CHANGELOG.md

Binary Refinery Changelog

Note

Almost every release contains bugfixes, but these are not usually included in the changelog. If a release contains only bugfix, it is marked as a 'bugfix release'. Otherwise, the changelog entries highlight only new or changed functionality.

Version 0.10.11

• The entire scripting deobfuscation system was rewritten from scratch. The old deobfuscation units (deob-ps1-*, deob-vba-*, deob-js-*) have been replaced by full parsers and emulators for PowerShell, VBA, and JavaScript. The new units are ps1, vba, js, and the universal defu which auto-detects the language.
• The new cmdarg unit extracts and unescapes arguments passed to wmic, cmd, start, or powershell commands.
• The xtnsis unit now includes a script decompiler that outputs human-readable NSIS source.
• The swf unit was added for extracting resources from Flash SWF files.
• The base unit is now purely block-based (RFC 4648 style). Big-integer encoding has been split into a new bigint unit.
• The escvb unit and the corresponding vbastr pattern for carve now support concatenations of quoted string literals and vb constants.
• Several compression libraries (aplib, blz, lzjb, lzw, xpress, mscf) were cythonized for improved performance.

Version 0.10.10

• Improves xtrar performance by cythonizing parts of the code.
• Fixes a regression where xor file would no longer read the contents of a file named file from disk.

Version 0.10.9

• Improved help text descriptions for carve and xtp exposing named patterns with description.

Version 0.10.8 - bugfix release

Version 0.10.7 - bugfix release

Version 0.10.6

• The modular dependency system was simplified; there are now simply the options default, extended, and all on top of base refinery.
• The unq unit for restoring AV-quarantined files was added. It currently has near zero test coverage and is based entirely on unquarantining lore.

Version 0.10.5

The release contains the new -g switch in binref for AI agents.

Version 0.10.4

This is a test release for enabling trusted publishing with PyPi.

Version 0.10.3

• The xtsf unit was extended with support for SetupFactory 10.
• The hostname regex pattern was renamed to host.
• The py7zr dependency was eliminated and refinery's 7zip unpacker now supports all archives.
• The png unit was added for extracting PNG data chunks.
• The xlmdeobf backend code was inlined as a third party module that ships with refinery, while swapping out its recursive dependencies for what we already have: This further trims refinery's dependency tree.
• A --brief (-b) mode was added to the binref command. This is primarily intended as a reference document to get an overview of all units.
• This command together with a newly authored SKILL.md file are the first attempts to teach language model based AI agents to use binary refinery.
• Compatibility with PowerShell was increased: Previously, complex framing syntax at the end of a unit would have to be quoted in PowerShell to work, this is no longer the case.

Version 0.10.2

• Adds the xtsf unit for extracting SetupFactory binaries.
• Adds the cbor unit for parsing the CBOR format.
• Adds the asn1 unit for generic ASN.1 structure parsing.
• The argon2i unit was renamed to argon2 and supports all variants of Argon2 now.
• Adds the xtrar unit for extracting RAR archives.
• Adds the xtdmg unit for extracting DMG volumes.
• Adds carve-elf, carve-pdf,
• Adds the a3xs unit for extracting only the script part of an AutoIt3 binary.
• Adds the sm3, aria, twofish, scrypt, and simon cryptographic units.
• Adds the plist, puny, and qp units for various common encodings.
• Adds the dsstore unit for parsing MacOS' .DS_Store files.
• The mscf unit now also supports LZMS.
• Adds dncode unit for extracting the MSIL code from functions in a .NET assembly.
• The xtxs unit was renamed to xtmdb and is now supported via xt.
• The lnk unit now supports URL shortcuts much better.
• The xtvba unit was renamed to vbamc to better align with vbapc.

A large number of external dependencies was eliminated from the code base, replacing them with internal parsers. This includes:

• asn1crypto
• LnkParse3
• phpdeserialize
• access-parser
• libzbar
• pycdlib
• olefile
• oletools
• extract-msg

In most cases, this improves the support of those formats with one exception: The qr unit can no longer parse arbitrary bar codes but is restricted to QR codes.

Version 0.10.1 - bugfix release

Version 0.10.0

• Switches -T in the units alu and pemeta were removed to add a new global option named --try: This option allows forwarding the input when a unit fails to process it.
• Adds a proper parser and emulator for Batch files which is exposed via the bat unit. The unit now has some switches to control how emulated commands are synthesized.
• A number of CPU-intensive tasks have been extended with a parallel Cython implementation, specifically PKZIP crypto, AutoIt decryption & decompression, and the 7zip ported code for LZX and deflate.
• Execution performance has been improved by several hundred milliseconds by delaying imports and entirely avoiding unnecessary imports in some cases.
• The idbmeta unit was added to read metadata from older IDA databases.
• The unit carve-tar for carving TAR files was added.
• The unit carve-png for carving PNG files was added.

Version 0.9.26 - bugfix release

Version 0.9.25

• The docmeta unit now exposes more of the embedded metadata of Word documents. . The xj0 unit no longer has size restrictions on extracted strings.

Version 0.9.24

• Adds the jpeg unit for parsing out JPG data streams and some metadata.

Version 0.9.23 - bugfix release

Version 0.9.22 - bugfix release

Version 0.9.21

• The edit unit was added which allows overwriting any part of the input data with a binary string.
• The xtzip unit now supports Deflate64.

Version 0.9.20

• Adds the xtrpa unit for extracting RenPy archives, because Karsten did it and why not.

Version 0.9.19

• The xtinno (and thereby xt) now support Inno Setup up to version 6.6.1.
• The innopwd unit has improved logging and can intercept passwords in more cases.

Version 0.9.18 - bugfix release

Version 0.9.17 - bugfix release

Version 0.9.16 - bugfix release

Version 0.9.15 - bugfix release

Version 0.9.14 - bugfix release

Version 0.9.13

• The xtdmp unit was added for extracting files from Minidumps.
• Refinery was extended by a custom ZIP archive parser which can detect data caves and should support all archives (and probably more) than what was previously supported.

Version 0.9.12 - bugfix release

Version 0.9.11 - bugfix release

Version 0.9.10

• The hlg, hls, and hlb units were added for easier access to different style flavors of syntax highlighting.
• The bruteforce unit now accepts integer intervals as argument rather than Python slices. This means that 1:4 includes 4.
• Various units for FNV hashing were added, i.e. fnv0, fnv1, fnv1a, and shortcuts for various bit sizes.
• This release also includes significant performance improvements to the .NET parser.

