Source code for spack.relocate_text

# Copyright Spack Project Developers. See COPYRIGHT file for details.
#
# SPDX-License-Identifier: (Apache-2.0 OR MIT)

"""This module contains pure-Python classes and functions for replacing
paths inside text files and binaries."""

import re
from typing import IO, Dict, Iterable, List, Union

from llnl.util.lang import PatternBytes

import spack.error

Prefix = Union[str, bytes]
PrefixToPrefix = Union[Dict[str, str], Dict[bytes, bytes]]


[docs] def encode_path(p: Prefix) -> bytes: return p if isinstance(p, bytes) else p.encode("utf-8")
def _prefix_to_prefix_as_bytes(prefix_to_prefix: PrefixToPrefix) -> Dict[bytes, bytes]: return {encode_path(k): encode_path(v) for (k, v) in prefix_to_prefix.items()}
[docs] def utf8_path_to_binary_regex(prefix: str) -> PatternBytes: """Create a binary regex that matches the input path in utf8""" prefix_bytes = re.escape(prefix).encode("utf-8") return re.compile(b"(?<![\\w\\-_/])([\\w\\-_]*?)%s([\\w\\-_/]*)" % prefix_bytes)
def _byte_strings_to_single_binary_regex(prefixes: Iterable[bytes]) -> PatternBytes: all_prefixes = b"|".join(re.escape(p) for p in prefixes) return re.compile(b"(?<![\\w\\-_/])([\\w\\-_]*?)(%s)([\\w\\-_/]*)" % all_prefixes)
[docs] def utf8_paths_to_single_binary_regex(prefixes: Iterable[str]) -> PatternBytes: """Create a (binary) regex that matches any input path in utf8""" return _byte_strings_to_single_binary_regex(p.encode("utf-8") for p in prefixes)
[docs] def filter_identity_mappings(prefix_to_prefix: Dict[bytes, bytes]) -> Dict[bytes, bytes]: """Drop mappings that are not changed.""" # NOTE: we don't guard against the following case: # [/abc/def -> /abc/def, /abc -> /x] *will* be simplified to # [/abc -> /x], meaning that after this simplification /abc/def will be # mapped to /x/def instead of /abc/def. This should not be a problem. return {k: v for k, v in prefix_to_prefix.items() if k != v}
[docs] class PrefixReplacer: """Base class for applying a prefix to prefix map to a list of binaries or text files. Derived classes implement _apply_to_file to do the actual work, which is different when it comes to binaries and text files.""" def __init__(self, prefix_to_prefix: Dict[bytes, bytes]) -> None: """ Arguments: prefix_to_prefix: An ordered mapping from prefix to prefix. The order is relevant to support substring fallbacks, for example ``[("/first/sub", "/x"), ("/first", "/y")]`` will ensure /first/sub is matched and replaced before /first. """ self.prefix_to_prefix = filter_identity_mappings(prefix_to_prefix) @property def is_noop(self) -> bool: """Returns true when the prefix to prefix map is mapping everything to the same location (identity) or there are no prefixes to replace.""" return not self.prefix_to_prefix
[docs] def apply(self, filenames: Iterable[str]) -> List[str]: """Returns a list of files that were modified""" changed_files = [] if self.is_noop: return [] for filename in filenames: if self.apply_to_filename(filename): changed_files.append(filename) return changed_files
[docs] def apply_to_filename(self, filename: str) -> bool: if self.is_noop: return False with open(filename, "rb+") as f: return self.apply_to_file(f)
[docs] def apply_to_file(self, f: IO[bytes]) -> bool: if self.is_noop: return False return self._apply_to_file(f)
def _apply_to_file(self, f: IO) -> bool: raise NotImplementedError("Derived classes must implement this method")
[docs] class TextFilePrefixReplacer(PrefixReplacer): """This class applies prefix to prefix mappings for relocation on text files. Note that UTF-8 encoding is assumed.""" def __init__(self, prefix_to_prefix: Dict[bytes, bytes]): """ prefix_to_prefix (OrderedDict): OrderedDictionary where the keys are bytes representing the old prefixes and the values are the new. """ super().__init__(prefix_to_prefix) # Single regex for all paths. self.regex = _byte_strings_to_single_binary_regex(self.prefix_to_prefix.keys())
[docs] @classmethod def from_strings_or_bytes(cls, prefix_to_prefix: PrefixToPrefix) -> "TextFilePrefixReplacer": """Create a TextFilePrefixReplacer from an ordered prefix to prefix map.""" return cls(_prefix_to_prefix_as_bytes(prefix_to_prefix))
def _apply_to_file(self, f: IO) -> bool: """Text replacement implementation simply reads the entire file in memory and applies the combined regex.""" replacement = lambda m: m.group(1) + self.prefix_to_prefix[m.group(2)] + m.group(3) data = f.read() new_data = re.sub(self.regex, replacement, data) if id(data) == id(new_data): return False f.seek(0) f.write(new_data) f.truncate() return True
[docs] class BinaryFilePrefixReplacer(PrefixReplacer): def __init__(self, prefix_to_prefix: Dict[bytes, bytes], suffix_safety_size: int = 7) -> None: """ prefix_to_prefix: Ordered dictionary where the keys are bytes representing the old prefixes and the values are the new suffix_safety_size: in case of null terminated strings, what size of the suffix should remain to avoid aliasing issues? """ assert suffix_safety_size >= 0 super().__init__(prefix_to_prefix) self.suffix_safety_size = suffix_safety_size self.regex = self.binary_text_regex(self.prefix_to_prefix.keys(), suffix_safety_size)
