9. Release Notes

This following summarizes the most important changes in recent Spicy releases. For an exhaustive list of all changes, see the CHANGES file coming with the distribution.

9.1. Version 1.3

New Functionality

  • Add optimizer removing unused %random-access or %filter functionality

    If a unit has e.g., a %random-access attribute Spicy emits additional code to track and update offsets. If the %random-access functionality is not used this leads to unneeded code being emitted which causes unneeded overhead, both during JIT and during execution.

    We now emit such feature-dependent code under a feature flag (effectively a global boolean constant) which is by default on. Additionally, we added an optimizer pass which detects whether a feature is used and can disable unused feature functionality (switching the feature flag to off), and can then remove unreachable code behind such disabled feature flags by performing basic constant folding.

  • Add optimizer pass removing unused sink functionality

    By default any unit declared public can be used as a sink. To support sink behavior additional code is emitted and invoked at runtime, regardless of whether the unit is used as a sink or not.

    We now detect unused sink functionality and avoid emitting it.

  • GH-934: Allow $$ in place of self in unit convert attributes.

Changed Functionality

  • GH-941: Allow use of units with all defaulted parameters as entry points.
  • We added precompilation support for libspicy.h.
  • Drop support for end-of-life Fedora 32, and add support for Fedora 34.

Bug fixes

  • Correctly handle lookups for NULL library symbols.
  • Use safe integers for size functions in the runtime library.
  • Make it possible to build on ARM64.
  • Fix building with gcc-11.

Documentation

9.2. Version 1.2

New Functionality

  • GH-913: Add support for switch-level &parse-at and &parse-from attributes inside a unit.

  • Add optimizer pass removing unimplemented functions and methods.

    This introduces a global pass triggered after all individual input ASTs have been finalized, but before we generate any C++ code. We then strip out any unimplemented member functions (typically Spicy hooks), both their definitions as well as their uses.

    In order to correctly handle previously generated C++ files which might have been generated with different optimization settings, we disallow optimizations if we detect that a C++ input file was generated by us.

Changed Functionality

  • Add validation of unit switch attributes. We previously silently ignored unsupported attributes; now errors are raised.
  • Remove configure option --build-zeek-plugin. Spicy no longer supports building the Zeek plugin/analyzers in-tree. This used to be available primarily for development purposes, but became challenging to maintain.
  • Add environment variable HILTI_CXX_INCLUDE_DIRS to specify additional C++ include directories when compiling generated code.
  • GH-940: Add runtime check for parsing progress during loops.

Bug fixes

  • Fix computation of unset locations.
  • Fix accidental truncating conversion in integer code.

Documentation

9.3. Version 1.1

New Functionality

  • GH-844: Add support for &size attribute to unit switch statement.

  • GH-26: Add %skip, %skip-pre and %skip-post properties for skipping input matching a regular expression before any further input processing takes place.

  • Extend library functionality provided by the spicy module:

    • crc32_init()/crc32_add() compute CRC32 checksums.
    • mktime() creates a time value from individual components.
    • zlib_init() initializes a ZlibStream with a given window bits argument.
    • Zlib now accepts a window bits parameter.

  • Add a new find() method to units for that searches for a bytes sequence inside their input data, forward or backward from a given starting position.

  • Add support for &chunked when parsing bytes data with &until or &until_including.

  • Add encode() method to string for conversion to bytes.

  • Extend parsing of void fields:

    • Add support for &eod to skip all data until the end of the current input is encountered.
    • Add support for &until to skip all data until a deliminator is encountered. The deliminator will be extracted from the stream before continuing.

  • Port Spicy to Apple silicon.

  • Add Dockerfile for OpenSUSE 15.2.

Changed Functionality

  • Reject void fields with names.
  • Lower minimum required Python version to 3.2.
  • GH-882: Lower minimum required Bison version to 3.0.