Version 0.9.9

• The hl unit was added for highlighting source code on the terminal.
• The xtmsi unit now supports Advanced Installer binaries which include the MSI with obfuscated header.
• The .NET parsing library was completely refactored, so there will certainly be bugs.

Version 0.9.8

• The sqlite unit was added, thanks to @oxitocin.
• The codebook unit was added.
• Several carve unit formats were renamed to shorter versions.

Version 0.9.7

• The d2p unit was added.

Version 0.9.6 - bugfix release

Version 0.9.5

• The carve unit now supports the strarray pattern for arrays of strings, but no decoder for it is implemented yet because it is unclear what that should be.
• The vstack unit has changed to allow for slice-style address arguments, more easily allowing you to specify a stop address.
• The unicorn and icicle-backed emulators in refinery (specifically vstack) now also have rudimentary support for API hooking and logging.
• The binref command now prints the current binary-refinery version.

Version 0.9.4

• The xtmsi unit now exposes several metadata streams that were discarded before, one important example being the digital signature.
• Changes have been made to how the xtxml unit constructs paths into the document structure, in cases where tag names are unique these will now be prioritized.
• The backend for the bat unit was swapped for a new and rudimentary emulator; this will be improved upon in future releases.
• The decompress unit was extended with a new heuristic tier: If any decompression produces output that is recognized as matching a known format, the unit will pick the best result from among all results with this property.
• Type hints across the refinery code base have undergone massive refactoring and should be better compatible with modern type checkers now.

Version 0.9.3 - bugfix release

Version 0.9.2

• The xtchm unit was added for extracting CHM (Windows Help) files.

Version 0.9.1 - bugfix release

Version 0.9.0

• The change to --join-path from v0.7.4 was reverted; this switch for path extraction units no longer alters the joined path based on files on disk. This makes its behavior completely deterministic again. Deconflicting paths is now done by dump: The unit implements the same logic in trying to find compatible paths on disk that do not conflict with existing files.
• The pdf unit has been extended with support for extracting images based on the mupdf library.
• The pdfcrypt unit was added for removing & adding passwords from & to PDF documents.
• There is now a qr unit for decoding barcodes (especially QR codes). It requires the ZBar shared library to work, but that really seems to be the only way it can be done.
• The z85 unit for yet another Base85 encoding scheme was added.
• The flz unit implementing FastLZ was added and is now included in the universal decompressor.
• The xtzip unit now allows extracting archives partially, i.e. a single password-protected file does not completely fail the extraction process. When decryption fails, it is also possible to extract the raw encrypted chunk using the --lenient option.

Version 0.8.28

• Adds the pbuf unit to decode ProtoBuf messages heuristically.

Version 0.8.27 - bugfix release

Version 0.8.26

• The djb2 unit was added for computing the DJB2 hash.
• The mscf decompression algorithms were added to the universal decompressor.

Version 0.8.25

• The xtinno unit now supports Inno Setup up to version 6.4.3.

Version 0.8.24 - bugfix release

Version 0.8.23

• The xt unit now also extracts AutoIt3 samples.
• API tracing via SpeakEasy in vstack can now be switched on and off with a separate switch.

Version 0.8.22

• The cfmt unit was renamed to pf, short for "Print Format".
• The carve-der unit was added.
• The argon2id key derivation unit was added.

Version 0.8.21

• The pecdb and pefix units were added.
• The pkw unit for decompressing PKWare was added.
• In the struct unit, it is now possible to peek a struct entry by specifying an alignment value of zero.

Version 0.8.20

• Adds the --tag and --aad flag to cipher units that support message authentication.

Version 0.8.19

• Thanks to @larsborn, the units dnasm and dnopc were added for disassembling MSIL.
• The iffc unit was added; for filtering chunks in a frame by size constraints.
• The xtxs unit was added; for extracting data from Microsoft Access databases.
• The pym unit for unmarshaling Python data has been improved with a cross-version parser.
• The carve-json unit now defaults to carving only dictionary values.
• The map unit was extended with the "default" parameter.
• The pop unit now supports using a single meta variable as the input source.
• The imgdb and imgtp units for image processing were added, and the transposition option was consequently removed from stego.
• The b2f (back to front) unit was added, a shortcut for pick ::-1.
• The HKDF key derivation unit was renamed to hkdf, in line with all other units now being lowercase.

Version 0.8.18 - bugfix release

Version 0.8.17

This version is functionally equivalent to the previous one, but refinery starts using the LIEF parser with this version. Switching from other executable parses to LIEF was the only change from the last version to this one, see #84.

Version 0.8.16

This is a bit of a botched release, don't use it; use 0.8.17 instead.

Version 0.8.15

• The tea and xtea units now offer the option to specify the number of rounds.
• The couple unit made stdout/stderr merging optional and discards stderr by default.
• The couple unit received the --noinput option to toggle this mode explicitly.
• The standard path formatting for xtxml and xthtml was changed to allow filtering for all elements of a certain tag easier.

Version 0.8.14

• The rtfc unit was added.
• A regular expression pattern named date was added and is now available in xtp.
• The output of the lnk unit was limited to essential information by default.
• The csb and csd shortcuts for carve can now use the --stripspace argument.
• Grouping into blocks in hexload and peek will now add separators in the ASCII preview.

Version 0.8.13 - bugfix release

Version 0.8.12 - bugfix release

Version 0.8.11 - bugfix release

Version 0.8.10 - bugfix release

Version 0.8.9 - bugfix release

Version 0.8.8

• The innopwd unit was added. It can emulate an Inno Setup installer in order to extract passwords that are encoded within the IFPS script.

Version 0.8.7 - bugfix release

Version 0.8.6

• The unit ps1str has been renamed to escps to match it's partner unit esc.
• The unit escvb was added to escape and unescape VB strings.
• An input forward format character was added to rex to support this common use case better.
• The dnfields unit was reworked to extract prettier paths based on the method, type, and namespace.

Version 0.8.5

• Adds the xtsim unit to extract smart install maker archives.
• Adds the lzx unit for LZX decompression.
• Adds LZX support to the xtcab unit.
• The xtcab unit now also suports multi-disk cabinet archives.

Version 0.8.4 - bugfix release

Version 0.8.3

• The lzma unit now supports a lot more (especially custom) LZMA formats.
• The xtinno unit underwent further improvements and can now extract embedded images, as well as decompression and decryption libraries.
• The vstack unit does again operate on unicorn version 2.0.1.post1.

