mirror of
https://github.com/PiBrewing/craftbeerpi4.git
synced 2024-11-14 02:58:16 +01:00
614 lines
25 KiB
Python
614 lines
25 KiB
Python
import copy
|
|
import itertools
|
|
from functools import partial
|
|
from typing import Iterable, List, Set, Tuple
|
|
|
|
from isort.format import format_simplified
|
|
|
|
from . import parse, sorting, wrap
|
|
from .comments import add_to_line as with_comments
|
|
from .identify import STATEMENT_DECLARATIONS
|
|
from .settings import DEFAULT_CONFIG, Config
|
|
|
|
|
|
def sorted_imports(
|
|
parsed: parse.ParsedContent,
|
|
config: Config = DEFAULT_CONFIG,
|
|
extension: str = "py",
|
|
import_type: str = "import",
|
|
) -> str:
|
|
"""Adds the imports back to the file.
|
|
|
|
(at the index of the first import) sorted alphabetically and split between groups
|
|
|
|
"""
|
|
if parsed.import_index == -1:
|
|
return _output_as_string(parsed.lines_without_imports, parsed.line_separator)
|
|
|
|
formatted_output: List[str] = parsed.lines_without_imports.copy()
|
|
remove_imports = [format_simplified(removal) for removal in config.remove_imports]
|
|
|
|
sections: Iterable[str] = itertools.chain(parsed.sections, config.forced_separate)
|
|
|
|
if config.no_sections:
|
|
parsed.imports["no_sections"] = {"straight": {}, "from": {}}
|
|
base_sections: Tuple[str, ...] = ()
|
|
for section in sections:
|
|
if section == "FUTURE":
|
|
base_sections = ("FUTURE",)
|
|
continue
|
|
parsed.imports["no_sections"]["straight"].update(
|
|
parsed.imports[section].get("straight", {})
|
|
)
|
|
parsed.imports["no_sections"]["from"].update(parsed.imports[section].get("from", {}))
|
|
sections = base_sections + ("no_sections",)
|
|
|
|
output: List[str] = []
|
|
seen_headings: Set[str] = set()
|
|
pending_lines_before = False
|
|
for section in sections:
|
|
straight_modules = parsed.imports[section]["straight"]
|
|
if not config.only_sections:
|
|
straight_modules = sorting.naturally(
|
|
straight_modules,
|
|
key=lambda key: sorting.module_key(
|
|
key, config, section_name=section, straight_import=True
|
|
),
|
|
)
|
|
|
|
from_modules = parsed.imports[section]["from"]
|
|
if not config.only_sections:
|
|
from_modules = sorting.naturally(
|
|
from_modules, key=lambda key: sorting.module_key(key, config, section_name=section)
|
|
)
|
|
|
|
straight_imports = _with_straight_imports(
|
|
parsed, config, straight_modules, section, remove_imports, import_type
|
|
)
|
|
from_imports = _with_from_imports(
|
|
parsed, config, from_modules, section, remove_imports, import_type
|
|
)
|
|
|
|
lines_between = [""] * (
|
|
config.lines_between_types if from_modules and straight_modules else 0
|
|
)
|
|
if config.from_first:
|
|
section_output = from_imports + lines_between + straight_imports
|
|
else:
|
|
section_output = straight_imports + lines_between + from_imports
|
|
|
|
if config.force_sort_within_sections:
|
|
# collapse comments
|
|
comments_above = []
|
|
new_section_output: List[str] = []
|
|
for line in section_output:
|
|
if not line:
|
|
continue
|
|
if line.startswith("#"):
|
|
comments_above.append(line)
|
|
elif comments_above:
|
|
new_section_output.append(_LineWithComments(line, comments_above))
|
|
comments_above = []
|
|
else:
|
|
new_section_output.append(line)
|
|
# only_sections options is not imposed if force_sort_within_sections is True
|
|
new_section_output = sorting.naturally(
|
|
new_section_output,
|
|
key=partial(
|
|
sorting.section_key,
|
|
order_by_type=config.order_by_type,
|
|
force_to_top=config.force_to_top,
|
|
lexicographical=config.lexicographical,
|
|
length_sort=config.length_sort,
|
|
reverse_relative=config.reverse_relative,
|
|
group_by_package=config.group_by_package,
|
|
),
|
|
)
|
|
|
|
# uncollapse comments
|
|
section_output = []
|
|
for line in new_section_output:
|
|
comments = getattr(line, "comments", ())
