summaryrefslogtreecommitdiffstats
path: root/cli_helpers/tabular_output/tabulate_adapter.py
diff options
context:
space:
mode:
Diffstat (limited to 'cli_helpers/tabular_output/tabulate_adapter.py')
-rw-r--r--cli_helpers/tabular_output/tabulate_adapter.py212
1 files changed, 212 insertions, 0 deletions
diff --git a/cli_helpers/tabular_output/tabulate_adapter.py b/cli_helpers/tabular_output/tabulate_adapter.py
new file mode 100644
index 0000000..2c557f8
--- /dev/null
+++ b/cli_helpers/tabular_output/tabulate_adapter.py
@@ -0,0 +1,212 @@
+# -*- coding: utf-8 -*-
+"""Format adapter for the tabulate module."""
+
+from __future__ import unicode_literals
+
+from cli_helpers.utils import filter_dict_by_key
+from cli_helpers.compat import Terminal256Formatter, Token, StringIO
+from .preprocessors import (
+ convert_to_string,
+ truncate_string,
+ override_missing_value,
+ style_output,
+ HAS_PYGMENTS,
+ escape_newlines,
+)
+
+import tabulate
+
+
+tabulate.MIN_PADDING = 0
+
+tabulate._table_formats["psql_unicode"] = tabulate.TableFormat(
+ lineabove=tabulate.Line("┌", "─", "┬", "┐"),
+ linebelowheader=tabulate.Line("├", "─", "┼", "┤"),
+ linebetweenrows=None,
+ linebelow=tabulate.Line("└", "─", "┴", "┘"),
+ headerrow=tabulate.DataRow("│", "│", "│"),
+ datarow=tabulate.DataRow("│", "│", "│"),
+ padding=1,
+ with_header_hide=None,
+)
+
+tabulate._table_formats["double"] = tabulate.TableFormat(
+ lineabove=tabulate.Line("╔", "═", "╦", "╗"),
+ linebelowheader=tabulate.Line("╠", "═", "╬", "╣"),
+ linebetweenrows=None,
+ linebelow=tabulate.Line("╚", "═", "╩", "╝"),
+ headerrow=tabulate.DataRow("║", "║", "║"),
+ datarow=tabulate.DataRow("║", "║", "║"),
+ padding=1,
+ with_header_hide=None,
+)
+
+tabulate._table_formats["ascii"] = tabulate.TableFormat(
+ lineabove=tabulate.Line("+", "-", "+", "+"),
+ linebelowheader=tabulate.Line("+", "-", "+", "+"),
+ linebetweenrows=None,
+ linebelow=tabulate.Line("+", "-", "+", "+"),
+ headerrow=tabulate.DataRow("|", "|", "|"),
+ datarow=tabulate.DataRow("|", "|", "|"),
+ padding=1,
+ with_header_hide=None,
+)
+
+tabulate._table_formats["ascii_escaped"] = tabulate.TableFormat(
+ lineabove=tabulate.Line("+", "-", "+", "+"),
+ linebelowheader=tabulate.Line("+", "-", "+", "+"),
+ linebetweenrows=None,
+ linebelow=tabulate.Line("+", "-", "+", "+"),
+ headerrow=tabulate.DataRow("|", "|", "|"),
+ datarow=tabulate.DataRow("|", "|", "|"),
+ padding=1,
+ with_header_hide=None,
+)
+
+# "minimal" is the same as "plain", but without headers
+tabulate._table_formats["minimal"] = tabulate._table_formats["plain"]
+
+tabulate.multiline_formats["psql_unicode"] = "psql_unicode"
+tabulate.multiline_formats["double"] = "double"
+tabulate.multiline_formats["ascii"] = "ascii"
+tabulate.multiline_formats["minimal"] = "minimal"
+
+supported_markup_formats = (
+ "mediawiki",
+ "html",
+ "latex",
+ "latex_booktabs",
+ "textile",
+ "moinmoin",
+ "jira",
+)
+supported_table_formats = (
+ "ascii",
+ "ascii_escaped",
+ "plain",
+ "simple",
+ "minimal",
+ "grid",
+ "fancy_grid",
+ "pipe",
+ "orgtbl",
+ "psql",
+ "psql_unicode",
+ "rst",
+ "github",
+ "double",
+)
+
+supported_formats = supported_markup_formats + supported_table_formats
+
+default_kwargs = {
+ "ascii": {"numalign": "left"},
+ "ascii_escaped": {"numalign": "left"},
+}