Version 0.8.2

• The xtinno unit was added to extracting files from InnoSetup installers.
• Related; the ifps and ifpsstr units now accept a string encoding argument.
• The jvdasm unit now has colored output.
• Thanks to @s3ven6, the speck cipher and maru hash units were added.
• The dnarrays unit was added for extracting hard-coded arrays from compiled .NET code.
• The iff unit was extended with an -ne switch.
• The xjl unit can now also collect the contents of one frame into a JSON list.

Version 0.8.1 - bugfix release

Version 0.8.0

• The binref command was changed to use conjunctive search logic by default.
• The copy: and cut: multibin handlers now accept arguments of the form [offset]:[length]:[step] instead of [start]:[end]:[step].
• The vstack unit now supports 3 emulator engines: unicorn, speakeasy, icicle. This is somewhat experimental and unicorn remains the default.
• As part of the changes to vstack, the vmemref unit was changed to use smda rather than angr.

Version 0.7.12 - bugfix release

Version 0.7.11 - bugfix release

Version 0.7.10 - bugfix release

Version 0.7.9

• Adds the morse unit for Morse code encoding and decoding.

Version 0.7.8

• The u16 unit is no longer limited to the little-endian variant of UTF-16.
• The snip unit was given a new argument --stream which allows each offset to be relative to the end of the previously extracted data.
• Path extraction units will now match paths case-insensitively when this does not cause ambiguity.
• The xtmsi unit now extracts all MSI tables as CSV on top of the JSON blob.
• The xtnsis unit now extracts setup.bin alongside setup.nsis, the former containing a full binary copy of the extracted header.
• The dedup unit now has an optional argument which can specify a meta variable to deduplicate by.

Version 0.7.7

• Adds the httprequest unit for parsing HTTP requests.
• Adds the b62 unit (thanks to @lukaskuzmiak).
• The uuenc unit was updated to remove reliance on the now deprecated uu module.
• Adds support for aPLib compressed data with headers.

Version 0.7.6

• Adds the brotli decompression unit.
• The pym unit was added which provides an interface to Python's marshal serialization.
• The units xsalsa, xchacha, and chacha20poly1305 were added. The latter only performs the decryption part of the scheme.
• Refinery pipelines used in Python code will now preserve the scope of a Chunk object when one is provided as input.
• The argument handlers prng and rng were added for random number generation.

Version 0.7.5

• The b65536 unit was added (thanks to @alphillips-lab).

Version 0.7.4

• The --join option of all path extraction units has been improved for producing paths that can always be used for dumping data to disk. This includes units to unpack archives, resources, or other embedded data that can be referenced by a name.
• The ef unit has a new option that specifies whether to follow (directory) symlinks / junctions or not.
• The key scaling method for autoxor was adjusted to produce less false positives when scanning for larger keys.
• Thanks to @alphillips-lab, the a3x unit is now capable of decrypting EA05 formatted scripts, previously only EA06 was supported.

Version 0.7.3 - bugfix release

Version 0.7.2

• The loop unit was enhanced with more options to abort execution based on regular expression patterns. It now also offers better control over terminating the execution when an error occurs.
• Conditional units (iff/iffp/iffx/iffs) were reworked to have less magic behavior. The -R switch now controls boolean negation and a separate switch controls whether chunks are hidden instead of being discarded. The -s switch was also removed from conditional units.
• The cull unit was removed from refinery.
• The units p1, p2, and p3 were added, which are shortcuts for picking the first 1, 2, or 3 chunks from a frame, respectively.
• Regular expression arguments now have a new handler f:, which initializes the regular expression entirely from one of the formats used in carve.

Version 0.7.1

• The global --iff option was added to units; this allows you to apply the unit only to formats that it knows it can handle.
• When using refinery in code, it is now possible to pipe a Chunk object directly to a pipeline.
• The csb and csd shortcuts were added for common applications of carve.
• The loop unit was added; it allows repeated application of a multibin suffix to the input data.
• To match the loop unit, the reduce unit now also works with a multibin suffix rather than with a pipeline string.
• The vstack unit now attempts to detect stack cookies and ignores them by default.
• Adds a deobfuscator for the kramer obfuscator.
• The xtmsi unit now automatically extracts embedded CAB files and infers the file names of these subfiles from the MSI manifest.

Version 0.7.0

• Raises minimum Python requirement to 3.8.
• Removes automatic escapes from cfmt; this now has to be done explicitly.
• The rsa unit can now output keys in Microsoft BLOB format.
• Adds the urn unit.
• Adds several multibin handlers to modify file system paths (pp/pb/pn/px).

Version 0.6.43

• Adds the hash units sha3-224, sha3-256, sha3-384, sha3-512, and keccak256.

Version 0.6.42 - bugfix release

Version 0.6.41 - bugfix release

Version 0.6.40 - bugfix release

Version 0.6.39

• Adds the b92 unit for Base92 encoding and decoding.
• Improves the performance of AutoIt3 unpacking in a3x.
• Adds the SymHash field to the machometa unit.

Version 0.6.38 - bugfix release

Version 0.6.37 - bugfix release

Version 0.6.36

• Adds the xtmacho unit which can unpack MachO fat binaries.

Version 0.6.35

• Adds the nrv2b, nrv2d, and nrv2e decompression units.
• Adds the fernet unit to decrypt messages in Fernet format.

Version 0.6.34

• The chop unit has a second argument now that allows to specify the step size. Also, The --into argument has been removed because this can be done more succinctly using the size meta variable and long division.
• The alu unit has been extended with a new helper function called M; it can be used to mask a value down to a certain number of bits.

Version 0.6.33

• The struct unit was extended with an additional format string character, g, for reading GUID values.

Version 0.6.32 - bugfix release

Version 0.6.31 - bugfix release

Version 0.6.30 - bugfix release

Version 0.6.29 - bugfix release

Version 0.6.28

• The reduce signature was changed; it is no longer possible to specify an initialization value, instead the first chunk in the frame is always used. Additionally, there is now an option to consume only a limited number of chunks.
• The queue unit has been removed in favor of two units qf (queue front) and qb (queue back) to queue chunks into the current frame.

Version 0.6.27

• The key derivation units DESDerive, CryptDeriveKey, and PasswordDeriveBytes have been renamed to deskd, mscdk, and mspdb, respectively, in order to match the common refinery unit naming convention of using indecipherable and consonant-heavy abbreviations.
• When passing integer arguments to the units xor, add, and sub, the block size is now automatically adjusted to the smallest size that will contain the given argument.

