Viua VM
Update issue repository
Remove the accidental io_close
Merge branch 'master' into devel


browse log



You can also use your local clone with git send-email.

Viua VM

A register-based, parallel virtual machine programmable in custom assembly lookalike language with strong emphasis on reliability, predictability, and concurrency.

Building the code
]$ ./configure      # (1) configure the project
]$ make -j 4        # (2) build it using 4 jobs

Note: the default build is run with -O0 (without any optimisations). You can compile Viua VM with higher optimisation leves, but keep in mind that the compilation might not succeed on machines with 4GB or less RAM (with -O3). This is tracked by issue #184 on GitHub, or by issue 7c06177872c3a718510a54e6513820f8fe0fb99b in the embedded issue repository.

Hello World in Viua VM
.function: main/0
    allocate_registers %3 local

    text %1 local "Hello World!\n"
    integer %2 local 1

    io_write %2 local %2 local %1 local
    io_wait void %2 local 1s

    izero %0 local

See README.asm file to see commented version of this sample program.

For more examples visit either documentation or Rosetta page for Viua VM. Check Weekly blog for news and developments in Viua VM.

Use cases

Viua is a runtime environment focused on reliability, predictability, and concurrency. It is suitable for writing long-running software that runs in the background providing infrastructure services:

  • servers (e.g. IRC)
  • message queues
  • system and application daemons (e.g. cron)

The VM is not ready to be used in places where performance matters the most. It is best to employ it for tasks which do not require human interaction or near real-time responses.

The VM is able to fully utilise all cores of the CPU it's running on (tested on a system with 8 hardware threads) so can generate high CPU loads, but is relatively light on RAM and should not contain any memory leaks (all runtime tests are run under Valgrind to ensure this).

The virtual machine is covered by more than 500 tests to provide safety, and guard against possible regressions. It ships with an assembler and a static analyser, but does not provide any higher-level language compiler.

Design goals
  • predictable execution: it is easier to reason about code when you know exactly how it will behave
  • predictable value lifetimes: in Viua you do not have to guess when the memory will be released, when objects will be destroyed, or remember about the possibility of a gargabe collector kicking in and interrupting your program; Viua manages resources without a GC in a strictly scope-based (where scope means virtual stack frame) manner
  • massive concurrency: Viua architecture supports spawning massive amounts of independent, VM-based lightweight processes that can run in parallel (Viua is capable of providing true parallelism given sufficient hardware resources, i.e. at least two CPU cores)
  • parallel I/O and FFI schedulers: I/O operations and FFI calls cannot block the VM as they are executed on dedicated schedulers, so block only the virtual process that called them without affecting other virtual processes
  • easy scatter/gather processing: processes communicate using messages but machine supports a join instruction - which synchronises virtual processes execution, it blocks calling process until called process finishes and receives return value of called process
  • safe inter-process communication via message-passing (with queueing)
  • soft-realtime capabilities: join and receive operations with timeouts (throwing exceptions if nothing has been received) make Viua a VM suitable to host soft-realtime programs
  • fast debugging: error handling is performed with exceptions (even across virtual processes), and unserviced exceptions cause the machine to generate precise and detailed stack traces; running programs are also debuggable with GDB
  • reliability: programs running on Viua should be able to run until they are told to stop, not until they crash; machine helps writing reliable programs by providing a framework for detailed exception-based error communication and a way to register a per-process watchdog that handles processes killed by runaway exceptions (the exception is serviced by watchdog instead of killing the whole VM)

Some features also supported by the VM:

  • separate compilation of Viua code modules
  • static and dynamic linking of Viua-native libraries
  • straightforward ways to use both dynamic and static function call dispatch
  • first-class functions
  • closures (with multiple way of capturing objects inside a closure)
  • passing function parameters by value, by move (non-copying pass), and by pointer
  • copy-free function returns
  • inter-function tail calls
  • several variants of fixed-size integer arithmetic: wrapping, checked, and saturating

For enhanced reliability, Viua assembler provides a built-in static analyser that is able to detect most common errors related to register manipulation at compile time (type mismatches, invalid register access). It provides traced errors whenever possible, i.e. when it detects an error and is able to trace execution path that would trigger it, a sequence of instructions (with source code locations) leading to the detected error is presented to the user instead of a single offending instruction.

Current limitations include:

  • severly limited introspection,
  • calling Viua code from C++ is not tested
  • debugging information encoded in compiled files is limited
  • speed: Viua is not the fastest VM around
  • VM cannot distinguish FFI calls that are CPU or I/O bound (WIP), so if the program performs many I/O operations it may saturate all FFI schedulers (which will effectively prevent the program from quickly executing more FFI calls; virtual processes are still running, and Viua functions are still freely callable, though)
  • lack of libraries
  • no easy way to perform I/O
Software state notice

Viua VM is an alpha-stage software. Even though great care is taken not to introduce bugs during development, it is inevitable that some will make their way into the codebase. Viua in its current state is not production ready; bytecode definition and format will be altered, opcodes may be removed and added without deprecation warning, and various external and internal APIs may change without prior notice. Suitable announcements will be made when the VM reaches beta, RC and release stages.


The way Viua works has mostly been influenced by C++ (static and dynamic method dispatch, move semantics), and Erlang (message passing, indepenedent VM-based lightweight processes as units of concurrency).

Programming in Viua

Viua can be programmed in an assembly-like language which must be compiled into bytecode. A typical session is shown below (assuming current working directory is the local clone of Viua repository):

]$ vi some_file.asm
 ]$ ./build/bin/vm/asm -o some_file.out some_file.asm
# static analysis or syntax errors...
 ]$ vi some_file.asm
 ]$ ./build/bin/vm/asm -o some_file.out some_file.asm
 ]$ ./build/bin/vm/kernel some_file.out
# runtime exceptions...
 ]$ vi some_file.asm
 ]$ ./build/bin/vm/asm -o some_file.out some_file.asm
 ]$ ./build/bin/vm/kernel some_file.out


Please read CONTRIBUTING.markdown for details on development setup, and the process for submitting pull requests, bug reports, and feature suggestions.

The code can be cloned either from Savannah or from Sr.ht.


The code is licensed under GNU GPL v3.

Contact information

Project website: viuavm.org

Maintainer: <marekjm at ozro dot pw>

Bugtracker: https://github.com/marekjm/viuavm/issues