Nix is an innovative, robust, fast, and high-quality solution that comes with a larger collection of packages and features than Conda. However, Windows support is poor (using it with Windows Subsystem for Linux still has many problems) and Nix can be harder to learn.
Conda has a more traditional approach to package management, is much more limited, slow, and lower-quality than Nix, but it supports Windows and it is simpler to use when it functions correctly.
Make sure you are using a 64-bit x86 Linux system. If you are using other systems, such as 32-bit x86, Nix will attempt to compile a number of dependencies from source on your machine. This may work, but the installation process will use a lot of CPU time, memory, and disk space.
First, install the Nix package manager. Some distributions provide a package for the Nix package manager, otherwise, it can be installed via the script on the `Nix website <http://nixos.org/nix/>`_.
Those channels track `nixpkgs 19.03 <https://github.com/NixOS/nixpkgs/tree/release-19.03>`_. You can check the latest status through the `Hydra interface <https://nixbld.m-labs.hk>`_. As the Nix package manager default installation uses the development version of nixpkgs, we need to tell it to switch to the release: ::
Nix won't install packages without verifying their cryptographic signature. Add the M-Labs public key by creating the file ``~/.config/nix/nix.conf`` with the following contents:
The easiest way to obtain ARTIQ is then to install it into the user environment with ``$ nix-env -f "<m-labs>" -iA artiq-env``. This provides a minimal installation of ARTIQ where the usual commands (``artiq_master``, ``artiq_dashboard``, ``artiq_run``, etc.) are available.
If you are getting the error message ``file 'm-labs' was not found in the Nix search path``, you are probably encountering `this Nix bug <https://github.com/NixOS/nix/issues/2709>`_. As a workaround, enter the command ``$ export NIX_PATH=~/.nix-defexpr/channels:$NIX_PATH`` and try again.
This installation is however quite limited, as Nix creates a dedicated Python environment for the ARTIQ commands alone. This means that other useful Python packages that you may want (pandas, matplotlib, ...) are not available to them, and this restriction also applies to the M-Labs packages containing board binaries, which means that ``artiq_flash`` will not automatically find them.
Installing multiple packages and making them visible to the ARTIQ commands requires using the Nix language. Create a file ``my-artiq-env.nix`` with the following contents:
::
let
# Contains the NixOS package collection. ARTIQ depends on some of them, and
# you may also want certain packages from there.
pkgs = import <nixpkgs> {};
# Contains the main ARTIQ packages, their dependencies, and board packages
Then spawn a shell containing the packages with ``$ nix-shell my-artiq-env.nix``. The ARTIQ commands with all the additional packages should now be available.
You can exit the shell by typing Control-D. The next time ``$ nix-shell my-artiq-env.nix`` is invoked, Nix uses the cached packages so the shell startup is fast.
You can edit this file according to your needs, and also create multiple ``.nix`` files that correspond to different sets of packages. If you are familiar with Conda, using Nix in this way is similar to having multiple Conda environments.
After installing either Anaconda or Miniconda, open a new terminal (also known as command line, console, or shell and denoted here as lines starting with ``$``) and verify the following command works::
Executing just ``conda`` should print the help of the ``conda`` command. If your shell does not find the ``conda`` command, make sure that the Conda binaries are in your ``$PATH``. If ``$ echo $PATH`` does not show the Conda directories, add them: execute ``$ export PATH=$HOME/miniconda3/bin:$PATH`` if you installed Conda into ``~/miniconda3``.
Download the `ARTIQ installer script <https://raw.githubusercontent.com/m-labs/artiq/master/conda/install-artiq.py>`_ and edit its beginning to define the Conda environment name (you can leave the default environment name if you are just getting started) and select the desired ARTIQ packages. Non-ARTIQ packages should be installed manually later. If you do not need to flash boards, the ``artiq`` package from the ``main`` Hydra build is sufficient.