Version 0.6.26

• Thanks to @EricFaehrmann, xtzip (and xt) now support doubly-loaded ZIP archives.

Version 0.6.25

• Fixes bugs that caused errors in Python 3.12 environments.

Version 0.6.24

• The paths extracted by xthtml, xtxml and xtjson now avoid the use of parentheses to work better on Bash.
• Adds the sosemanuk cipher unit.
• Improves the capabilities of the vbastr unit.

Version 0.6.23 - bugfix release

Version 0.6.22

• The peek unit in --decode mode now truncates long lines by default. Specifying the option twice has the same effect as the previous default, which is to wrap lines.
• The stego unit has been modified to generate a single output by default and provides a switch to generate individual rows or columns.

Version 0.6.21

• Adds the xtzpaq unit to unpack ZPAQ archives.

Version 0.6.20

• Includes the preliminary fix for the PowerShell problem. PowerShell versions 7.4 and beyond support native to native pipelines.

Version 0.6.19

• The b85 unit is now resilient against white space.
• The vsect unit can now extract "synthesized" sections. This also affects vsnip; it can now also extract data from, e.g., the header of an executable based on virtual addresses.
• The possible extras for the binary-refinery Python package have been expanded and the default install has been slimmed even further to avoid having to install too many dependencies for just the core utilities.

Version 0.6.18 - bugfix release

Version 0.6.17

• Adds the opc unit and removes the Angr option from asm.

Version 0.6.16

• The path formatting feature has been isolated in the xthtml and xtxml units.
• The vstack unit no longer extracts byte patches that consist exclusively of zero bytes because these were common false positives.

Version 0.6.15

• The vstack unit has received further improvements. CPU register initialization now works via meta variables instead of shell environment variables, more options have been added, and new heuristics: Values written to the stack that represent addresses into any mapped segment are now ignored by default.
• This release adds the "shell like" interface; by importing units from refinery.shell, they can be instantiated in Python by using string arguments that are interpreted as if the corresponding unit was being assembled from a shell command line.
• The lzw decompression unit was added.
• the xtmagtape unit was added to extract files from SIMH tape files. But why, you ask? It may forever remain a mystery.
• The hc256 cipher unit was added.
• The --more option was added to the struct unit to give access to unparsed rest data.
• The --length option was added to the snip unit as a qualit of life feature.
• The btoi handler can now receive a second argument that allows reading interlaced integers from a byte stream.
• Thanks to @alphillips-lab, the dnsfx unit was added for extracting .NET file bundles.
• The pestrip unit was renamed to pedebloat and petrim was renamed back to pestrip.

Version 0.6.14

• The trim unit can now remove padding and also perform case-insensitive trimming.
• The ngrams and bruteforce unit were added for simple brute forcing tasks.
• The vstack unit can now execute shellcode blobs. It also gained the ability to skip calls entirely, and registers can now be initialized by using shell environment variables.
• The salsa and chacha units can now be initialized with a 64-byte "key" which represents the entire initial state matrix.
• The perc unit now extracts resource languages.
• The yara: handler for regular expression arguments now has the even shorter shortcut Y: because I use it so much.
• A bug was fixed in the url unit which incorrectly decoded when using the --plus switch.

Version 0.6.13 - bugfix release

Version 0.6.12

• Adds the sm4 cipher unit.
• Adds the blabla cipher unit.

Version 0.6.11

• The machometa unit was added thanks to @cxiao.
• The pestrip unit was extended with more features, and the unit petrim was introduced as a unit to simply remove overlays.
• The xtnuitka unit was added to extract Nuitka archives.

Version 0.6.10

• Adds the pyc unit to decompile Python bytecode directly.
• Adds more options to the still quite experimental vstack unit.

Version 0.6.9

• The coloring in peek on Windows is now applied even if peek is not the last unit in the pipeline. This previously caused a bug, but in recent versions the bug was not reproducible.
• The bitsnip unit was added.
• PowerShell deobfuscation was augmented by two units to decode base64.

Version 0.6.8

• The pestrip unit has received some improvements and bugfixes; it should work more reliably now against bloated sections and resources.
• All stream cipher units have been given the --discard option which allows you to discard an arbitrary number of initial bytes from the keystream.
• Call tracing has been removed from vstack; it never really worked in practice and would require a lot more effort to do properly.

Version 0.6.7

• The pack unit can now also pack lists of floating-point numbers.
• The unit chaskey was added to support this cipher; it is used by the Donut framework.

Version 0.6.6

• A minor bug was fixed in pemeta that prevented some signatures from being parsed correctly.
• Archive extraction utilities now escalate fuzziness in 3 stages rather than just 2.

Version 0.6.5

• Slightly improves the script extraction and formatting in xtmsi.

Version 0.6.4

• The xtdoc unit now demangles file names in MSI archives correctly.
• The xtmsi unit was added for extracting MSI files and also stream metadata in a synthesized JSON document.
• The csv unit now has a reverse operation to convert simple JSON documents back to CSV format.
• Thanks to @larsborn, the tnetmtm unit was added for parsing MITMProxy traffic capture files.

Version 0.6.3

• The xtnode unit was added for extracting the contents of Node.js executables created with pkg or nexe.
• The xtzip unit now supports AES-encrypted archives via the pyzipper module.

Version 0.6.2 - bugfix release

Version 0.6.1 - bugfix release

Version 0.6.0

• The AutoIt decompiler unit a3x was added.
• Path extractor units have been reworked to be more consistent about when and when they do not use fuzzy matching on paths. Switches have been added to control this behavior.

Version 0.5.10

• The tea and xtea units now have a --swap switch which allows to switch them from little endian to big endian mode.
• The xxtea unit was re-worked to support being used as a proper block cipher. This is enabled by specifying the block size using the --block-size argument. By default, xxtea will continue to operate on the input as a single block: This is how XXTEA is often used in malicious samples.
• The rc5 and rc6 units have been updated to support the --segment-size option for CFB mode.

Version 0.5.9 - bugfix release

Version 0.5.8 - bugfix release

Version 0.5.7

• The pestrip and peoverlay default settings are the same again.

Version 0.5.6

• The pestrip unit has been extended with the capability to strip bloated resources and sections.
• The xtone unit was added to extract embedded files from OneNote documents.
• The color legend of the iemap unit is now optional and can be enabled with a switch.

