[burgholzer]

🤖 AI text below 🤖

Description

Replace DD iostreams with LLVM output streams and borrowed byte input. Python serialization and worker state transport no longer copy payloads into input stringstreams. File input uses LLVM buffers, and Graphviz launching uses LLVM's process API instead of platform-specific code.

Preserve the existing throwing parsing/file APIs and exception/RTTI build policy. Borrowed output streams remain the caller's responsibility; file overloads check writes and close their descriptors, including during unwinding. Binary data keeps its native representation. Text serialization preserves the existing grammar and uses locale-independent decimal conversion, including finite subnormal values.

C++ callers must replace std::ostream sinks with llvm::raw_ostream and pass byte ranges to Package::deserialize; filenames use Package::deserializeFile. Python signatures remain unchanged. Installed MQT::CoreDD propagates its LLVM include and link requirements.

This is the DD I/O extraction from #2545, based on main after #2839 merged. The broader owned-error APIs and exception/RTTI changes remain in #2545.

Validation

Rebased on main dedaa13f5f, retaining its exportToDot rename throughout the LLVM export implementation, bindings, and tests. The range-diff contains only the corresponding rename and formatting adaptations.

• Clang 23, LLVM/MLIR 23.1, ThinLTO and mold: 4,102 native tests passed; installed Python DD tests passed 64/64.
• Six saved binary payloads reserialize byte for byte, including embedded and trailing NULs; binary/text decoding matches independently specified numerical inputs. Binary layout remains architecture-dependent.
• Targeted mutations fail for broadened text grammar, subnormal rejection, partial trailing binary records and locale-sensitive decimal parsing.
• An independent allocation-failure probe passes 256 file-output cases. Removing unwind cleanup reproduces descriptor leaks and a fatal buffered write error.
• Installed CMake consumer, stub generation, repository lint and whole-file C++ lint passed. Independent Ponytail review findings are incorporated.
• Complete executable Sphinx/Doxygen documentation and generated-link checks pass locally; the precision parameter is documented and the rendered API entry was inspected in Chrome.

Local validation is on Linux. Windows/macOS execution and hosted CI are unverified; no performance claim is made.

AI assistance: Codex implemented, independently reviewed and locally validated this extraction. Human review remains required.

Checklist

• The pull request only contains commits that are focused and relevant to this change.
• I have added appropriate tests that cover the new/changed functionality.
• I have updated the documentation to reflect these changes.
• The changes follow the project's style guidelines and introduce no new warnings.
• The changes are fully tested and pass the CI checks.
• I have reviewed my own code changes.

If PR contains AI-assisted content:

• Any agent that created, edited, or submitted GitHub content was explicitly authorized for that scope, as required by our AI Usage Guidelines.
• Every agent-authored or agent-edited public text body begins with the visible disclosure 🤖 *AI text below* 🤖 (titles are exempt).
• I have disclosed AI assistance in the PR description.
• I confirm that I have personally reviewed and understood all AI-generated content, and accept full responsibility for it.

[codecov]

Codecov Report

❌ Patch coverage is 93.93939% with 16 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
include/mqt-core/dd/Export.hpp	81.1%	10 Missing ⚠️
src/dd/ComplexValue.cpp	93.1%	2 Missing ⚠️
src/dd/Export.cpp	97.7%	2 Missing ⚠️
src/dd/RealNumberUniqueTable.cpp	83.3%	2 Missing ⚠️

📢 Thoughts on this report? Let us know!

[burgholzer]

@denialhaag @simon1hofmann FYI: also extracted from #2545. I love to see all the standard library iostream stuff gone and replaced by cleaner and faster stream APIs from LLVM. There may be even more places where we could make such a change.
C++'s iostreams are notorious for causing a substantial amount of build time and general slowness.