NVC - VHDL Compiler and Simulator

Unreleased changes

Version 1.12.0 - 2024-04-07

• The --jit elaboration option no longer requires --no-save.
• Fixed a crash when subtype bounds depend on a package instance generic (#815).
• Fixed various issues in the implementation of guarded blocks and disconnection specifications (#829).
• The --std option now controls which VHDL standard the --install command uses to compile third-party libraries (#836).
• The prefix of an indexed or slice name now must be another name or function call, as required by the LRM (#835).
• The implementation of conversions in port maps has been reworked and fixes a number of long-standing issues (#843).
• Added support for inertial keyword in port maps (#843).
• Fixed a bug where 'last_value could give the wrong result if the signal has multiple sources.
• Updated to OSVVM 2023.09a and UVVM 2023.09.16 for nvc --install.
• The --exit-severity= option now also controls which severity level results in a non-zero exit code (#850).
• Improvements to waveform dumping for signals with record types or types with non-locally-static bounds (#851, #852).
• The parameter for attributes with dimensions such as 'length(N) may be any integer type when --relaxed is passed (#862).
• Resolved several other minor issues (#654, #854, #855, #859, #863).
• The default standard version was changed to VHDL-2008.
• The --vhpi-trace option now implies --vhpi-debug.
• The bounds of array aggregates are now calculated correctly in several corner cases (#873).
• Added supported for VHPI foreign subprograms.

Version 1.11.3 - 2024-02-04

• Fixed incorrect effective value when a signal has multiple sources due to port-collapsing optimisation (#824).
• Signals that appear in aggregate choice expressions like (S downto 0 => '0') are now included in the sensitivity list for concurrent statement equivalent processes, and in process (all) (#825).
• The direction of aggregates with range choices is now calculated correctly in VHDL-2008 mode (#826).
• Fixed a memory corruption bug when evaluating certain aggregates that contain a range association (#827).
• Resolved several other minor issues (#837, #839, #840).

Version 1.11.2 - 2024-01-04

• Fixed an incorrect length check in the equivalent process for non-static port map actuals (#817).
• Library file names for design units that contain extended identifiers such as /Foo/ are now encoded in a way that avoids illegal characters and case sensitivity issues on Windows and macOS (#816).
• Implemented vhpiIsNullP and vhpiIsDiscreteP for ranges.
• Fixed a crash when an aliased enumeration literal appears in a case choice expression (#819).
• Fixed calculation of longest static prefix with indexes and slices of a constant array (#820).
• Fixed a crash during elaboration when simplifying an if-statement that depends on a generate parameter (#821).
• Increased the limit on the number of subprogram arguments to allow compiling some Lattice IPs (from @sean-anderson-seco).

Version 1.11.1 - 2023-12-16

• Fix crash during elaboration when an if-statement branch is always taken (#812).
• VHPI plugins can now access types in instantiated packages (from @sean-anderson-seco).
• The --version output now includes the bare Git commit SHA if the program is built from a shallow clone with no tags (#813).

Version 1.11.0 - 2023-12-06

• New command --cover-export exports coverage data in the Cobertura XML format which is supported by most CI environments such as GitLab.
• Generics on internal instances can now be overridden with the -g elaboration option. For example -g uut.value=42.
• Implemented the 'reflect attribute and associated protected types from VHDL-2019.
• Added support for VHDL-2019 sequential block statements.
• Implemented the VHDL-2019 directory I/O functions in std.env.
• Added VHDL-2019 assert API (with @Blebowski).
• Implemented 'image, 'value and to_string for composite types in VHDL-2019.
• Implemented the “closely related record types” feature from VHDL-2019.
• Implemented the “composition with protected types” feature from VHDL-2019.
• The new --shuffle option runs processes in a random order which can help to identify code that depends on a particular execution order.
• Updated to OSVVM 2023.07 for nvc --install.
• Various enhancements and fixes to the VHPI implementation.
• Implemented the VHDL-2019 changes to instance_name and path_name for protected type variables.
• VHPI error messages are no longer reported as diagnostic messages on the console. The new --vhpi-debug option restores the old behaviour.
• Support for type conversions between arrays with closely related element types.
• Added support for FSM state coverage collection (from @Blebowski).
• An alias of a type now correctly creates implicit aliases for each predefined operator of that type (#776).
• Improve overload resolution where a partial named association implies the formal parameter must be an array (#793).
• Handling of implicit conversion for universal types has been reworked to better comply with the LRM.
• Fixed a crash when string literal characters have a type which is an alias to another type (#801).
• Added a warning when calling the predefined "=" and "/=" operators on arrays and the left and right hand sides have different lengths.
• Expressions like abs(x)**2.0 are now parsed correctly (#805).