Version 0.5.5

• Bugfix to account for changes in macOS libmagic which lead to not correctly identifying exe and dll extensions.
• Importing refinery no longer changes the names of log levels globally.

Version 0.5.4

• The xthtml unit can now extract attributes of HTML tags.
• The rijndael cipher unit was added.

Version 0.5.3 - bugfix release

Version 0.5.2

• The lzg unit was added.
• The lzf unit received several bugfixes and now supports the chunked format produced by the command-line lzf tool.
• The ntlm hash unit was added (thanks to @m0rv4i for the contribution)
• The vmemref and vstack units were added; both are still experimental and not thoroughly tested.
• The min and max units were added to simplify the pattern sorted [| pick 0 ] to a single unit.

Version 0.5.1 - bugfix release

Version 0.5.0

This release changes the way in which meta variables are handled, they now have a scope:

• By default, variables cease to exist when the frame ends in which they were defined.
• Variables remain visible in child frames.
• When a variable is re-defined in a child frame, this definition shadows the previous one: When the child frame ends, the variable is restored to the value it had in the parent frame.
• Some units like pop can propagate variables to the parent scope as well.
• The units mvg and mvc were introduced to manage scoping of variables, the unit wm was removed.

Changes unrelated to meta variable redesign:

• The unit vaddr was added to convert integer meta variables from virtual address to file offset and vice versa.

Version 0.4.49

• The pkcs7sig unit was added.
• The pemeta unit now also displays the module name stored in the export directory.
• The dedup unit now uses MD5 instead of Python's built-in hash function because of the high risk of collisions.

Version 0.4.48

• Block cipher unit backed by pycryptodome (i.e. aes, des, des3) now support additional arguments for some of the less commonly used block cipher modes.
• The rsa and rsakey unit now also support a simple key format of the form [modulus]:[exponent] where both modulus and exponent are hex-encoded numbers in big endian representation for a textbook RSA round.
• The perc unit now has the --pretty option to fix bitmap and icon resources by adding the necessary headers (which are missing from the raw resource data).
• The pcap and pcap-http unit now sort streams by the occurrence of the first packet.

Version 0.4.47

• By default, the ef unit does no longer use glob-patterns on posix systems. The behavior can be explicitly adjusted using new command-line flags.
• Adds the queue unit.
• The names of urlencode patterns for carve were shortened.
• Adds the xtnsis unit to the units used in xt.
• The pemeta unit has improved RICH header data and displays RICH header counts.

Version 0.4.46 - bugfix release

Version 0.4.45 - bugfix release

Version 0.4.44

• Adds the lzf unit for LZF compression and decompression.

Version 0.4.43

• Adds the qlz unit for QuickLZ decompression.

Version 0.4.42 - bugfix release

Version 0.4.41

• Adds the carve-lnk unit to carve Windows Shortcut files.
• Adds the carve-rtf unit to carve RTF documents.
• Adds the subfiles unit which unifies all structured file format carvers.
• The b64 unit now automatically detects and switches to the urlsafe encoding variant.
• Adds the xtnsis unit which can extract files from NSIS archives and provide a rudimentary disassembly of the setup script.
• Adds the ifps and ifpsstr units to disassemble and extract strings from compiled Pascal script files.

Version 0.4.40 - bugfix release

Version 0.4.39

• Adds the vbapc and vbastr units which can extract (decompiled) VBA p-code and VBA strings from (potentially stomped) Word documents.
• Adds the somewhat experimental xkey and autoxor units that can (sometimes) automatically decrypt XOR-encrypted files using frequency analysis. These units are still work in progress, though.
• Adds the mscf unit which implements part of the Microsoft Compression API formats, with LZMS currently missing.
• Adds the b58 unit which does base58 encoding (used to encode Bitcoin addresses, for example). Simultaneously, the base unit was adjusted to no longer strip leading zero bytes unless explicitly instructed to do so.
• Adds variable conversions to pop: It is now possible to prefix a variable with a sequence of multibin handlers to convert input data before storing it in the variable.
• When executing the put unit without a second argument, it now stores the contents of the current chunk in the specified variable.

Version 0.4.38

Fixes a critical bug in the meta variable propagation logic.

Version 0.4.37

• Adds the jcalg unit.
• Adds the byteswap unit.

Version 0.4.36 - bugfix release

Version 0.4.35

• Adds the lzip unit.
• Reworks the serpent unit to work with real-world examples and adds a --swap option to change the block byte order to become compatible with other implementations.
• Changes the peek design and fixes problems with colored output on Windows.

Version 0.4.34 - bugfix release

Version 0.4.33 - bugfix release

Version 0.4.32 - bugfix release

Version 0.4.31

• The (still somewhat experimental) xt unit was added which attempts to extract data from known archive formats.
• The xtasar unit was added which can extract data from ASAR files.
• The lnk unit was added which is a thin wrapper around the LnkParse3 library which extracts metadata from Windows shortcut files.
• The urlfix unit was added which can strip URL indicators of fragments and query strings.
• The iff unit has gained several new features.
• The xjl unit was added, it converts JSON-lists to a sequence of JSON chunks.
• The xvar: handler was renamed to eat: (it is similar to var:, except that it removes the variable after use).
• The xlxtr unit now supports XLSB format by virtue of the pyxlsb2 library.
• The base32 unit was made more robust against invalid paddings.
• The peek unit design was changed yet again and colorization was added to the hexdump preview. It can be disabled through the -g switch.

Version 0.4.30 - bugfix release

Version 0.4.29

• Unit execution time has been improved significantly.
• The rc5, rc6, and xxtea cipher units were added.

Version 0.4.28

• Adds the option to completely disable the PowerShell band-aid introduced in 0.4.27 to allow using the Use-RawPipeline module.

Version 0.4.27

• Adds several VBA/VBS deobfuscation units and a deob-vba unit that applies all of them, similar to deob-ps1.
• Adds the camellia cipher unit.
• Adds the new struct unit format character w for decoded wide strings.
• The dnfields unit was extended and now also extracts string fields which are assigned a unique value.
• Implements a better PowerShell band-aid and displays a warning message.

Version 0.4.26

Adds various convenience output options in the Python REPL and adds documentation for those.