[docs] @classmethod def binary_text_regex( cls, binary_prefixes: Iterable[bytes], suffix_safety_size: int = 7 ) -> PatternBytes: """Create a regex that looks for exact matches of prefixes, and also tries to match a C-string type null terminator in a small lookahead window. Arguments: binary_prefixes: Iterable of byte strings of prefixes to match suffix_safety_size: Sizeof the lookahed for null-terminated string. """ # Note: it's important not to use capture groups for the prefix, since it destroys # performance due to common prefix optimization. return re.compile( b"(" + b"|".join(re.escape(p) for p in binary_prefixes) + b")([^\0]{0,%d}\0)?" % suffix_safety_size )
[docs] @classmethod def from_strings_or_bytes( cls, prefix_to_prefix: PrefixToPrefix, suffix_safety_size: int = 7 ) -> "BinaryFilePrefixReplacer": """Create a BinaryFilePrefixReplacer from an ordered prefix to prefix map. Arguments: prefix_to_prefix: Ordered mapping of prefix to prefix. suffix_safety_size: Number of bytes to retain at the end of a C-string to avoid binary string-aliasing issues. """ return cls(_prefix_to_prefix_as_bytes(prefix_to_prefix), suffix_safety_size)
def _apply_to_file(self, f: IO[bytes]) -> bool: """ Given a file opened in rb+ mode, apply the string replacements as specified by an ordered dictionary of prefix to prefix mappings. This method takes special care of null-terminated C-strings. C-string constants are problematic because compilers and linkers optimize readonly strings for space by aliasing those that share a common suffix (only suffix since all of them are null terminated). See https://github.com/spack/spack/pull/31739 and https://github.com/spack/spack/pull/32253 for details. Our logic matches the original prefix with a ``suffix_safety_size + 1`` lookahead for null bytes. If no null terminator is found, we simply pad with leading /, assuming that it's a long C-string; the full C-string after replacement has a large suffix in common with its original value. If there *is* a null terminator we can do the same as long as the replacement has a sufficiently long common suffix with the original prefix. As a last resort when the replacement does not have a long enough common suffix, we can try to shorten the string, but this only works if the new length is sufficiently short (typically the case when going from large padding -> normal path) If the replacement string is longer, or all of the above fails, we error out. Arguments: f: file opened in rb+ mode Returns: bool: True if file was modified """ assert f.tell() == 0 # We *could* read binary data in chunks to avoid loading all in memory, but it's nasty to # deal with matches across boundaries, so let's stick to something simple. modified = False for match in self.regex.finditer(f.read()): # The matching prefix (old) and its replacement (new) old = match.group(1) new = self.prefix_to_prefix[old] # Did we find a trailing null within a N + 1 bytes window after the prefix? null_terminated = match.end(0) > match.end(1) # Suffix string length, excluding the null byte. Only makes sense if null_terminated suffix_strlen = match.end(0) - match.end(1) - 1 # How many bytes are we shrinking our string? bytes_shorter = len(old) - len(new) # We can't make strings larger. if bytes_shorter < 0: raise CannotGrowString(old, new) # If we don't know whether this is a null terminated C-string (we're looking only N + 1 # bytes ahead), or if it is and we have a common suffix, we can simply pad with leading # dir separators. elif ( not null_terminated or suffix_strlen >= self.suffix_safety_size # == is enough, but let's be defensive or old[-self.suffix_safety_size + suffix_strlen :] == new[-self.suffix_safety_size + suffix_strlen :] ): replacement = b"/" * bytes_shorter + new # If it *was* null terminated, all that matters is that we can leave N bytes of old # suffix in place. Note that > is required since we also insert an additional null # terminator. elif bytes_shorter > self.suffix_safety_size: replacement = new + match.group(2) # includes the trailing null # Otherwise... we can't :( else: raise CannotShrinkCString(old, new, match.group()[:-1]) f.seek(match.start()) f.write(replacement) modified = True return modified
[docs] class BinaryTextReplaceError(spack.error.SpackError): def __init__(self, msg): msg += ( " To fix this, compile with more padding " "(config:install_tree:padded_length), or install to a shorter prefix." ) super().__init__(msg)
[docs] class CannotGrowString(BinaryTextReplaceError): def __init__(self, old, new): return super().__init__( f"Cannot replace {old!r} with {new!r} because the new prefix is longer." )
[docs] class CannotShrinkCString(BinaryTextReplaceError): def __init__(self, old, new, full_old_string): # Just interpolate binary string to not risk issues with invalid unicode, which would be # really bad user experience: error in error. We have no clue if we actually deal with a # real C-string nor what encoding it has. super().__init__( f"Cannot replace {old!r} with {new!r} in the C-string {full_old_string!r}." )