|
|
if comments:
|
|
section_output.extend(comments)
|
|
section_output.append(str(line))
|
|
|
|
section_name = section
|
|
no_lines_before = section_name in config.no_lines_before
|
|
|
|
if section_output:
|
|
if section_name in parsed.place_imports:
|
|
parsed.place_imports[section_name] = section_output
|
|
continue
|
|
|
|
section_title = config.import_headings.get(section_name.lower(), "")
|
|
if section_title and section_title not in seen_headings:
|
|
if config.dedup_headings:
|
|
seen_headings.add(section_title)
|
|
section_comment = f"# {section_title}"
|
|
if section_comment not in parsed.lines_without_imports[0:1]:
|
|
section_output.insert(0, section_comment)
|
|
|
|
if pending_lines_before or not no_lines_before:
|
|
output += [""] * config.lines_between_sections
|
|
|
|
output += section_output
|
|
|
|
pending_lines_before = False
|
|
else:
|
|
pending_lines_before = pending_lines_before or not no_lines_before
|
|
|
|
if config.ensure_newline_before_comments:
|
|
output = _ensure_newline_before_comment(output)
|
|
|
|
while output and output[-1].strip() == "":
|
|
output.pop() # pragma: no cover
|
|
while output and output[0].strip() == "":
|
|
output.pop(0)
|
|
|
|
if config.formatting_function:
|
|
output = config.formatting_function(
|
|
parsed.line_separator.join(output), extension, config
|
|
).splitlines()
|
|
|
|
output_at = 0
|
|
if parsed.import_index < parsed.original_line_count:
|
|
output_at = parsed.import_index
|
|
formatted_output[output_at:0] = output
|
|
|
|
imports_tail = output_at + len(output)
|
|
while [
|
|
character.strip() for character in formatted_output[imports_tail : imports_tail + 1]
|
|
] == [""]:
|
|
formatted_output.pop(imports_tail)
|
|
|
|
if len(formatted_output) > imports_tail:
|
|
next_construct = ""
|
|
tail = formatted_output[imports_tail:]
|
|
|
|
for index, line in enumerate(tail):
|
|
should_skip, in_quote, *_ = parse.skip_line(
|
|
line,
|
|
in_quote="",
|
|
index=len(formatted_output),
|
|
section_comments=config.section_comments,
|
|
needs_import=False,
|
|
)
|
|
if not should_skip and line.strip():
|
|
if (
|
|
line.strip().startswith("#")
|
|
and len(tail) > (index + 1)
|
|
and tail[index + 1].strip()
|
|
):
|
|
continue
|
|
next_construct = line
|
|
break
|
|
if in_quote:
|
|
next_construct = line
|
|
break
|
|
|
|
if config.lines_after_imports != -1:
|
|
formatted_output[imports_tail:0] = ["" for line in range(config.lines_after_imports)]
|
|
elif extension != "pyi" and next_construct.startswith(STATEMENT_DECLARATIONS):
|
|
formatted_output[imports_tail:0] = ["", ""]
|
|
else:
|
|
formatted_output[imports_tail:0] = [""]
|
|
|
|
if parsed.place_imports:
|
|
new_out_lines = []
|
|
for index, line in enumerate(formatted_output):
|
|
new_out_lines.append(line)
|
|
if line in parsed.import_placements:
|
|
new_out_lines.extend(parsed.place_imports[parsed.import_placements[line]])
|
|
if (
|
|
len(formatted_output) <= (index + 1)
|
|
or formatted_output[index + 1].strip() != ""
|
|
):
|
|
new_out_lines.append("")
|
|
formatted_output = new_out_lines
|
|
|
|
return _output_as_string(formatted_output, parsed.line_separator)
|
|
|
|
|
|
def _with_from_imports(
|
|
parsed: parse.ParsedContent,
|
|
config: Config,
|
|
from_modules: Iterable[str],
|
|
section: str,
|
|
remove_imports: List[str],
|
|
import_type: str,
|
|
) -> List[str]:
|
|
output: List[str] = []
|
|
for module in from_modules:
|
|
if module in remove_imports:
|
|
continue
|
|
|
|
import_start = f"from {module} {import_type} "
|
|
from_imports = list(parsed.imports[section]["from"][module])
|
|
if not config.no_inline_sort or (
|
|
config.force_single_line and module not in config.single_line_exclusions
|
|
):
|
|
if not config.only_sections:
|
|
from_imports = sorting.naturally(
|
|
from_imports,
|
|
key=lambda key: sorting.module_key(
|