Version 0.4.25 - bugfix release

Version 0.4.24

• Adds the szdd decompression unit.
• Adds the lzjb decompression unit.
• Adds an option to the iff unit to check for the existence of a certain meta variable.
• The xtpyi unit now uses both uncompyle6 and decompyle3, even though they currently appear to have feature parity at best - there is some hope that one of them will support Python 3.9 in the future.
• Adds the groupby unit.
• Adds the isaac cipher unit.
• Adds the bat unit for deobfuscating batch scripts.

Version 0.4.23 - bugfix release

Version 0.4.22

• Adds the ripemd160 and ripemd128 units.
• Adds the xtw unit for extracting cryptocurrency wallet addresses.
• Adds the iemap unit to display a colored entropy heatmap.
• Introduces new syntax to the struct unit for handling byte alignment.
• The rsakey unit supports a new option to output the public key portion of a private key.
• The pemeta unit now computes the size of the PE file based on header information.
• Several switches for comparison operators were added to the iff unit.

Version 0.4.21

• Thanks to @baderj, the unit xlmdeobf was added which wraps the extremely useful XLMMacroDeobfuscator tool for extracting and deobfuscating Excel V4 macros.
• Adds the carve-7z unit for carving 7zip archives from blobs.

Version 0.4.20

• Renames the blockop unit to alu.
• Removes the shortcut unit carveb64z.
• Renames a number of command-line switches for carve, xtp, and other pattern extraction units.
• Adds a default argument to resub that makes it strip whitespace from the input by default.

Version 0.4.19

Improves performance by replacing an import of pkg_resources with equivalent functionality from importlib. On a test machine, this removes between 250 and 500 milliseconds from the execution time of any single unit.

Version 0.4.18

Changes the format for the binary formatter used in struct, rex, resub, and cfmt. It now uses a reverse multibin handler instead of parsing the modifier like a command-line pipeline.

Version 0.4.17 - bugfix release

Version 0.4.16 - bugfix release

Version 0.4.15

• Adds the lzo unit

Version 0.4.14

• The winreg unit is now able to extract data from Windows registry editor exports (i.e. .reg files).
• The key derivation units pbkdf2 and pbkdf1 use a more forgiving decoder to better cover the Rfc2898DeriveBytes class, which offers a call signature that receives an arbitrary byte string as password.
• The string regular expression pattern now excludes literal line breaks within the string.

Version 0.4.13

• Base64 regular expression patterns were improved to account for correct character counts.
• The dexstr unit was added.
• The index meta variable is now automatically populated within frames.
• The n40 string decryption unit was added.
• The xtpyi unit now extracts Python disassembly when decompilation fails.
• The lzma unit now correctly decompresses output produced by PyLZMA.

Version 0.4.12 - bugfix release

Version 0.4.11

• The doctxt unit was added; courtesy of @baderj

Version 0.4.10 - bugfix release

Version 0.4.9 - bugfix release

Version 0.4.8

• Adds the serpent unit.

Version 0.4.7

• Adds the xtpdf unit for extracting embedded objects from PDF documents.
• The accu: handler now supports pre-configured finite state machines for well-known rand() implementations.

Version 0.4.6

• The officectypt unit now supports the Excel default password VelvetSweatshop.
• The ci property has been removed from the output of peek --meta.
• The following units were added: xj0, evtx
• The hexdmp unit was renamed once more to hexload, and its pattern matching was improved.
• The asm unit was completely redesigned using an Angr-based fallback to produce better disassembly.
• The pcap-http unit now extracts the URL from whence the data was downloaded.
• The rep unit received some performance improvements.
• The refinery dependencies were cleaned up considerably.
• Blockwise operations no longer require numpy to be reasonably fast by implementing a dynamic inlining step.

Version 0.4.5

• Adds the cswap unit.
• The index counter of blockop now starts at zero.
• An option was added to the swap unit to swap the contents of two meta variables. This can also be used to rename a meta variable.
• An option was added to xtpyi to unpack, but not decompile the contents of a PYZ.
• Adds the --bare option to esc and uses it in peek.
• Adds the --meta option to ef. The ef unit now also descends into dot-directories and lists dot-files.
• The __init__.pkl file containing the unit lookup cache was moved into the distribution.

Version 0.4.4

• Adds the xtvba unit to extract Office document macros.
• Adds the pcap unit to extract TCP streams from packet capture files.
• Adds the xthtml unit to extract components of HTML documents.
• The htm unit has been renamed to htmlesc.
• The default sort order of sorted has been changed to descending.
• The pemeta and pkcs7 units now also extract certificate thumbprints.

Version 0.4.3

• Fixes an issue with applying ppjscript to obfuscated JavaScript files.
• Adds Murmur Hash units
• Adds xtpyi unit to extract PyInstaller-packed archives.
• Logging now uses the Python logging module.

Version 0.4.2 - bugfix release

Version 0.4.1

• Significantly improves unit loading time which had regressed due to the changes in 0.4.0.

Version 0.4.0

This release removes the setup-venv helper scripts and instead uses a slightly less ugly hack to resolve dependencies before running the refinery setup by declaring every dependency a build dependency in pyproject.toml. Any kind of installation should work seamlessly through pip.

Version 0.3.38 - bugfix release

Updates build system.

Version 0.3.37 - bugfix release

Version 0.3.36 - bugfix release

Version 0.3.35 - bugfix release

Version 0.3.34

• Fixes critical bug in deployment.
• Adds the msgpack unit.
• Adds the cull unit and changes the behaviour of conditional units to make filtered chunks invisible instead of removing them. Conditional units have been renamed to iff, iffs, iffx, and iifp.

Version 0.3.33

• Adds the xfcc unit, which replaces the intersection unit.
• The cm unit can now be used to remove meta variables.
• JSON dumps no longer use hex encoding for big integers as JSON has no size limit on integer expressions.
• The struct unit was significantly redesigned and the lprefix unit removed because it can now be trivially implemented with struct.
• The ifexpr unit has been renamed to iff and the iffp unit was added.
• The field names in dnfields have been altered to more closely resemble file names.
• Adds a list of default passwords to archive units.

Version 0.3.32

• Renames the fread unit to ef.
• Metadata / Format string expression parsing is now more flexible.

Version 0.3.31

• Adds the intersection unit.

Version 0.3.30

• Adds the xtjson and xtxml units for extracting data from JSON and XML files.
• Slight redesigns of lprefix, peek, xtmail, and cfmt.
• Refinery now has (very weak) support for PowerShell.
• Adds the --tabular option to ppjson to produce a flattened jason output.
• Changes to the in-code pipe syntax:
  • data | unit | unit is an iterable over output chunks
  • data | unit | unit | callable invokes callable with a bytearray containign all concatenated chunks
  • connected pipelines (data | unit | ... | unit) can be passed to str and bytes
• Path extraction units (like fread, xtzip) offer better control over the path variable.
• Variable merging was added to the pop unit.
• The cm unit only populates size and index by default, never performing a full scan unless explicitly requested.