Version 1.10.4 - 2023-10-16

• Fixed compatibility with LLVM 17.
• Fixed an intermittent crash when using the --jit elaboration option.
• Improved overload resolution when subprogram argument is known to have character type.
• Improved bounds checking for port map actuals.
• Fixed a crash with aggregates for record types that contain an unconstrained field (#768).
• Fix spurious “checksum 00000000” error when analysing multiple design units and one contains use work.all (#769).
• Fix crash when using the --relaxed analysis option and multiple warnings are printed (#770).

Version 1.10.3 - 2023-09-17

• Fixed memory corruption in rare circumstances with functions declared in process declarative regions (#751).
• Fixed a race condition when creating library directories that occasionally led to build failures with make -j (#746).
• Aliases of protected type methods are now allowed.
• Fixed a crash when a pragma appears outside of a design unit (#752).
• Fixed a crash analysing an alias of an alias of a subprogram (#755).
• Fixed an elaboration failure when an allocator expression has an array type with an unconstrained element (#756).
• Fixed a memory leak in the --print-deps command.
• Fixed a crash when evaluating globally static expressions during elaboration with coverage enabled (#759).
• Fixed an analysis crash where a predefined function for a type declared in a package is overridden in the package body only (#760).

Version 1.10.2 - 2023-08-20

• Fixed a crash due to an array bounds check being incorrectly optimised out (#747).
• The type of string literals and aggregates in generic map associations such as G => X"00" where G has a generic type can now be determined correctly (#750).
• Fixed a spurious “duplicate declaration” error if a use clause for an instantiated packages appears in the same declarative region that instantiated the package (#750).
• Improved checking for configuration specifications.
• Empty bit string literals are now parsed correctly.
• Fixed stack corruption when a function returns an alias of one of its array arguments.

Version 1.10.1 - 2023-07-28

• Fixed incorrect sensitivity list generation with concurrent statements such as x <= C(y) where C is a constant array.
• Fixed an incorrect optimisation which caused 'event to return the wrong value in rare circumstances.
• Decimal bit string literals more than 64 bits wide are now supported (#731).
• The format of fractional time values returned by the standard to_string function was changed to match other simulators.
• Fixed a crash when constant folding a locally static expression in a package body (#742).
• Added support for reading command line arguments in VHPI (from @Forty-Bot).
• Fixed a compilation error when using the predefined minimum/maximum functions with arrays of physical types.
• Overloaded protected procedure calls could read the wrong value of protected type variables in rare circumstances.

