403Webshell
Server IP : 66.29.132.122  /  Your IP : 3.145.90.207
Web Server : LiteSpeed
System : Linux business142.web-hosting.com 4.18.0-553.lve.el8.x86_64 #1 SMP Mon May 27 15:27:34 UTC 2024 x86_64
User : admazpex ( 531)
PHP Version : 7.2.34
Disable Function : NONE
MySQL : OFF  |  cURL : ON  |  WGET : ON  |  Perl : ON  |  Python : ON  |  Sudo : OFF  |  Pkexec : OFF
Directory :  /proc/self/root/proc/self/root/proc/thread-self/root/proc/thread-self/root/opt/cloudlinux/venv/lib/python3.11/site-packages/guppy/gsl/

Upload File :
current_dir [ Writeable ] document_root [ Writeable ]

 

Command :


[ Back ]     

Current File : /proc/self/root/proc/self/root/proc/thread-self/root/proc/thread-self/root/opt/cloudlinux/venv/lib/python3.11/site-packages/guppy/gsl/Latex.py
class Doc2Latex:
    sizes = ('tiny', 'scriptsize', 'footnotesize', 'small',
             'normalsize', 'large', 'Large', 'LARGE', 'huge', 'Huge')

    def __init__(self, mod, doc, node):
        self.mod = mod
        self.doc = doc
        self.encoder = Encoder(mod)
        self.encode = self.encoder.encode
        self.node = node
        self.out = []
        self.ms = []
        self.mode = None
        self.cur_style = 'rm'
        self.cur_size = list(self.sizes).index('normalsize')
        self.document_lang = None
        self.document_title = None
        self.document_metas = []
        self.latex_list_nesting = 0
        self.latex_mode = 0
        self.noindent = 0
        self.authors = []

        node.accept(self)

    def _visit_children(self, node):
        E = self.mod.ReportedError
        for ch in node.children:
            try:
                ch.accept(self)
            except E:
                pass

    def abs_size(self, size, node):
        osize = self.cur_size
        si = size
        if si < 0:
            si = 0
        elif si >= len(self.sizes):
            si = len(self.sizes) - 1
        self.append('{\\%s ' % self.sizes[si])
        self.cur_size = si
        if self.cur_style != 'rm':
            self.style(self.cur_style, node)
        else:
            node.arg_accept(self)
        self.append('}')
        self.cur_size = osize

    def append(self, x):
        self.out.append(x)

    def changed_size(self, delta, node):
        self.abs_size(self.cur_size + delta, node)

    def error(self, msg, *args, **kwds):
        msg = 'Doc2Latex: ' + msg
        self.doc.env.error(msg, *args, **kwds)

    def get_latex(self):
        return ''.join(self.out)

    def no_children(self, node):
        if node.children:
            self.error('No children allowed for %r.' %
                       node.tag, node.children[0])

    def style(self, style, node):
        self.append('{\\%s ' % style)
        ostyle = self.cur_style
        self.cur_style = style
        node.arg_accept(self)
        self.cur_style = ostyle
        if style == 'em':
            self.append('\\/}')
        else:
            self.append('\\/}')

    def visit_a(self, node):
        pass

    def visit_author(self, node):
        self.authors.append(node.arg)
        self.no_children(node)

    def visit_big(self, node):
        self.changed_size(1, node)

    def visit_block(self, node):
        self._visit_children(node)

    def visit_blockquote(self, node):
        self.append('\\begin{quote}\n')
        self.latex_list_nesting += 1

        node.arg_accept(self)

        self.latex_list_nesting -= 1
        self.append('\\end{quote}\n')

    char_table = {
        'nbsp': '~',
    }

    def visit_char(self, node):
        char = node.arg.strip()
        c = self.char_table.get(char)
        if c is None:
            self.error('No such character: %r.' % char, node)
            c = char
        self.append(c)

    def visit_code(self, node):
        self.style('tt', node)

    def visit_comment(self, node):
        pass

    def visit_dd(self, node):
        self.ms.append('dd')
        step = 24
        ls = (self.ms.count('dd') + self.latex_list_nesting) * step
        self.append('{\\par \\noindent  \\leftskip = %d pt ' % ls)
        for i, v in enumerate(('i', 'ii', 'iii', 'iv', 'v', 'vi')[self.latex_list_nesting:]):
            self.append(' \\leftmargin%s = %d pt ' % (v, ls + (i + 1) * step))
        node.arg_accept(self)
        self.append('\\par}\n')
        self.ms.pop()