Version 0.3.29

• Meta variables are now allowed in struct formats, and struct assumes no alignment by default.
• The pemeta unit now has support for RICH header data.
• The rsakey unit was added.
• The pop unit was extended by an option to discard chunks.
• Several new archive extractors are now available: xt7z, xtace, xtiso, and xtcpio.
• The xlxtr unit was refactored and generates more metadata.
• The sorted unit can sort by metadata variables now.
• The swap unit can now swap with an empty variable, which will empty the chunk body.

Version 0.3.28

• The trivia unit was renamed to cm for "common meta".
• The pemeta unit can now display PE header information, .NET header flags, and supports a table view instead of the JSON output.
• Python expressions all across multibin arguments no longer restrict the operators that can be used.
• The domain regular expression was updated with new TLDs and the artificial TDLs .coin and .bazar.
• The terminate unit was added.
• The struct unit was added.

Version 0.3.27

• Adds the ifexpr and ifstr units for filtering framed data.
• The pemeta unit now also extracts the EntryPointToken field from the .NET header.

Version 0.3.26

• The hexview unit was removed, instead the hexdmp unit was created. By default, this unit converts hexdumps back to binary, the previous functionality of hexview is now available as the reverse operation of hexdmp.
• Adds the dnblob unit.
• The drp unit underwent major refactoring with the goal to improve both speed and quality of results. Two options were added to help control these new settings.

Version 0.3.25 - bugfix release

Version 0.3.24

• Adds the xtrtf unit to extract embedded objects from RTF documents.
• Adds the officecrypt unit to decode password-protected Office documents.
• Improves PKCS7 parsing and fixes some cases where pemeta did not display the details of the digital signature.
• Adds brieflz support to the universal decompress unit.

Version 0.3.23

• Unification of (nearly) all multibin handlers. Only the yara: and escape: handlers remain to regular expression type arguments.
• Adds the multibin handlers accu, reduce, cycle, and take.
• Alters the le and be handlers to support both conversion from integer to byte string and vice versa.
• Renames the unpack handler to btoi and adds the btoi handler which performs the inverse operation.
• Command line switches for the lprefix unit changed.
• Adds the global --lenient option which is now required to admit partial results as output.

Version 0.3.22

• Adds the blz unit for BriefLZ compression and decompression.

Version 0.3.21

• Adds the xtdoc unit which can extract more files from Office documents than xtzip.
• Adds the trivia unit which can be used to attach certain meta variables. Moving forward, this will be the preferred way to access simple invariants of a binary chunk. For now, it can attach the integer variables size and index, containing the size of the data in bytes and the chunk index within the current frame, respectively. The eval: handler for numeric multibin values no longer accepts the special variable N to represent the chunk size as this functionality can be recovered by preprocessing each chunk with trivia and using the variable size instead of N.
• The carve-pe unit is now a path extractor unit (TL/DR: More command line options).

Version 0.3.20 - documentation

Version 0.3.19 - bufgfix release

Version 0.3.18

• Changes the interface for the frame squeeze mechanic
• Adds option to pefile to compute carve size based on virtual section sizes & offsets.

Version 0.3.17

• Using hex escape sequences in the replacement string for resub now works as expected.
• The yara: modifier for regular expression based units now accepts lowercase hex characters.
• The imphash unit's performance was improved slightly.
• Additional options for the pecarve unit.
• Adds the ppjscript unit (wrapper around jsbeautifier).
• The vsnip unit can now extract more than one memory region.
• Adds a count restriction to the resplit and resub units.

Version 0.3.16

• The interface for cipher units has been changed; the encryption mode is no longer a mandatory argument. Better handling of various cipher block chaining modes has been implemented.
• Conservative option added to peoverlay and pestrip.

Version 0.3.15

• The salsa and chacha cipher units now have pure Python implementations that allow you to specify the number of rounds. The PyCryptodome interfaces still exist, now as units salsa20 and chacha20.
• The HMAC unit was added to support simple HMAC based key derivation.
• The dump unit stream mode has been adjusted so that it is possible to write consecutive data to a file inside a nested frame.

Version 0.3.14

• The cfmt unit has been reworked to support more common modern Python format string syntax.
• The output of crc32 and adler32 checksum hashes has been altered to use the correct byte order.
• The rabbit unit was added which implements the RABBIT stream cipher.

Version 0.3.13

• The mpush, mpop, and mput units have been renamed to simply push, pop, and put.
• The autoxor unit has been transformed into the drp unit, the behavior of autoxor can be achieved using xor drp:copy:all.
• Data types of .NET fields are better detected by dnfields now, but a proper parser for type signatures is still missing.

Version 0.3.12

• The gz unit was deprecated because the zl unit covers its usecase (and does a better job at it).
• The lprefix unit for parsing length-prefixed data was added.
• Parsing of managed .NET string resources via the dnmr unit was fixed, these would previously be returned unparsed.
• The binpng unit has been improved and renamed to stego, a more flexible unit to extract data from images.

Version 0.3.11

• The peslice, elfslice, and pesect units have been removed.
• In their place, the cross-format units vsnip and vsect can now be used to extract data from virtual addresses and sections of PE, ELF, and MachO files.

Version 0.3.10 - bugfix release

Version 0.3.9

• adds md2 and md4 hashing algorithms
• the CryptDeriveKey unit now also mirrors the API call for SHA2 based hashing algorithms
• message type attachments in Outlook email formats are now supported by xtmail

Version 0.3.8

• The interface of the memory slicing units peslice and elfslice has changed.
• Python expression parser and numeric arguments have been refactored.

Version 0.3.7

• Removes the --install-option capability introduced in 0.3.5, see pip/#8748 for more information.
• The xttar unit was added.
• The lzma unit can now return partial results for buffers with junk bytes at the end.

Version 0.3.6

• The ifrex unit was added.
• The jvstr unit was added.
• A source distribution manifest was added to fix errors that occurred during source installs.

Version 0.3.5

• Using pip install --install-option=library binary-refinery or a REFINERY_PREFIX environment variable with value ! will now install the binary refinery without any command line scripts, only as a library.