|
key,
|
|
config,
|
|
True,
|
|
config.force_alphabetical_sort_within_sections,
|
|
section_name=section,
|
|
),
|
|
)
|
|
if remove_imports:
|
|
from_imports = [
|
|
line for line in from_imports if f"{module}.{line}" not in remove_imports
|
|
]
|
|
|
|
sub_modules = [f"{module}.{from_import}" for from_import in from_imports]
|
|
as_imports = {
|
|
from_import: [
|
|
f"{from_import} as {as_module}" for as_module in parsed.as_map["from"][sub_module]
|
|
]
|
|
for from_import, sub_module in zip(from_imports, sub_modules)
|
|
if sub_module in parsed.as_map["from"]
|
|
}
|
|
if config.combine_as_imports and not ("*" in from_imports and config.combine_star):
|
|
if not config.no_inline_sort:
|
|
for as_import in as_imports:
|
|
if not config.only_sections:
|
|
as_imports[as_import] = sorting.naturally(as_imports[as_import])
|
|
for from_import in copy.copy(from_imports):
|
|
if from_import in as_imports:
|
|
idx = from_imports.index(from_import)
|
|
if parsed.imports[section]["from"][module][from_import]:
|
|
from_imports[(idx + 1) : (idx + 1)] = as_imports.pop(from_import)
|
|
else:
|
|
from_imports[idx : (idx + 1)] = as_imports.pop(from_import)
|
|
|
|
only_show_as_imports = False
|
|
comments = parsed.categorized_comments["from"].pop(module, ())
|
|
above_comments = parsed.categorized_comments["above"]["from"].pop(module, None)
|
|
while from_imports:
|
|
if above_comments:
|
|
output.extend(above_comments)
|
|
above_comments = None
|
|
|
|
if "*" in from_imports and config.combine_star:
|
|
import_statement = wrap.line(
|
|
with_comments(
|
|
_with_star_comments(parsed, module, list(comments or ())),
|
|
f"{import_start}*",
|
|
removed=config.ignore_comments,
|
|
comment_prefix=config.comment_prefix,
|
|
),
|
|
parsed.line_separator,
|
|
config,
|
|
)
|
|
from_imports = [
|
|
from_import for from_import in from_imports if from_import in as_imports
|
|
]
|
|
only_show_as_imports = True
|
|
elif config.force_single_line and module not in config.single_line_exclusions:
|
|
import_statement = ""
|
|
while from_imports:
|
|
from_import = from_imports.pop(0)
|
|
single_import_line = with_comments(
|
|
comments,
|
|
import_start + from_import,
|
|
removed=config.ignore_comments,
|
|
comment_prefix=config.comment_prefix,
|
|
)
|
|
comment = (
|
|
parsed.categorized_comments["nested"].get(module, {}).pop(from_import, None)
|
|
)
|
|
if comment:
|
|
single_import_line += (
|
|
f"{comments and ';' or config.comment_prefix} " f"{comment}"
|
|
)
|
|
if from_import in as_imports:
|
|
if (
|
|
parsed.imports[section]["from"][module][from_import]
|
|
and not only_show_as_imports
|
|
):
|
|
output.append(
|
|
wrap.line(single_import_line, parsed.line_separator, config)
|
|
)
|
|
from_comments = parsed.categorized_comments["straight"].get(
|
|
f"{module}.{from_import}"
|
|
)
|
|
|
|
if not config.only_sections:
|
|
output.extend(
|
|
with_comments(
|
|
from_comments,
|
|
wrap.line(
|
|
import_start + as_import, parsed.line_separator, config
|
|
),
|
|
removed=config.ignore_comments,
|
|
comment_prefix=config.comment_prefix,
|
|
)
|
|
for as_import in sorting.naturally(as_imports[from_import])
|
|
)
|
|
|
|
else:
|
|
output.extend(
|
|
with_comments(
|
|
from_comments,
|
|
wrap.line(
|
|
import_start + as_import, parsed.line_separator, config
|
|
),
|
|
removed=config.ignore_comments,
|
|
comment_prefix=config.comment_prefix,
|
|
)
|
|
for as_import in as_imports[from_import]
|
|
)
|
|
else:
|
|
output.append(wrap.line(single_import_line, parsed.line_separator, config))
|
|
comments = None
|
|
else:
|
|
while from_imports and from_imports[0] in as_imports:
|
|
from_import = from_imports.pop(0)
|
|
|
|
if not config.only_sections:
|
|
as_imports[from_import] = sorting.naturally(as_imports[from_import])
|
|
from_comments = (
|
|
parsed.categorized_comments["straight"].get(f"{module}.{from_import}") or []