    def visit_default(self, node):
        self.error('I don\'t know what to generate for the tag %r.' %
                   node.tag, node)

    def visit_define(self, node):
        # xxx
        self._visit_children(node)

    def visit_dl(self, node):
        if self.ms and self.ms[-1] == 'dt':
            self.visit_dd(node)
        else:
            self.append('{\\par \\noindent\n')
            self._visit_children(node)
            self.append('\\par}\n')

    def visit_dt(self, node):
        self.ms.append('dt')
        self.append('{\\par \\pagebreak[%f] \\noindent \\hangindent = 12 pt \\hangafter = 1 \n' % (
                    3.4-0.1*len(self.ms),
                    ))
        node.arg_accept(self)
        self.append('\\par}\n')
        self.ms.pop()

    def visit_document(self, node):
        self._visit_children(node)

    def visit_document_lang(self, node):
        if self.document_lang is not None:
            self.error('Duplicate document lang directive.', node)
        self.document_lang = node

    def visit_document_title(self, node):
        if self.document_title is not None:
            self.error('Duplicate document title directive.', node)
        self.document_title = node

    def visit_exdefs(self, node):
        self.symplace = {}
        for ch in node.children:
            syms = [x.strip() for x in ch.arg.split(',')]
            for sym in syms:
                self.symplace[sym] = ch.tag

    def visit_em(self, node):
        self.style('em', node)

    def visit_enumerate(self, node):
        self.append('\\begin{enumerate}\n')
        for c in node.children:
            self.append('\\item ')
            c.accept(self)
        self.append('\\end{enumerate}\n')

    def visit_h0(self, node):
        # Not a html header,
        # we may treat this as 'new page' or chapter here
        # and some larger divisor in html.
        self.visit_hx(node)

    def visit_h1(self, node):
        self.visit_hx(node)

    def visit_h2(self, node):
        self.visit_hx(node)

    def visit_h3(self, node):
        self.visit_hx(node)

    def visit_h4(self, node):
        self.visit_hx(node)

    def visit_h5(self, node):
        self.visit_hx(node)

    def visit_h6(self, node):
        self.visit_hx(node)

    def visit_hx(self, node):
        n = int(node.tag[1:])
        if self.mode == 'man_page':
            self.append('{\\par \\pagebreak[%d] \\vskip %d pt \\noindent\n' % (
                [4, 3, 3, 2, 2, 1, 1][n],
                (12 - 2 * n)))
            self.abs_size(len(self.sizes) - n - 2, self.mod.node_of_taci(
                '', '',
                [self.mod.node_of_taci('strong', node.arg, node.children)]))
            self.append('\\par \\vskip %d pt\n} \\noindent\n' % (12 - 2 * n))
            self.noindent = 1
            # self.append('\\end{list}\n')
        else:
            self.append('\\%s{' % self.mod.section_table[n])
            node.arg_accept(self)
            self.append('}\n')

    def visit_itemize(self, node):
        self.append('\\begin{itemize}\n')
        self.latex_list_nesting += 1
        for c in node.children:
            self.append('\\item ')
            c.accept(self)

        self.latex_list_nesting -= 1
        self.append('\\end{itemize}\n')

    def visit_latex(self, node):
        self.latex_mode += 1
        node.arg_accept(self)
        self.latex_mode -= 1

    def visit_li(self, node):
        self.append('\\item ')
        node.arg_accept(self)

    def visit_link_to(self, node):
        # xxx
        name = node.arg
        self.append(' {\\em ')
        if not node.children:
            self.append(self.encode(name))
        else:
            self._visit_children(node)
        self.append('\\/}')

    def visit_link_to_extern(self, node):
        # xxx
        name = node.arg
        doc = node.children[0].arg
        children = node.children[1:]
        self.append(' {\\em ')
        if not children:
            self.append(self.encode(name))
        else:
            for ch in children:
                ch.accept(self)
        self.append('\\/}')

    def visit_link_to_local(self, node):
        # xxx
        name = node.arg
        self.append(' {\\em ')
        if not node.children:
            self.append(self.encode(name))
        else:
            self._visit_children(node)
        self.append('\\/}')