Version 1.10.0 - 2023-07-14

• The Zstandard compression library is now a build dependency. Install libzstd-dev or similar.
• The integer type is now 64-bit in VHDL-2019 mode.
• The VUnit VHDL libraries can now be installed with nvc --install vunit but please note this does not install the Python infrastructure.
• Updated to OSVVM 2023.05 and UVVM 2023.03.21 for nvc --install.
• Conditional expressions are now allowed in constant, signal, and variable declarations in VHDL-2019 mode.
• Conditional return statements are now supported in VHDL-2019.
• Added support for the “function knows vector size” feature in VHDL-2019.
• Entity ports with variable class and protected type are now supported in VHDL-2019 mode.
• The xpm_vhdl project which provides VHDL models of the Xilinx XPM macros can now be installed with nvc --install xpm_vhdl.
• Many improvements to the VHPI implementation (from @Forty-Bot).
• vhpi_put_value with vhpiDepositPropagate mode is now supported.
• The Synopsys std_logic_misc package is now compiled for 2008 (#696).
• Fixed an issue where leading NUL characters in a report message would prevent the entire message being printed (#700).
• Added support for interfaces in VHDL-2019 including mode view declarations, mode view indications, and the 'converse attribute.
• Added support for VHDL-2008 matching select? statements (#705).
• Added support for the new 'designated_subtype and 'index attributes in VHDL-2019.
• Implemented the date/time functions from std.env in VHDL-2019.
• The default exit severity was changed from error to failure. This means a failing assertion no longer immediately terminates the simulation. The old behaviour can be restored with --exit-severity=error.
• Comparison operators as well as minimum/maximum functions are now defined for all scalar array types in VHDL-2019.
• Added support for selected signal and variable sequential assignment statements.
• The -a analysis command now accepts an -f list option where list is a text file containing a list of files to analyse. Alternatively this may be written @list.
• Accesses to protected types and files are now allowed in VHDL-2019.
• Fixed a crash when indexing a null array (#734).
• Named and range choices are now supported in aggregate targets of variable and signal assignments (#712).
• The synopsys.attributes package is no longer distributed or built as part of the standard libraries.

Version 1.9.2 - 2023-05-01

• Fix elaboration errors with recursive entity instantiation (#668).
• Updated to latest GtkWave FST writer library.
• Fix crash when an external name is used in a wait expression (#674).
• Fix stack corruption when passing large numbers of arguments (#665).
• Protected procedure calls with out or inout signal arguments now create drivers (#675).
• Fixed a crash when a next or exit statement appears inside process (all) (#676).
• The --jit elaboration options now works on Intel Macs (#680).
• Fixed a corner case where a forced signal with no drivers had the wrong effective value after release (#681).
• Fixed a crash when a signal assignment statement contains only unaffected (#677).

Version 1.9.1 - 2023-04-15

• Fix build errors and warnings on MSYS2 Clang x64 environment.
• Fix build failure due to missing SHT_X86_64_UNWIND on Alpine and Ubuntu 18.04 (#666).
• Elaboration now works correctly when the --with-system-cc= configure argument is given a non-absolute path (#667).
• Fixed a crash when a subprogram declared in a for-generate block is called during elaboration (#668).
• Real to physical or 64-bit integer conversion now produces the correct result when evaluated during elaboration (#669).
• Matching case? statements are now handled correctly in inside process (all) (#670).
• Pass -no_fixup_chains to linker on macOS to avoid warning.
• ; can now be used as a separator in the --work and --map arguments on Windows in addition to : (#671).

Version 1.9.0 - 2023-04-07

• Code generation has been rewritten to enable faster elaboration and “just-in-time” compilation in the future.
• Now compatible with LLVM 16.
• Implemented the VHDL-2019 call path reporting API.
• The elsif VHDL-2019 conditional analysis directive now works correctly (#604).
• The 'transaction implicit signal no longer incurs a delta-cycle delay.
• x'ascending now reports the correct result if x has unconstrained array type and null range.
• The predefined "=" operator on record types now always uses the predefined equality comparison for fields even in the presence of a user-defined "=" operator.
• It is no longer necessary on Windows to link VHPI plugins at elaboration time with NVC_FOREIGN_OBJ. Use the --load option to load the plugin at run time as on other operating systems.
• The experimental --jit elaboration option defers native code generation until run time. This can dramatically reduce total test time for short-running simulations.
• Statements like wait for X where X is negative but not a constant now produce an error at run time (#633).
• NVC is now supported by VUnit.
• Implicit signal attributes like 'transaction are now considered static signal names (#640).
• Added support for fine-grained coverage collection via --coverage-spec elaboration option (from @Blebowski).
• The ABI for passing unconstrained arrays to foreign subprograms changed slightly, see the manual for details.
• Implemented new file I/O operations from VHDL-2019.
• Added analysis option --define to set user-defined conditional analysis identifiers (from @Blebowski).
• Optional support for using ZSTD to compress library files if libzstd-dev is installed.
• ISO-8859-1 extended characters are now handled properly in identifiers and when printing to the terminal.
• The new configure option --disable-default-paths disables the default library search paths (#652).
• Subtype indications used as case range choices no longer crash during analysis (#655).
• The default standard version was changed to VHDL-2002 and will likely change again to -2008 in a future release. Users are recommended to use the --std= option to specify an explicit standard revision to avoid any compatibility issues.
• Fixed a crash when elaborating a port map which contains a subtype of a record (#662).
• Implemented VHDL-2019 syntax relaxations for empty records and trailing semicolon in interface lists (from @bpadalino).
• A Bash auto-completion script is now installed by default. Run configure with --without-bash-completion to disable this.

