2))
    if set1 == set2:
        return True
    print(sorted(list(set1)))
    print(sorted(list(set2)))
    return False


if __name__ == "__main__":
    pass