    def visit_link_to_unresolved(self, node):
        # xxx
        name = node.arg
        self.append(' {\\em ')
        if not node.children:
            self.append(self.encode(name))
        else:
            self._visit_children(node)
        self.append('\\/}')

    def visit_literal_block(self, node):
        self.append('{\\ttfamily \\raggedright \\noindent')
        self.encoder.literal_block = 1
        self.encoder.insert_none_breaking_blanks = 1

        node.arg_accept(self)
        self.encoder.literal_block = 0
        self.encoder.insert_none_breaking_blanks = 0
        self.append('}\n')

    def visit_lp(self, node):
        self.latex_mode += 1
        self.visit_paragraph(node)
        self.latex_mode -= 1

    def visit_man_page_mode(self, node):
        omode = self.mode
        self.mode = 'man_page'
        self._visit_children(node)
        self.mode = omode

    def visit_meta(self, node):
        self.document_metas.append(node)

    def visit_ol(self, node):
        self.append('\\begin{enumerate}\n')
        self._visit_children(node)
        self.append('\\end{enumerate}\n')

    def visit_p(self, node):
        self.visit_paragraph(node)

    def visit_paragraph(self, node):
        self.append('{\\par ')
        if self.noindent:
            self.append('\\parindent = 0 pt ')
            self.noindent = 0
        self.append('\n')
        node.arg_accept(self)
        self.append(' \\par}\n')

    def visit_pre(self, node):
        # I couldn't use Latex verbatim environment
        # since it didn't respected leftskip
        # so the environment became misplaced (within dd)

        text = node.arg.strip()
        if text:
            text += '\n'
        text = text + node.get_text()
        text = text.expandtabs()
        lines = text.split('\n')
        if lines and not lines[-1]:
            lines.pop()
        if not lines:
            return
        self.append('\\par\n')
        self.encoder.insert_none_breaking_blanks += 1
        self.encoder.literal += 1
        first = 1
        self.append('{\\tt{%s}}\n' % self.encode(lines[0]))
        for line in lines[1:]:
            self.append(
                '{ \\par \\parindent = 0 pt \\parskip = 0 pt \\tt{%s} }\n' %
                self.encode(line))
        self.encoder.insert_none_breaking_blanks -= 1
        self.encoder.literal -= 1
        self.append('\\par\n')

    def visit_small(self, node):
        self.changed_size(-1, node)

    def visit_spc_colonkind(self, node):
        self.append('~{\\bf :} ')

    def visit_spc_mapsto(self, node):
        self.append(' \\(\mapsto \\) ')

    def visit_string(self, node):
        self._visit_children(node)

    def visit_strong(self, node):
        self.style('bf', node)

    def visit_sub(self, node):
        self.append('\\raisebox{-.6ex}{')
        self.changed_size(-1, node)
        self.append('}')

    def visit_sup(self, node):
        self.append('\\raisebox{.6ex}{')
        self.changed_size(-1, node)
        self.append('}')

    def visit_symbol(self, node):
        self.visit_text(node)

    def visit_table(self, node):
        Table(self, node)

    def visit_text(self, node):
        if self.latex_mode:
            self.append(node.arg)
        else:
            text = node.arg
            text = self.encoder.encode(text)
            self.append(text)
        self._visit_children(node)

    def visit_to_document_only(self, node):
        self._visit_children(node)

    def visit_to_html_only(self, node):
        pass

    def visit_to_tester_only(self, node):
        pass

    def visit_tt(self, node):

        self.append('\\texttt{')
        self.encoder.literal = 1
        node.arg_accept(self)
        self.encoder.literal = 0
        self.append('}')

    def visit_ul(self, node):
        self.append('\\begin{itemize}\n')
        self._visit_children(node)
        self.append('\\end{itemize}\n')

    def visit_var(self, node):
        self.style('em', node)


class Table(Doc2Latex):
    many_hlines = 1  # Use extra many hlines.. looks good, a matter of taste.

