Deprecated Items

This appendix attempts to document the things in SAW that have been deprecated and removed over recent releases, in more detail than will fit in the change log and release notes.

Warning When Used

The following elements are currently deprecated and warn when used; this has happened since SAW 1.5 was released, so they will be/are that way in SAW 1.6. They are expected to be hidden by default in SAW 1.7.

  • The prove_extcore builtin; you can just use prove_print instead now.

  • The add_prelude_eqs and add_cryptol_eqs builtins; use add_core_thms instead.

  • The assume_unsat builtin, which was informally deprecated years ago; use admit instead. admit takes an extra string argument that’s supposed to state why you’re admitting the theorem without proof.

  • The type SetupValue (an old name for LLVMValue)

  • The type CrucibleMethodSpec (an old name for LLVMSpec)

  • The type CrucibleSetup (an old name for LLVMSetup)

The following elements were deprecated as of SAW 1.5, and currently warn. They will be hidden by default when we remove Boolector from what4-solvers. This will happen at some point after SAW 1.6; the time frame depends mostly on the maintenance burden.

  • The commands related to the Boolector solver, which has been replaced upstream with Bitwuzla. All boolector commands have equivalent bitwuzla commands.

Hidden by Default

The following elements were previously experimental and will be deprecated and hidden by default as of SAW 1.6:

  • The experimental builtin crucible_llvm_verify_x86 (an old name for llvm_verify_x86), which was left in place in SAW 1.5 due to some downstream uses that needed to be migrated first.

There are not yet any (non-experimental) elements that were first deprecated as of SAW 1.6 and are now hidden by default. (Such elements would continue to be hidden by default in SAW 1.7 and would be removed in SAW 1.8.)

The following elements were deprecated as of SAW 1.5 (or earlier), and are hidden by default as of SAW 1.6. They are expected to be removed in SAW 1.7.

  • The commands related to the CVC4 solver; proofs should be migrated to CVC5. Each cvc4 builtin has a corresponding cvc5 builtin. If you have proofs that work with cvc4 but time out in cvc5, and you cannot find another solver that can handle them instead, please file an issue (with us or CVC5 upstream as seems best).

  • The external_aig_solver builtin, which was renamed to arbitrary_aig.

  • The external_cnf_solver builtin, which was renamed to arbitrary_cnf.

  • The w4_offline_smtlib2 builtin, which was renamed to offline_w4_smtlib2 for consistency.

  • The llvm_struct builtin; the current name for it is llvm_alias, because it looks up typedef names. If you are trying to create an LLVM struct type from its contents, as many historical callers were found to be, you want llvm_struct_type instead.

  • The old coq names of the Rocq backend; change coq to rocq.

  • The many old crucible names of LLVM backend functions; e.g. crucible_null is now llvm_null. Most of these names have been outdated for years.

  • Similarly, the crucible_java_extract builtin, which is an old name for jvm_extract.

Removed

The following elements have been removed entirely:

  • The env builtin (use the :env REPL command instead)

  • Certain old crucible names for LLVM backend functions that never made it past “experimental”:

    • crucible_fresh_cryptol_var

    • crucible_alloc_with_size

    • crucible_points_to_array_prefix

    • crucible_llvm_compositional_extract

    • crucible_llvm_array_size_profile

  • The addsimp' and addsimps' builtins, which added unproven rewrite rules to a Simpset. Don’t do that; it’s borrowing trouble. If you can’t prove your rewrite rules, for whatever, reason, use admit explicitly and then use the ordinary addsimp and/or addsimps bulitins to add them to Simpsets.

  • The crucible_setup_val_to_term builtin. This was a partial inverse for llvm_term; that is, it turned an LLVMValue value back to a Term if it was one that could be represented. In general LLVM specs should be written so they lift values from Term to LLVMValue and not the other way around.

  • The unused type JavaSetup; any stray references to it should be changed to JVMSetup.

  • The extract_uninterp builtin. This was a way to rewrite a Term to make functions uninterpreted (generalize it over all functions that are functions in place of specific functions with definitions) and also extract the uses of those functions for further processing. This depended on problematic internal functionality, and also relied on being able to identify which applications of a given function are actually the same, which is difficult in general and not the right approach.

  • The Heapster, MRSolver, and Monadify features, which proved unmaintainable and inconsistent with efforts to build SAW an assurance case.

  • The sbv_uninterpreted and read_sbv builtins and their Uninterp type. This was code for importing from an obsolete file format.

  • The callcc builtin, whose behavior was erratic, whose internal implementation was insupportable, and which had only one known use that was itself better done some other way.

Also, the following SAWScript language-level behavior has been removed:

  • It used to be possible to update top-level variable bindings by just binding the same variable again. This would update some uses and not others depending on the then-very-dodgy evaluation semantics of SAWScript. Now if you bind the same variable again, it just shadows the old variable in the namespace and does not replace it.

    For uses that really need the mutability behavior, you can now, at the top level only, do let rebindable ..., which lets you bind a variable that you can update with another let rebindable ..., and the semantics of the update are predictable.