Version 1.8.2 - 2023-02-14

• Fixed “failed to suspend thread” crash on macOS.
• Fix incorrect coverage scope nesting with array case statements (from @Blebowski).
• Expressions like FOO(X)'DELAYED now work as expected rather than producing a fatal error (#603).
• Fixed a bug where data was not propagated through inout ports in certain conditions (#609).
• The -gNAME=VALUE option to set generic values from the command line now works correctly for subtypes of enumeration types (#618).
• Fixed crash when creating an array of record subtypes where the subtype declaration has an element constraint (#615).

Version 1.8.1 - 2023-01-23

• Initial signal values for certain types were not dumped correctly in FST files on Windows (#596).
• Missing thirdparty/str-two-way.h from distribution tarball (#599).

Version 1.8.0 - 2023-01-22

• The --disable-opt and --native elaborate options which were deprecated in version 1.3 have been removed.
• The JSON dumper which was unmaintained for several years has been removed.
• VHDL-2019 mode can be enabled with --std=2019. Please note there is very limited support for this standard at present.
• The new --no-save elaboration option skips saving the elaborated design and other generated files to the working library.
• Added support for else and elsif in generate statements (#510).
• Xilinx Vivado vendor libraries can now be compiled with nvc --install vivado.
• LLVM 8.0 or later is now required due to deprecation of non-opaque pointers.
• Altera/Intel Quartus vendor libraries can now be compiled with nvc --install quartus.
• The nvc --version output now includes the commit hash if built from a Git checkout.
• The new --gtkw run option writes a .gtkw save file for GtkWave containing all the signals in the design (suggested by @amb5l).
• libffi is now a build-time dependency.
• Negation of the smallest negative value of a type such as -integer'left now produces an error.
• Default OSVVM version updated to 2022.11.
• case .. generate statements are now supported in VHDL-2008.
• Coverage implementation was reworked and now collects statement, branch, expression and toggle metrics (from @Blebowski).
• The --make command is deprecated and will be repurposed in a later release. Use the new --print-deps command instead to generate Makefile dependencies.

Version 1.7.2 - 2022-10-16

• Fixed build on FreeBSD/arm (#534).
• Fixed crash with generic package instantiation (#520).
• Now compatible with LLVM 15.0.
• Fixed calculation of longest static prefix with 'RANGE expression (#542).
• MOD and REM are now defined for physical types in VHDL-2008 (#540).
• Signal declarations are now allowed in entity declarative part (#547).
• Protected objects are now allowed in concurrent procedure calls (#547).
• Opening a file with APPEND_MODE now actually appends to the file instead of truncating (#551).