    def __init__(self, d2l, node):
        self.d2l = d2l
        self.__dict__.update(d2l.__dict__)
        self.node = node
        self.out = []
        self.rows = []
        self.colwidth = None

        self._visit_children(node)

        maxcols = 0
        for row in self.rows:
            if len(row.columns) > maxcols:
                maxcols = len(row.columns)

        if not maxcols:
            return  # Empty table
        if self.colwidth is not None:
            if not len(self.colwidth) == maxcols:
                self.error("Wrong number of column width specifications (%d) vs\n"
                           "    max columns in table (%d)." % (
                               len(self.colwidth), maxcols),
                           node)
        else:
            self.colwidth = [1.0/maxcols]*maxcols
        ap = self.d2l.append
        ap('\n\\begin{longtable}[c]{|%s|}\n' % ('|'.join(['p{%.2g\\linewidth}' % cw
                                                          for cw in self.colwidth])))
        if self.many_hlines:
            ap('\\hline\n')
        for row in self.rows:
            for col in row.columns:
                ap(''.join(col.data))
                if col is row.columns[-1]:
                    if self.many_hlines:
                        ap('\\\\\n')
                        ap('\\hline\n')
                    else:
                        if row is not self.rows[-1]:
                            ap('\\\\\n')
                else:
                    ap('&\n')
            if row.is_head:
                ap('\\hline\n')
                ap('\\endhead\n')
        ap('\n\\end{longtable}\n')

    def visit_colgroup(self, node):
        colwidth = []

        for c in node.children:
            if c.tag != "col_width":
                self.error('Unrecognized colgroup option: %r' % c.tag, c)
            cg = c.arg
            if cg.endswith('%'):
                cg = cg[:-1]
                cg = float(cg)/100.0
            else:
                cg = float(cg)
            colwidth.append(cg)
        self.colwidth = colwidth

    def visit_options(self, node):
        pass

    def visit_thead(self, node):
        self._visit_children(node)
        self.rows[-1].is_head = 1

    def visit_tr(self, node):
        self.rows.append(Row(self, node))


class Row(Doc2Latex):
    is_head = 0

    def __init__(self, table, node):
        self.__dict__.update(table.__dict__)
        self.columns = []
        self._visit_children(node)

    def visit_td(self, node):
        self.columns.append(Column(self, node))

    def visit_th(self, node):
        self.columns.append(Column(self, node))


class Column(Doc2Latex):
    def __init__(self, row, node):
        self.__dict__.update(row.__dict__)
        self.data = []
        self.append = self.data.append
        node.arg_accept(self)


class Babel:
    """Language specifics for LaTeX."""
    # country code by a.schlock.
    # partly manually converted from iso and babel stuff, dialects and some
    _ISO639_TO_BABEL = {
        'no': 'norsk',  # XXX added by hand ( forget about nynorsk?)
        'gd': 'scottish',  # XXX added by hand
        'hu': 'magyar',  # XXX added by hand
        'pt': 'portuguese',  # XXX added by hand
        'sl': 'slovenian',
        'af': 'afrikaans',
        'bg': 'bulgarian',
        'br': 'breton',
        'ca': 'catalan',
        'cs': 'czech',
        'cy': 'welsh',
        'da': 'danish',
        'fr': 'french',
        # french, francais, canadien, acadian
        'de': 'ngerman',  # XXX rather than german
        # ngerman, naustrian, german, germanb, austrian
        'el': 'greek',
        'en': 'english',
        # english, USenglish, american, UKenglish, british, canadian
        'eo': 'esperanto',
        'es': 'spanish',
        'et': 'estonian',
        'eu': 'basque',
        'fi': 'finnish',
        'ga': 'irish',
        'gl': 'galician',
        'he': 'hebrew',
        'hr': 'croatian',
        'hu': 'hungarian',
        'is': 'icelandic',
        'it': 'italian',
        'la': 'latin',
        'nl': 'dutch',
        'pl': 'polish',
        'pt': 'portuguese',
        'ro': 'romanian',
        'ru': 'russian',
        'sk': 'slovak',
        'sr': 'serbian',
        'sv': 'swedish',
        'tr': 'turkish',
        'uk': 'ukrainian'
    }

    def __init__(self, mod):
        self.language = mod.language_code
        self.re = mod.re