+headless_formats = ("minimal",)
+
+
+def get_preprocessors(format_name):
+ common_formatters = (
+ override_missing_value,
+ convert_to_string,
+ truncate_string,
+ style_output,
+ )
+
+ if tabulate.multiline_formats.get(format_name):
+ return common_formatters + (style_output_table(format_name),)
+ else:
+ return common_formatters + (escape_newlines, style_output_table(format_name))
+
+
+def style_output_table(format_name=""):
+ def style_output(
+ data,
+ headers,
+ style=None,
+ table_separator_token=Token.Output.TableSeparator,
+ **_,
+ ):
+ """Style the *table* a(e.g. bold, italic, and colors)
+
+ .. NOTE::
+ This requires the `Pygments <http://pygments.org/>`_ library to
+ be installed. You can install it with CLI Helpers as an extra::
+ $ pip install cli_helpers[styles]
+
+ Example usage::
+
+ from cli_helpers.tabular_output import tabulate_adapter
+ from pygments.style import Style
+ from pygments.token import Token
+
+ class YourStyle(Style):
+ default_style = ""
+ styles = {
+ Token.Output.TableSeparator: '#ansigray'
+ }
+
+ headers = ('First Name', 'Last Name')
+ data = [['Fred', 'Roberts'], ['George', 'Smith']]
+ style_output_table = tabulate_adapter.style_output_table('psql')
+ style_output_table(data, headers, style=CliStyle)
+
+ data, headers = style_output(data, headers, style=YourStyle)
+ output = tabulate_adapter.adapter(data, headers, style=YourStyle)
+
+ :param iterable data: An :term:`iterable` (e.g. list) of rows.
+ :param iterable headers: The column headers.
+ :param str/pygments.style.Style style: A Pygments style. You can `create
+ your own styles <https://pygments.org/docs/styles#creating-own-styles>`_.
+ :param str table_separator_token: The token type to be used for the table separator.
+ :return: data and headers.
+ :rtype: tuple
+
+ """
+ if style and HAS_PYGMENTS and format_name in supported_table_formats:
+ formatter = Terminal256Formatter(style=style)
+
+ def style_field(token, field):
+ """Get the styled text for a *field* using *token* type."""
+ s = StringIO()
+ formatter.format(((token, field),), s)
+ return s.getvalue()
+
+ def addColorInElt(elt):
+ if not elt:
+ return elt
+ if elt.__class__ == tabulate.Line:
+ return tabulate.Line(
+ *(style_field(table_separator_token, val) for val in elt)
+ )
+ if elt.__class__ == tabulate.DataRow:
+ return tabulate.DataRow(
+ *(style_field(table_separator_token, val) for val in elt)
+ )
+ return elt
+
+ srcfmt = tabulate._table_formats[format_name]
+ newfmt = tabulate.TableFormat(*(addColorInElt(val) for val in srcfmt))
+ tabulate._table_formats[format_name] = newfmt
+
+ return iter(data), headers
+
+ return style_output
+
+
+def adapter(data, headers, table_format=None, preserve_whitespace=False, **kwargs):
+ """Wrap tabulate inside a function for TabularOutputFormatter."""
+ keys = ("floatfmt", "numalign", "stralign", "showindex", "disable_numparse")
+ tkwargs = {"tablefmt": table_format}
+ tkwargs.update(filter_dict_by_key(kwargs, keys))
+
+ if table_format in supported_markup_formats:
+ tkwargs.update(numalign=None, stralign=None)
+
+ tabulate.PRESERVE_WHITESPACE = preserve_whitespace
+
+ tkwargs.update(default_kwargs.get(table_format, {}))
+ if table_format in headless_formats:
+ headers = []
+ return iter(tabulate.tabulate(data, headers, **tkwargs).split("\n"))