Revision history for llvm-pretty

0.14.0.0 -- 2026-01-22

• Changes to support LLVM 19 (some of these changes are not backward-compatible):

  • Changes to LayoutSpec for DataLayout:

    • Size specification fields use a common sub-structure Storage which itself contains an Alignment common sub-structure: IntegerSize, VectorSize, FloatSize, and StackObjSize.
    • The pointer size specification field uses a PointerSize sub-structure that itself contains a Storage sub-structure.
    • Updated AggregateSize to make first field optional (it was dropped in LLVM 4) and the remaining fields are now provided via the Alignment sub-structure.
    • Added ProgramAddrSpace, GlobalAddrSpace, and AllocaAddrSpace constructors, each defined via an AddressSpace sub-structure.
    • Added NonIntegralPointerSpaces to record address spaces with an unspecified bitwise representation.
    • Added GoffMangling, WindowsX86CoffMangling, and XCoffMangling forms of Mangling.

  • Added GEPAttr flags for GEP instruction and constant expression, with RangeSpec for the latter.

  • Added numExtraInhabitants to DIBasicType.

  • Added support for DebugRecord parsing, specifically for:

    • FUNC_CODE_DEBUG_RECORD_VALUE
    • FUNC_CODE_DEBUG_RECORD_DECLARE
    • FUNC_CODE_DEBUG_RECORD_ASSIGN
    • FUNC_CODE_DEBUG_RECORD_VALUE_SIMPLE
    • FUNC_CODE_DEBUG_RECORD_LABEL

    This adds a new field to both the Result and Effect constructors of the Stmt type.

  • Pretty-printing with an LLVM version >= 19 now generates an error for icmp, fcmp, and shl constant expressions that are no longer supported as of LLVM 19.

• Changes to cast-related instructions:

  • Add a Bool field to ZExt which, if True, indicates that the argument must be non-negative. This is used by LLVM 18 and up.
  • Add a Bool field to UiToFp which, if True, indicates that the argument must be non-negative. This is used by LLVM 19 and up.
  • Add Bool fields to Trunc to check if the truncation would cause unsigned or signed overflow. These are used by LLVM 20 and up.

• Add a Bool field to ICmp, which indicates that the arguments must have the same sign. This is used by LLVM 20 and up.

• Add dlAtomGroup and dlAtomRank fields to DebugLoc', which were introduced in LLVM 21.

• Add dilColumn, dilIsArtificial, and dilCoroSuspendIdx fields to DILabel', which were introduced in LLVM 21.

• The following debug-related fields have had their types changed from Word64 to Maybe (ValMd' lab):

  • DIBasicType': dibtSize
  • DICompositeType': dictSize and dictOffset
  • DIDerivedType': didtSize and didtOffset

  This allows them to encode non-constant sizes and offsets (a capability used by Ada, for instance) in LLVM 21 or later.

• Added the bbStmtModifier which can be used to modify individual statements as they are emitted from the top-level instruction generator functions (e.g to add Debug Metadata to each statement).

• Fix a bug that would cause indirectbr statements to be pretty-printed incorrectly.

0.13.1.0 (October 2025)

• Add a FunctionPointerAlign constructor to LayoutSpec.

0.13.0.0 (March 2025)

• Changed some of the signatures of helper functions in the AST to make them more flexible by using Type' ident rather than Type in their signatures (the latter fixes ident to be Ident). Changed functions: isAlias, isPrimTypeOf, isVector, isVectorOf, isArray, and isPointer.

0.12.1.0 (August 2024)

• Fix for printing NaN and infinite floating point values.

• Add support for more AtomicRWOps.

0.12.0.0 (January 2024)

• Add preliminary support for LLVM versions up through 17.