        # pdflatex does not produce double quotes for ngerman in tt.
        self.double_quote_replacment = None
        if self.re.search('^de', self.language):
            #self.quotes = ("\"`", "\"'")
            self.quotes = ('{\\glqq}', '{\\grqq}')
            self.double_quote_replacment = "{\\dq}"
        else:
            self.quotes = ("``", "''")
        self.quote_index = 0

    def next_quote(self):
        q = self.quotes[self.quote_index]
        self.quote_index = (self.quote_index+1) % 2
        return q

    def quote_quotes(self, text):
        t = None
        for part in text.split('"'):
            if t == None:
                t = part
            else:
                t += self.next_quote() + part
        return t

    def double_quotes_in_tt(self, text):
        if not self.double_quote_replacment:
            return text
        return text.replace('"', self.double_quote_replacment)

    def get_language(self):
        if self.language in self._ISO639_TO_BABEL:
            return self._ISO639_TO_BABEL[self.language]
        else:
            # support dialects.
            l = self.language.split("_")[0]
            if l in self._ISO639_TO_BABEL:
                return self._ISO639_TO_BABEL[l]
        return None


class Encoder:
    literal_block = 0
    literal = 0
    mathmode = 0
    verbatim = 0
    insert_newline = 0
    mbox_newline = 0
    insert_none_breaking_blanks = 0

    latex_equivalents = {
        '\u00A0': '~',
        '\u2013': '{--}',
        '\u2014': '{---}',
        '\u2018': '`',
        '\u2019': '\'',
        '\u201A': ',',
        '\u201C': '``',
        '\u201D': '\'\'',
        '\u201E': ',,',
        '\u2020': '{\\dag}',
        '\u2021': '{\\ddag}',
        '\u2026': '{\\dots}',
        '\u2122': '{\\texttrademark}',
        '\u21d4': '{$\\Leftrightarrow$}',
    }

    def __init__(self, mod):
        self.mod = mod
        self.re = mod.re
        self.babel = Babel(mod)
        self.font_encoding = mod.font_encoding
        self.latex_encoding = self.to_latex_encoding(mod.output_encoding)

    def to_latex_encoding(self, docutils_encoding):
        """
        Translate docutils encoding name into latex's.

        Default fallback method is remove "-" and "_" chars from docutils_encoding.

        """
        tr = {"iso-8859-1": "latin1",     # west european
              "iso-8859-2": "latin2",     # east european
              "iso-8859-3": "latin3",     # esperanto, maltese
              "iso-8859-4": "latin4",     # north european,scandinavian, baltic
              "iso-8859-5": "iso88595",   # cyrillic (ISO)
              "iso-8859-9": "latin5",     # turkish
              "iso-8859-15": "latin9",    # latin9, update to latin1.
              "mac_cyrillic": "maccyr",   # cyrillic (on Mac)
              "windows-1251": "cp1251",   # cyrillic (on Windows)
              "koi8-r": "koi8-r",         # cyrillic (Russian)
              "koi8-u": "koi8-u",         # cyrillic (Ukrainian)
              "windows-1250": "cp1250",   #
              "windows-1252": "cp1252",   #
              "us-ascii": "ascii",        # ASCII (US)
              # unmatched encodings
              # "": "applemac",
              # "": "ansinew",  # windows 3.1 ansi
              # "": "ascii",    # ASCII encoding for the range 32--127.
              # "": "cp437",    # dos latine us
              # "": "cp850",    # dos latin 1
              # "": "cp852",    # dos latin 2
              # "": "decmulti",
              # "": "latin10",
              # "iso-8859-6": ""   # arabic
              # "iso-8859-7": ""   # greek
              # "iso-8859-8": ""   # hebrew
              # "iso-8859-10": ""   # latin6, more complete iso-8859-4
              }
        if docutils_encoding.lower() in tr:
            return tr[docutils_encoding.lower()]
        return docutils_encoding.translate(str.maketrans("", "", "_-")).lower()

    def unicode_to_latex(self, text):
        # see LaTeX codec
        # http://aspn.activestate.com/ASPN/Cookbook/Python/Recipe/252124
        # Only some special chracters are translated, for documents with many
        # utf-8 chars one should use the LaTeX unicode package.
        for uchar in list(self.latex_equivalents.keys()):
            text = text.replace(uchar, self.latex_equivalents[uchar])
        return text

    def encode(self, text):
        """
        Encode special characters in `text` & return.
            # $ % & ~ _ ^ \ { }
        Escaping with a backslash does not help with backslashes, ~ and ^.