Version 1.7.1 - 2022-08-31

• Added missing textio WRITE [LINE, REAL, STRING] in VHDL-2008.
• Added support for FreeBSD/powerpc (#503, #504, from @pkubaj).
• Fixed “missing vcode unit” error during elaboration (#502).
• Fixed crash with recursive entity instantiation (#502).
• Fixed error with expressions like X'ELEMENT'LENGTH (#508).
• Added support for FreeBSD/i386.
• Fixed crash reading resolved value of record signal (#502).
• Improved folding of for-generate expressions (#514).
• Fixed memory leak when forcing signals.
• Fixed crash with type conversion in generic map (#518).
• Fixed crash with expressions like X'DELAYED'STABLE (#517).
• External names now work with record signals (#520).
• Xilinx Vivado vendor libraries can now be compiled with nvc --install vivado.
• VITAL libraries are now built by default on Windows.
• Fixed build with old versions of Glibc.
• Resolution functions with nested record types now behave correctly (#516).
• Avoid repeated stat(2) calls when accessing library.
• Fixed very slow elaboration in some cases with function calls in generic map expressions.
• Fixed intermittent segfault during elaboration (#506).
• Fixed incorrect constant folding of case statements with range choices.
• Fixed crash with constrained array declared in entity.
• Real valued signals can now be dumped in FST files (#524).
• Fixed signal assignment delay with side effects being evaluated twice (#527).
• An error is now reported for duplicate choices in array-type case statements (#528).

Version 1.7.0 - 2022-08-07

• Breaking change: In-tree builds are no longer supported: use a separate build directory instead.
• Breaking change: The --force-init command is deprecated and has no effect.
• Added support for VHDL-2008 type generics on packages and entities.
• Diagnostic messages have been enhanced with more contextual information.
• Added support for record element constraints and record fields with unconstrained array types.
• Alias of multidimensional array allowed in VHDL-2008 mode.
• Implemented VHDL-2008 rules for aggregates with slices.
• VHPI is now always enabled at build time and the --enable-vhpi configure option has no effect.
• Arithmetic operations that overflow the underlying machine type now produce an error (#101).
• Added support for VHDL-2008 force/release assignments.
• Basic support for external names in VHDL-2008.
• Matching case case? statements are supported in VHDL-2008 mode.
• Fixed several bugs in the implementation of guarded signals.
• Implemented VHDL-2008 rules for generic visibility.
• Shared variable declaration permitted in entity declaration.
• Case expression no longer requires a locally static subtype in VHDL-2008 mode (#460).
• The VHDL heap is now garbage collected as required by VHDL-2019 and the deallocate operator has no effect other than setting the access to null.
• A new global option -H specifies the size of the simulation heap and defaults to 16 megabytes.
• Concurrent procedure call allowed in entity statement part.
• Added support for 'SUBTYPE and 'ELEMENT attributes in VHDL-2008.
• The new top-level --init command creates a new empty library directory.
• The -a analysis command now reads from the standard input if the file name is -.
• Added support for array element constraints in VHDL-2008.
• The --prefer-explicit analysis option which was deprecated before the 1.0 release has been removed.
• A new --relaxed analysis option enables “relaxed rules” mode. This has the same effect as enabling all the existing --relax= options. However some constructs will still produce warnings.
• The --relax= analysis option is deprecated and is now equivalent to passing --relaxed. The individual options are ignored.
• Added support for generic subprograms in VHDL-2008.
• New command --install allows easy installation of common third-party packages such as OSVVM and UVVM.
• Identifiers in waveform dumps are now in lower case instead of upper case.
• The function CURRENT_DELTA_CYCLE in NVC.SIM_PKG can be used to query the current delta cycle number.

Version 1.6.2 - 2022-04-03

• Fix make -j with GNU make (#440).
• Subtraction from zero could return the wrong result in some circumstances.
• Fix incorrect code generation for access-to-array.
• Fix assertion failure with nested context.
• Generic declarations are now allowed to hide the entity name.
• Fix spurious error about duplicate declarations when using VHDL-2008 context declarations.
• Fix incorrect default generic value when component and entity specify generics in different order (#448).
• Postponed processes with sensitivity lists now work correctly.
• Unit tests no longer fail when built without debug symbols (#445).
• Simulation now correctly terminates when time reaches TIME'HIGH.
• Assertion failure and report messages were not printed during the initialisation phase of the simulation.

