v0.9.0

API Breaks

• Remove the BinaryFragment.Data field and replace it with BinaryFragment.RawData, which contains the decoded but still compressed data from the fragment. Clients should use the io.Reader returned by the new BinaryFragment.Data() method to access the uncompressed data. (#76)
• Apply() and BinaryApplier.ApplyFragment() enforce a default maximum binary fragment size of 100 MiB. Clients may adjust or disable this limit by passing the WithMaxBinaryFragmentBytes option to their apply calls. (#77)

Improvements

• Short source files are reported as conflicts instead of EOF errors during applies (#63)
• Clients can limit the size of binary fragments during applies using the WithMaxBinaryFragmentBytes option (#77)
• BinaryFragment.String() returns the same content lines as were originally parsed (assuming normalized whitespace) instead of an equivalent encoding that parsed to the same value but was not the same text (#76)

Fixes

• Fix a panic or memory exhaustion issue when applying text fragments with very large old positions (#75)
• Fix unbounded decompression leading to memory exhaustion when parsing binary fragments (#76)
• Fix unbounded writes to the client-provided io.Writer when applying binary fragments (#77)

v0.8.1

Fixes

• Parse binary file headers that include file paths (#55)
• Fix binary file headers when formatting a patch as a string (#56)

v0.8.0

Improvements

• Add the File.String() method to convert a parsed file back to a patch representation (#48)

Changes

• Set the module's Go version to 1.21 to due to usage of generics and the slices package in test code

v0.7.4

Fixes

• Return the input as the preamble, instead of an empty string, when a patch has no files (#46)

v0.7.3

Improvements

• Accept invalid (according to RFC 5322) email addresses in From: lines of mail-formatted patches (#44)
• More closely follow Git's logic when parsing identities with ParsePatchIdentity (#44)
  • Extract email addresses even when they do not appear in angle brackets
  • Consider content that appears after the email address as part of the name
  • Unquote escaped characters in quoted strings and RFC 5322 comments
  • Normalize runs of white space in names to single space characters

v0.7.2

Improvements

• Accept more inputs in ParsePatchIdentity (#42)
  • Allow identity strings with empty email address (must still include <> markers)
  • When an identity contains only an email address, also use it to set the Name field

v0.7.1

Fixes

• Fix error parsing mode lines with trailing whitespace (#38)

v0.7.0

API Breaks

• Remove the Applier type and replace it with TextApplier and BinaryApplier. This does not impact clients that only use the top-level Apply function (#32).

Improvements

• Add an option to control which content is removed from subject lines when parsing the headers of mail-formatted patches (#36)
• Simplify internal logic for patch header parsing (#34)
• Run tests and lint checks with Go 1.19 (#35)

v0.6.1

Fixes

• Fix out-of-bounds panic in timestamp parsing (#28)
• Fix panic when parsing certain invalid text fragments (#29)
• Return an error when parsing text fragments that contain only context lines (#29)
• Use ioutil to retain compatibility with Go versions prior to 1.16 (#31)

v0.6.0

API Breaks

• Improve the behavior of the LineReaderAt interface when the final line does not have a newline character. This change is only relevant for clients that define LineReaderAt implementations and should not affect most users of the library. If you use a custom LineReaderAt implementation, please review the documentation and update your implementation to avoid errors (#27).

Fixes

• Fix apply errors (unexpected EOF) with files that are multiples of 1024 bytes and have no final newline (#27).