            < > are only available in math-mode or tt font. (really ?)
            $ starts math- mode.
        AND quotes:

        """
        if self.verbatim:
            return text
        # compile the regexps once. do it here so one can see them.
        #
        # first the braces.
        if 'encode_re_braces' not in self.__dict__:
            self.encode_re_braces = self.re.compile(r'([{}])')
        text = self.encode_re_braces.sub(r'{\\\1}', text)
        if 'encode_re_bslash' not in self.__dict__:
            # find backslash: except in the form '{\{}' or '{\}}'.
            self.encode_re_bslash = self.re.compile(r'(?<!{)(\\)(?![{}]})')
        # then the backslash: except in the form from line above:
        # either '{\{}' or '{\}}'.
        text = self.encode_re_bslash.sub(r'{\\textbackslash}', text)

        # then dollar
        text = text.replace("$", '{\\$}')
        if not (self.literal_block or self.literal or self.mathmode):
            # the vertical bar: in mathmode |,\vert or \mid
            #   in textmode \textbar
            text = text.replace("|", '{\\textbar}')
            text = text.replace("<", '{\\textless}')
            text = text.replace(">", '{\\textgreater}')
        # then
        text = text.replace("&", '{\\&}')
        # the ^:
        # * verb|^| does not work in mbox.
        # * mathmode has wedge. hat{~} would also work.
        # text = text.replace("^", '{\\ensuremath{^\\wedge}}')
        text = text.replace("^", '{\\textasciicircum}')
        text = text.replace("%", '{\\%}')
        text = text.replace("#", '{\\#}')
        text = text.replace("~", '{\\textasciitilde}')
        # Separate compound characters, e.g. "--" to "-{}-".  (The
        # actual separation is done later; see below.)
        separate_chars = '-'
        if self.literal_block or self.literal:
            # In monospace-font, we also separate ",,", "``" and "''"
            # and some other characters which can't occur in
            # non-literal text.
            separate_chars += ',`\'"<>'
            # pdflatex does not produce doublequotes for ngerman.
            text = self.babel.double_quotes_in_tt(text)
            if self.font_encoding == 'OT1':
                # We're using OT1 font-encoding and have to replace
                # underscore by underlined blank, because this has
                # correct width.
                text = text.replace('_', '{\\underline{ }}')
                # And the tt-backslash doesn't work in OT1, so we use
                # a mirrored slash.
                text = text.replace('\\textbackslash', '\\reflectbox{/}')
            else:
                text = text.replace('_', '{\\_}')
        else:
            text = self.babel.quote_quotes(text)
            text = text.replace("_", '{\\_}')
        for char in separate_chars * 2:
            # Do it twice ("* 2") becaues otherwise we would replace
            # "---" by "-{}--".
            text = text.replace(char + char, char + '{}' + char)
        if self.insert_newline or self.literal_block:
            # Insert a blank before the newline, to avoid
            # ! LaTeX Error: There's no line here to end.
            text = text.replace("\n", '~\\\\\n')
        elif self.mbox_newline:
            if self.literal_block:
                closings = "}" * len(self.literal_block_stack)
                openings = "".join(self.literal_block_stack)
            else:
                closings = ""
                openings = ""
            text = text.replace(
                "\n", "%s}\\\\\n\\mbox{%s" % (closings, openings))
        # lines starting with "[" give errors.
        text = text.replace('[', '{[}')
        if self.insert_none_breaking_blanks:
            text = text.replace(' ', '~')
        if self.latex_encoding != 'utf8':
            text = self.unicode_to_latex(text)
        return text


class _GLUECLAMP_:
    _imports_ = (
        '_parent:SpecNodes',
        '_parent.SpecNodes:node_of_taci',
        '_parent.Main:ReportedError',
        '_root:re',
    )

    font_encoding = ''
    double_quote_replacment = ''
    language_code = ''
    output_encoding = ''

    section_table = {
        0: 'part',
        1: 'chapter',
        2: 'section',
        3: 'subsection',
        4: 'subsubsection',
        5: 'paragraph',
        6: 'subparagraph'
    }

    def doc2text(self, doc, node):
        d2l = Doc2Latex(self, doc, node)
        return d2l.get_latex()

    def doc2filer(self, doc, node, name, dir, opts, IO):
        text = self.doc2text(doc, node)
        path = IO.path.join(dir, '%s.tex' % name)
        node = self.node_of_taci('write_file', path, [
                                 self.node_of_taci('text', text)])
        return node

Youez - 2016 - github.com/yon3zu
LinuXploit