Version 0.3.4 - bugfix release

Version 0.3.3 - bugfix release

Version 0.3.2

• It is now possible to use local refinery units (i.e. a Python script in the current director which contains a refinery unit that is not abstract) for multibin prefixes and in any other situation where units are dynamically loaded.
• The pesect unit was added.
• The resub and resplit units no longer offer options that have no bearing on their behavior.
• The lz4 unit was added with a pure Python implementation of LZ4 decompression.
• The jvdasm unit for disassembling Java class files was added.

Version 0.3.1 - bugfix release

Version 0.3.0

• The autoxor unit was added.
• The cfmt unit was added.
• The License of Binary Refinery was changed to 3-Clause BSD.

Version 0.2.1

• The netbios unit was added.
• The stretch unit was added.
• The hc128 cipher unit was added.
• The unit dnrc was split into dnrc for extracting .NET resources and dnmr for unpacking managed .NET resources.
• Several units that extract items from container formats have received a unified interface. So far, this interface applies to xtmail, xtzip, winreg, dnfields, dnrc, and dnmr.
• When using named match groups for the rex unit, these matches are now forwarded as metadata within frames.
• The xtzip unit was given an optional archive password parameter.
• The xtmail unit can now extract headers in text and json format.

Version 0.2.0

• Test coverage was increased
• The recode unit can now autodetect input encoding.
• Several bugfixes were performed on the vbe unit.
• More bandaids were added to PowerShell deobfuscation.
• The pestrip and peoverlay units were added.
• Interface retrofitting was completed.

Version 0.1.9

• Fixes a tiny bug in the PyPI display of the readme file, and completes changelog from previous version.

Version 0.1.8

• The rsa unit was improved and can handle the Microsoft blob format now.
• PowerShell deobfuscation was improved, but that doesn't change the fact that this would be much better with a proper parser.
• The b32 for base32 encoding and decoding was added.
• Preliminary support for meta variables has been added with the mpush, mpop, and mput units. This feature is experimental and not well documented yet.
• The --squeeze/-Z option was added to all units that produce multiple outputs: It disables the default separation of these outputs by line breaks.
• Pattern extraction units such as rex will now preserve the order of extracted strings, even when the --longest option is used.
• The suggested PATH environment variabe modification from the Linux installer script was corrected; The previous variant would make the refinery virtual environment take precedence over the global python executables.

Version 0.1.7

• The dump unit has been refactored to make it easier to use; Formatting of file names is done automatically now unless the flag -p or --plain is specified to prevent string formatting.
• The snip unit can now remove bytes from the input.
• The dnfields unit was added.
• The ppjson unit can now minify json by specifying 0 as the desired indentation width.
• The dsjava unit was improved, although it remains a work in progress.
• The fread unit received a linewise mode.

Version 0.1.6

• After some incomplete attempts to improve backwards compatibility, the package now simply requires Python 3.7.

Version 0.1.5

• Units can now be written with a Python __init__ constructor and deduce the command line interface from this constructor. A decorator class was added to help enriching the parameter list of the constructor with information on how to translate these into command line parameters. The goal is to eventually retrofit all units to follow this standard.
• The pemeta unit has more features now.
• The couple unit was added; it is an adapter to turn any stdin/stdout based command line tool into a refinery unit.
• The carve-xml unit was added.
• The dnstr unit was added.

Version 0.1.4

• All hashing prefixes for multibin expressions have been implemented as separate units, i.e. sha256 and md5 are now units that output the corresponding hash of the input data.
• The xtmail unit was added which can extract the body and attachments of email documents, both Outlook and MIME formats.
• The framed format was extended with rudimentary support for metadata in framed chunks. This is currently used by the xtzip and xtmail units to attach a name property to emitted chunks which contains the file name information from the parsed data. The dump unit now has a --meta option to read this name property and use it as the file name for dumping. The --meta options defaults to using the SHA256 hash of the data as the file name if no corresponding metadata is present.
• The pemeta unit was added.
• The carve-json unit was added.
• The peslice and elfslice units were given a unified interface.
• The b85 for base 85 encoding and decoding was added.

Version 0.1.3

• Fixes a bug in the .NET header parser where the tables were sometimes parsed in the wrong order.

Version 0.1.2

• The xtzip unit has been added, which can extract data from zip archives.
• The carve-zip unit has been added. It can carve ZIP files from buffers, similar to carve-pe for PE files.
• The rsa unit has finally been added.
• The rncrypt unit has been added.
• The dncfx unit has been added; it extracts the strings from ConfuserEx obfuscated .NET binaries.
• Adds support for TrendMicro Clicktime URL guards in the urlguards unit.

Version 0.1.1

• Several tests were added, testing now uses malshare to test units against real world samples. To properly execute tests, the environment variable MALSHARE_API needs to contain a valid malshare API key.
• A numpy import that always occured during any unit load was moved into the peek unit code to reduce import time of other units.
• Issues with wheel installation on Windows were fixed.

Version 0.1.0

• It is now possible to instantiate units in code with arguments of type bytes and have it work as expected, i.e. xor(B's3cr3t') will construct a xor unit that decrypts using the byte string key s3cr3t.
• The rex unit can now apply an arbitrary number of transformations to each match and return the results as separate outputs.
• The urlguards unit now supports ProofPoint V3 guarded URLs.
• Thanks to the recent fix of #29 in javaobj, the dsjava (deserialize Java serialized data) unit should now work. However, since there are currently no tests, bugs should be expected.

Version 0.0.6

• Processing of data in frames is no longer interrupted by errors in one unit.
• The global --lenient (or -L) flag has been added: It allows refinery units to return partial results. This behavior is disabled by default because it usually means that an error occurred during processing.
• The virtual environment setup script has received bug fixes for problems with absolute paths.

Version 0.0.5

• This changelog was added.
• The unit jsonfmt has been renamed to ppjson (for pretty-print json).
• The unit ppxml (pretty-print xml) was added.
• The unit carve-pe (carve PE files) was added.
• The unit winreg (read Windows registry hives) was added, also adding a dependency on the python-registry package (also on GitHub).
• .NET managed resource extraction was improved, although it is still not perfect.
• The unit sorted now only sorts the chunks of the input stream that are in scope.
• The unit dedup can no longer sort the input stream because sorted can do this.
• PowerShell deobfuscation and their test coverage was improved.
• Cryptographic units have been refactored; the salsa and chacha units now take a --nonce parameter rather than an --iv parameter, as they should.