Bug fixes

  • GH-872: Fix missing normalization of enum label IDs.
  • GH-878: Fix casting integers to enums.
  • GH-889: Fix hook handling for anonymous void fields.
  • GH-901: Fix type resolution bug in &convert.
  • Fix handling of &size attribute for anonymous void fields.
  • Fix missing update to input position before running %done hook.
  • Add validation rejecting $$ in hooks not supporting it.
  • Make sure container sizes are runtime integers.
  • Fix missing operator<< for enums when generating debug code.
  • GH-917: Default-initialize forwarding fields without type arguments.

Documentation

  • GH-37: Add documentation on how to skip data with void fields.

9.4. Migrating from the old prototype

Below we summarize language changes in Spicy compared to the original research prototype. Note that some of the prototype’s more advanced functionality has not yet been ported to the new code base; see the corresponding list on GitHub for what’s still missing.

Changes:

  • Renamed export linkage to public.

  • Renamed %byteorder property to %byte-order.

  • Renamed &byteorder attribute to &byte-order.

  • Renamed &bitorder attribute to &bit-order.

  • All unit-level properties now need to conclude with a semicolon (e.g., %random-access;).

  • Renamed &length attribute to &size.

  • Renamed &until_including attribute to &until-including.

  • Replaced &parse with separate &parse-from (taking a “bytes” instance) and &parse-at (taking a stream iterator) attributes.

  • Attributes no longer accept their arguments in parentheses, it now must <attr>=expr. (Before, both versions were accepted.)

  • uint<N> and int<N> are no longer accepted, use uintN/intN instead (which worked before already as well)

  • list<T> is no longer supported, use vector<T> instead.

  • New syntax for parsing sequences: Use x: int8[5] instead of x: vector<int8> &length=5. For lists of unknown size, use x: int8[]. When parsing sequences sub-units, use: x: Item[]; or, if further arguments/attributes are required, x: (Item(1,2,3))[]. (The latter syntax isn’t great, but the old syntax was ambiguous.)

  • New syntax for functions: function f(<params>) [: <result>] instead of <result> f(<params>)

  • Renamed runtime support module from Spicy to spicy (so use import spicy)

  • In units, variables are now initialized to default values by default. Previously, that was (inconsistently) happening only for variables of type sink. To revert to the old behaviour, add “&optional” to the variable.

  • Renamed type double to real.

  • Generally, types don’t coerce implicitly to bool anymore except in specific language contexts, such as in statements with boolean conditions.

  • Units using any of the random access methods (e.g., input()), now need to explicitly add a unit property %random-access.

  • Filters can now be implemented in Spicy itself. The pre-built filter::Base64Decode and filter::Zlib provide the base64 and zlib functionality of the previously built-in filters.

  • {unit,sink}::add_filter are renamed to {unit,sink}::connect_filter.

  • Enums don’t coerce to bool anymore, need to manually compare to Undef.

  • Coercion to bool now happens only in certain contexts, like if-conditions (similar to C++).

  • The sink method sequence has been renamed to sequence_number.

  • The effect of the sink method set_initial_sequence_number no longer persists when reconnecting a different unit to a sink.

  • &transient is no longer a supported unit field attribute. The same effect can now be achieved through an anonymous field (also see next point).

  • $$ can now be generally used in hooks to refer to the just parsed value. That’s particularly useful inside hooks for anonymous fields, including fields that previously were &transient (see above). Previously, “$$” worked only for container elements in foreach hooks (which still operates the same way).

  • Fields of type real are parsed with &type attribute (e.g., &type=Spicy::RealType::IEEE754_Double). They used to &precision attributes with a different enum type.

  • Assigning to unit fields and variables no longer triggers any hooks. That also means that hooks are generally no longer supported for variables (This is tricky to implement, not clear it’s worth the effort.)

  • When importing modules, module names are now case-sensitive.

  • When parsing vectors/lists of integers of a given length, use &count instead of &length.

  • Zeek plugin:

    • Bro::dpd_confirm() has been renamed to zeek::confirm_protocol(). There’s also a corresponding zeek::reject_protocol().
    • To auto-export enums to Zeek, they need to be declared public.