|
|
)
|
|
if (
|
|
parsed.imports[section]["from"][module][from_import]
|
|
and not only_show_as_imports
|
|
):
|
|
specific_comment = (
|
|
parsed.categorized_comments["nested"]
|
|
.get(module, {})
|
|
.pop(from_import, None)
|
|
)
|
|
if specific_comment:
|
|
from_comments.append(specific_comment)
|
|
output.append(
|
|
wrap.line(
|
|
with_comments(
|
|
from_comments,
|
|
import_start + from_import,
|
|
removed=config.ignore_comments,
|
|
comment_prefix=config.comment_prefix,
|
|
),
|
|
parsed.line_separator,
|
|
config,
|
|
)
|
|
)
|
|
from_comments = []
|
|
|
|
for as_import in as_imports[from_import]:
|
|
specific_comment = (
|
|
parsed.categorized_comments["nested"]
|
|
.get(module, {})
|
|
.pop(as_import, None)
|
|
)
|
|
if specific_comment:
|
|
from_comments.append(specific_comment)
|
|
|
|
output.append(
|
|
wrap.line(
|
|
with_comments(
|
|
from_comments,
|
|
import_start + as_import,
|
|
removed=config.ignore_comments,
|
|
comment_prefix=config.comment_prefix,
|
|
),
|
|
parsed.line_separator,
|
|
config,
|
|
)
|
|
)
|
|
|
|
from_comments = []
|
|
|
|
if "*" in from_imports:
|
|
output.append(
|
|
with_comments(
|
|
_with_star_comments(parsed, module, list(comments or ())),
|
|
f"{import_start}*",
|
|
removed=config.ignore_comments,
|
|
comment_prefix=config.comment_prefix,
|
|
)
|
|
)
|
|
from_imports.remove("*")
|
|
comments = None
|
|
|
|
for from_import in copy.copy(from_imports):
|
|
comment = (
|
|
parsed.categorized_comments["nested"].get(module, {}).pop(from_import, None)
|
|
)
|
|
if comment:
|
|
single_import_line = with_comments(
|
|
comments,
|
|
import_start + from_import,
|
|
removed=config.ignore_comments,
|
|
comment_prefix=config.comment_prefix,
|
|
)
|
|
single_import_line += (
|
|
f"{comments and ';' or config.comment_prefix} " f"{comment}"
|
|
)
|
|
output.append(wrap.line(single_import_line, parsed.line_separator, config))
|
|
from_imports.remove(from_import)
|
|
comments = None
|
|
|
|
from_import_section = []
|
|
while from_imports and (
|
|
from_imports[0] not in as_imports
|
|
or (
|
|
config.combine_as_imports
|
|
and parsed.imports[section]["from"][module][from_import]
|
|
)
|
|
):
|
|
from_import_section.append(from_imports.pop(0))
|
|
if config.combine_as_imports:
|
|
comments = (comments or []) + list(
|
|
parsed.categorized_comments["from"].pop(f"{module}.__combined_as__", ())
|
|
)
|
|
import_statement = with_comments(
|
|
comments,
|
|
import_start + (", ").join(from_import_section),
|
|
removed=config.ignore_comments,
|
|
comment_prefix=config.comment_prefix,
|
|
)
|
|
if not from_import_section:
|
|
import_statement = ""
|
|
|
|
do_multiline_reformat = False
|
|
|
|
force_grid_wrap = config.force_grid_wrap
|
|
if force_grid_wrap and len(from_import_section) >= force_grid_wrap:
|
|
do_multiline_reformat = True
|
|
|
|
if len(import_statement) > config.line_length and len(from_import_section) > 1:
|
|
do_multiline_reformat = True
|
|
|
|
# If line too long AND have imports AND we are
|
|
# NOT using GRID or VERTICAL wrap modes
|
|
if (
|
|
len(import_statement) > config.line_length
|
|
and len(from_import_section) > 0
|
|
and config.multi_line_output
|
|
not in (wrap.Modes.GRID, wrap.Modes.VERTICAL) # type: ignore
|
|
):
|
|
do_multiline_reformat = True
|
|
|
|
if do_multiline_reformat:
|
|
import_statement = wrap.import_statement(
|
|
import_start=import_start,
|
|
from_imports=from_import_section,
|
|
comments=comments,
|
|
line_separator=parsed.line_separator,
|
|
config=config,
|
|
)
|
|
if config.multi_line_output == wrap.Modes.GRID: # type: ignore
|
|
other_import_statement = wrap.import_statement(
|
|
import_start=import_start,
|
|
from_imports=from_import_section,
|
|
comments=comments,
|
|
line_separator=parsed.line_separator,
|
|
config=config,
|
|
multi_line_output=wrap.Modes.VERTICAL_GRID, # type: ignore
