[ENG-2416] Add write_sta_db and read_sta_db commands

[gigeresk]

This PR adds StaDBs, which are serialized STA sessions. They can be written with write_sta_db and read with read_sta_db. Reading an StaDB session will clear state before restoring the serialized session.
Rewrites after restore are byte-idempotent; as such, some data structures need to be sorted before serialization.
StaDBs are considered valid across builds of OpenSTA with the same stadb_version and stadbAbiGuard(), which is a hash of sizeof for Vertex, Edge, Liberty, etc. Not shareable across libc++/libstdc++ builds due to various sizeof differences.

Covered in serialization

Liberties (NLDM), SDC, Netlists, Graph and search (e.g. report_checks)

Not covered in serialization (out of scope)

Parasitics (SPEF) are omitted, MCMM (i.e. anything >1 Scene throws), CCS is dropped, statistical throws

Open questions:

• Making StaDb a friend to access private members is less code on the core but may raise concerns upstream, leave this as future work?

[greptile-apps]

Greptile Summary

The PR adds serialization and restoration of supported OpenSTA sessions through write_sta_db and read_sta_db.

• Adds the StaDB binary container, codecs, and serializers for Liberty, network, graph, SDC, and search state.
• Integrates the new Tcl and C++ APIs into the build and documentation.
• Adds regression coverage for snapshot fidelity and deterministic rewrites.

Confidence Score: 0/5

The PR is not safe to merge because crafted StaDB files can still terminate OpenSTA through oversized allocations, and a failed restore still destroys the active session.

The new 64 GiB section and 512 GiB file caps still allow infeasible attacker-controlled vector allocations, while readStaDb clears the existing session before fallible section decoding and provides no rollback when decoding fails.

Files Needing Attention: stadb/StaDbFormat.hh, stadb/StaDbFile.cc, and stadb/StaDbReader.cc

Security Review

The new allocation limits do not close the previously reported denial-of-service paths: crafted input can still request a decompressed allocation up to 64 GiB or a whole-file allocation up to 512 GiB before being rejected.

How this was verified: Both accepted sizes flow directly into vector allocations, and the configured limits remain far above realistically available process memory.

Important Files Changed

Filename	Overview
stadb/StaDbFormat.hh	Introduces allocation caps, but 64 GiB per section and 512 GiB per file remain large enough for the previously reported memory-exhaustion paths.
stadb/StaDbFile.cc	Adds pre-allocation checks, but accepted attacker-controlled sizes still flow directly into potentially infeasible vector allocations.
stadb/StaDbReader.cc	Adds cleanup for failed in-place restores, but still destroys the previous session and leaves the caller with an empty one.
stadb/StaDbGraph.cc	Restores graph, levelization, delay, and period-check state; malformed collection counts also reinforce the need for effective aggregate resource limits.
stadb/StaDbWriter.cc	Coordinates deterministic serialization of supported session sections and rejects unsupported session configurations.

Sequence Diagram

sequenceDiagram
  participant Tcl as read_sta_db
  participant File as DbFileReader
  participant STA as Active Sta
  participant Restore as Section decoders
  Tcl->>File: read and validate container
  File-->>Tcl: validated sections
  Tcl->>STA: clearSession()
  Tcl->>Restore: restore Liberty/network/graph/SDC/search
  alt restore succeeds
    Restore-->>Tcl: restored session
  else section decoding throws
    Restore-->>Tcl: exception
    Tcl->>STA: clearSession()
    Tcl-->>Tcl: rethrow with empty session
  end

Loading

_{Reviews (3): Last reviewed commit: "File safety + sta docs" | Re-trigger Greptile}

[gigeresk]

@greptile

[gigeresk]

@greptile

[linear-code]

ENG-2416