Version 1.6.1 - 2022-02-05

• Fix compatibility with BSD make (#440).
• Fix an out-of-memory condition when initialising processes (#441).
• Fix incorrect generic value in binding specification (#442).
• Fix spurious error in overload resolution if required arguments follow optional arguments (#443).
• Fix intermittent crash generating code for enumerated types (#444).
• Minor elaboration performance optimisation.

Version 1.6.0 - 2022-01-25

• Name resolution and overload resolution has been completely rewritten which should fix a number of long-standing issues.
• The elaboration phase was largely written which fixes a number of long-standing issues and significantly improves elaboration speed.
• VHDL-2008 IEEE standard libraries are now built and installed in addition to the VHDL-1993 libraries.
• The VHDL-1993 standard libraries are now derived from the Apache 2.0 licensed sources from VHDL-2019.
• There is now a project website at https://www.nickg.me.uk/nvc/. Please link to this in preference to the GitHub project page.
• Compiled VHDL code now includes DWARF debug information which is used for runtime stack trace if libdw or libdwarf is installed.
• Added support for VHDL-2008 reduction operators, match operators, and condition conversion.
• Added support for VHDL-2008 element resolution.
• Variable assignment now supports aggregate targets.
• The --relax=impure option allows pure functions to call impure functions.
• Added support for VHDL-2008 “all” sensitised processes.
• Added support for ports and generics in block statements.
• Added support for the ’BASE attribute.
• Type name now allowed in element association choice (#407).
• Implement textio READ procedure for REAL.
• LLVM 6.0 or later is now required to build.
• Added support for MINIMUM, MAXIMUM, and TO_STRING predefined operators in VHDL-2008.
• VCD files are now generated from FST data in a similar manner to fst2vcd(1). This should improve compatibility with other tools.
• Added support for ’LAST_ACTIVE attribute (#423).
• Added support for ’DRIVING and ’DRIVING_VALUE attributes.
• Added a new option --ieee-warnings=off to disable warning messages from the standard IEEE packages.
• Support for configurations has been significantly improved (#372).
• Added support for VHDL-2008 delimited comments.
• Added support for guard expressions on blocks.
• Added support for guarded signals.
• Added support for HREAD, HWRITE, and other TEXTIO additions in VHDL-2008.
• Code generation now happens in parallel when LLVM is built with multi-threading enabled.
• Link time optimisation (LTO) is now enabled for release builds where supported.
• The default assertion failure message for certain simple scalar comparisons now shows the values of the left and right hand sides.
• Added support for VHDL-2008 conditional variable assignment statements.
• Added support for VHDL-2008 extended bit string literals.
• Non-globally-static actuals allowed in port maps in VHDL-2008 mode.
• Added support for VHDL-2008 sequential conditional signal assignment statements.
• Added basic support for package generics and package instantiation.
• Nested arrays can now be included in the waveform dump but only if the --dump-arrays option is passed. This is disabled my default due the significant performance and memory overhead.
• Added support for record types in waveform dump (#216).
• Added support for foreign subprograms using the VHPIDIRECT protocol.
• Library build is now reproducible when running make -j.
• Fix a constant folding crash with nested records.
• Fixed a crash when a record aggregate contains an “others” association and the fields have array types with different lengths.
• Fixed a stack overflow when a subprogram with unconstrained array arguments is called repeatedly in a loop (#414).
• Fixed intermittent crash when evaluating nested constant records (#425).
• Fixed missing import libraries on Windows (#424).
• Standard libraries are now installed under $prefix/lib/nvc/ instead of $prefix/share/nvc.
• New configure option --disable-vital disables building the VITAL packages whose license status is unclear.
• Support for the LXT wave output format, which was deprecated in version 1.5, has been removed. Use the default FST format instead.
• The fetch-ieee.sh script which did nothing since the last release has been removed.
• The --codegen command, which has been deprecated since 1.3, was removed.
• The --profile option now prints internal simulation statistics instead of the top processes by CPU time.