|
|
)
|
|
if max(len(x) for x in import_statement.split("\n")) > config.line_length:
|
|
import_statement = other_import_statement
|
|
if not do_multiline_reformat and len(import_statement) > config.line_length:
|
|
import_statement = wrap.line(import_statement, parsed.line_separator, config)
|
|
|
|
if import_statement:
|
|
output.append(import_statement)
|
|
return output
|
|
|
|
|
|
def _with_straight_imports(
|
|
parsed: parse.ParsedContent,
|
|
config: Config,
|
|
straight_modules: Iterable[str],
|
|
section: str,
|
|
remove_imports: List[str],
|
|
import_type: str,
|
|
) -> List[str]:
|
|
output: List[str] = []
|
|
|
|
as_imports = any((module in parsed.as_map["straight"] for module in straight_modules))
|
|
|
|
# combine_straight_imports only works for bare imports, 'as' imports not included
|
|
if config.combine_straight_imports and not as_imports:
|
|
if not straight_modules:
|
|
return []
|
|
|
|
above_comments: List[str] = []
|
|
inline_comments: List[str] = []
|
|
|
|
for module in straight_modules:
|
|
if module in parsed.categorized_comments["above"]["straight"]:
|
|
above_comments.extend(parsed.categorized_comments["above"]["straight"].pop(module))
|
|
if module in parsed.categorized_comments["straight"]:
|
|
inline_comments.extend(parsed.categorized_comments["straight"][module])
|
|
|
|
combined_straight_imports = ", ".join(straight_modules)
|
|
if inline_comments:
|
|
combined_inline_comments = " ".join(inline_comments)
|
|
else:
|
|
combined_inline_comments = ""
|
|
|
|
output.extend(above_comments)
|
|
|
|
if combined_inline_comments:
|
|
output.append(
|
|
f"{import_type} {combined_straight_imports} # {combined_inline_comments}"
|
|
)
|
|
else:
|
|
output.append(f"{import_type} {combined_straight_imports}")
|
|
|
|
return output
|
|
|
|
for module in straight_modules:
|
|
if module in remove_imports:
|
|
continue
|
|
|
|
import_definition = []
|
|
if module in parsed.as_map["straight"]:
|
|
if parsed.imports[section]["straight"][module]:
|
|
import_definition.append(f"{import_type} {module}")
|
|
import_definition.extend(
|
|
f"{import_type} {module} as {as_import}"
|
|
for as_import in parsed.as_map["straight"][module]
|
|
)
|
|
else:
|
|
import_definition.append(f"{import_type} {module}")
|
|
|
|
comments_above = parsed.categorized_comments["above"]["straight"].pop(module, None)
|
|
if comments_above:
|
|
output.extend(comments_above)
|
|
output.extend(
|
|
with_comments(
|
|
parsed.categorized_comments["straight"].get(module),
|
|
idef,
|
|
removed=config.ignore_comments,
|
|
comment_prefix=config.comment_prefix,
|
|
)
|
|
for idef in import_definition
|
|
)
|
|
|
|
return output
|
|
|
|
|
|
def _output_as_string(lines: List[str], line_separator: str) -> str:
|
|
return line_separator.join(_normalize_empty_lines(lines))
|
|
|
|
|
|
def _normalize_empty_lines(lines: List[str]) -> List[str]:
|
|
while lines and lines[-1].strip() == "":
|
|
lines.pop(-1)
|
|
|
|
lines.append("")
|
|
return lines
|
|
|
|
|
|
class _LineWithComments(str):
|
|
def __new__(cls, value, comments):
|
|
instance = super().__new__(cls, value) # type: ignore
|
|
instance.comments = comments
|
|
return instance
|
|
|
|
|
|
def _ensure_newline_before_comment(output):
|
|
new_output: List[str] = []
|
|
|
|
def is_comment(line):
|
|
return line and line.startswith("#")
|
|
|
|
for line, prev_line in zip(output, [None] + output):
|
|
if is_comment(line) and prev_line != "" and not is_comment(prev_line):
|
|
new_output.append("")
|
|
new_output.append(line)
|
|
return new_output
|
|
|
|
|
|
def _with_star_comments(parsed: parse.ParsedContent, module: str, comments: List[str]) -> List[str]:
|
|
star_comment = parsed.categorized_comments["nested"].get(module, {}).pop("*", None)
|
|
if star_comment:
|
|
return comments + [star_comment]
|
|
return comments
|