Version 1.5.3 - 2021-11-13

• Handle access(2) returning EPERM in macOS sandbox (#421).
• Fix race when multiple processes concurrently update a library.
• Fix --syntax command when file contains multiple design units.
• Allow constant folding of nand/nor/xor/xnor.
• Fix potential out of memory condition when evaluating complex assert expressions.
• Fix incorrect result of mod operator with negative operands.
• Fixed intermittent crash when evaluating nested constant records (#425).
• Buffer too small for printing TIME’HIGH (#98).

Version 1.5.2 - 2021-07-28

• Link libexecinfo on FreeBSD.
• Implement textio READ procedure for BIT and TIME (#408).
• Fixed a crash when a long running procedure suspends in a loop (#412).
• Fix static linking with LLVM 12.0.
• Fix crash when assigning to a signal declared in a package.
• Fix incorrect recording of dependencies which caused a failure to load generated DLLs on Windows.
• Fix file locking error when a library is located on NFS (#417).
• Optimise loading large library index from disk.
• Fix a crash when using ’VALUE with enumeration subtypes (#419).
• Fix a crash when a signal with more than 256 elements is declared in a package (#420).

Version 1.5.1 - 2021-04-09

• Fix a compiler warning in vcode.c.
• Disable VHDL backtrace on non-Linux systems as the symbol names cannot be parsed reliably (#385).
• Update to latest pc_from_uncontext.m4 for Apple M1 support.
• Fix incorrect application of LRM rules for building an equivalent wait statement for concurrent statements.
• Library build is now reproducible when running make -j (#409).
• Fix assertion failure with nested record type (#404).
• Use Pandoc to generate the manual page.

Version 1.5 - 2020-07-19

• IEEE library sources are now distributed
• Updated FST library to match GtkWave 3.3.79
• The LXT wave output format is deprecated, use FST instead
• Fix incorrect file name in assertion message
• Fix crash while recovering from parse error
• Add --dump-json command to print AST as JSON (from Sebastien Van Cauwenberghe)
• Fix crash when using LLVM 7 and later
• Fix spurious assertion failure in std.textio.readline
• Reals are now rounded to the nearest integer as specified by the LRM
• Fix crash when constant folding uses too much memory
• Improved memory management in evaluator (thanks to Frank Mori Hess)
• Various other minor fixes and improvements

Version 1.4 - 2018-07-16

• Windows with MSYS2 is now fully supported
• Continuous integration for OS X and Windows using Travis and AppVeyor
• Record resolution functions are now supported
• The --relax=prefer-explicit option now works correctly

Version 1.3 - 2017-12-31

• Native code is always generated on platforms that support it
• Initial support for VHDL-2017 conditional compilation blocks
• The TCL shell has been removed
• Various code generation and constant folding fixes
• Fix crash when build IEEE library from non-clean tree
• ORC is used for JIT if LLVM version is 3.9 or later
• LLVM 3.8 or later is now required
• Added -O0 to -O3 options to control LLVM optimisation level
• The --disable-opt option is deprecated, use -O0 instead
• The elaborate --native option is now a global option and should be placed before the -e command
• The --codegen command is deprecated, use --native when analysing to generate native shared libraries for packages

Version 1.2 - 2017-05-04

• Compile time evaluation now uses vcode
• Record aggregates can now be constant folded (#305)
• The TCL shell is deprecated and will be removed in the next release
• Compatible with LLVM 4.0
• Functions returning records can now be constant folded (#315)
• LLVM 3.4 or later is now required

Version 1.1 - 2016-11-05

• Improvements to based literal parsing
• Improved stack traces on Linux when libdw-dev installed
• Fixed elaboration of subprograms in entity declarations
• Improved elaboration performance when design contains many signals
• Fixed various bugs involving record signals
• Fixed sensitivity list generation for concurrent procedure calls
• Improvements to simulation performance
• Implemented vhpi_get_phys
• Added command --list to print all units in a library
• Rewrote the run_regr program in C so make check no longer requires Ruby
• The --relax=generic-static option is now renamed to --relax=locally-static and applies in wider range of contexts

Version 1.0 - 2016-05-01

• First stable release