Controllers for third-party devices (e.g. Thorlabs TCube, Lab Brick Digital Attenuator, etc.) that are not shipped with ARTIQ can also be installed with this script. Browse `Hydra <https://nixbld.m-labs.hk/>`_ to find the names of the corresponding packages, and list them at the beginning of the script.
Run ``$ nix-channel --update`` to retrieve information about the latest versions, and then either reinstall ARTIQ into the user environment (``$ nix-env -i python3.6-artiq``) or re-run the ``nix-shell`` command.
To rollback to the previous version, use ``$ nix-channel --rollback`` and then re-do the second step. You can switch between versions by passing a parameter to ``--rollback`` (see the ``nix-channel`` documentation).
When upgrading ARTIQ or when testing different versions it is recommended that new Conda environments are created instead of upgrading the packages in existing environments.
To install the latest version, just select a different environment name and run the installer script again.
Switching between Conda environments using commands such as ``$ conda deactivate artiq-6`` and ``$ conda activate artiq-5`` is the recommended way to roll back to previous versions of ARTIQ.
You may need to reflash the gateware and firmware of the core device to keep it synchronized with the software.
If you have purchased a pre-assembled system from M-Labs or QUARTIQ, the gateware and firmware are already flashed and you can skip those steps, unless you want to replace them with a different version of ARTIQ.
With Nix, add ``m-labs.openocd`` to the shell packages. Be careful not to add ``pkgs.openocd`` instead - this would install OpenOCD from the NixOS package collection, which does not support ARTIQ boards.
With Conda, the ``artiq`` package installs ``openocd`` automatically but it can also be installed explicitly on both Linux and Windows::
On Linux, first ensure that the current user belongs to the ``plugdev`` group (i.e. ``plugdev`` shown when you run ``$ groups``). If it does not, run ``$ sudo adduser $USER plugdev`` and re-login.
If you installed OpenOCD on Linux using Conda and are using the Conda environment ``artiq``, then execute the statements below. If you are using a different environment, you will have to replace ``artiq`` with the name of your environment::
For Kasli, insert a SFP/RJ45 transceiver (normally included with purchases from M-Labs and QUARTIQ) into the SFP0 port and connect it to a gigabit Ethernet port in your network. It is necessary that the port be gigabit - 10/100 ports cannot be used. If you need to interface Kasli with 10/100 network equipment, connect them through a gigabit switch.
If you purchased a device from M-Labs, it already comes with a valid MAC address and an IP address, usually ``192.168.1.75``. Once you can reach this IP, it can be changed with: ::
Check that you can ping the device. If ping fails, check that the Ethernet link LED is ON - on Kasli, it is the LED next to the SFP0 connector. As a next step, look at the messages emitted on the UART during boot. Use a program such as flterm or PuTTY to connect to the device's serial port at 115200bps 8-N-1 and reboot the device. On Kasli, the serial port is on FTDI channel 2 with v1.1 hardware (with channel 0 being JTAG) and on FTDI channel 1 with v1.0 hardware.
The idle kernel is the kernel (some piece of code running on the core device) which the core device runs whenever it is not connected to a PC via Ethernet.
This kernel is therefore stored in the :ref:`core device configuration flash storage <core-device-flash-storage>`.
To flash the idle kernel, first compile the idle experiment. The idle experiment's ``run()`` method must be a kernel: it must be decorated with the ``@kernel`` decorator (see :ref:`next topic <connecting-to-the-core-device>` for more information about kernels). Since the core device is not connected to the PC, RPCs (calling Python code running on the PC from the kernel) are forbidden in the idle experiment. Then write it into the core device configuration flash storage: ::
The startup kernel is executed once when the core device powers up. It should initialize DDSes, set up TTL directions, etc. Proceed as with the idle kernel, but using the ``startup_kernel`` key in the ``artiq_coremgmt`` command.
For DRTIO systems, the startup kernel should wait until the desired destinations (including local RTIO) are up, using :meth:`artiq.coredevice.Core.get_rtio_destination_status`.
If you are using DRTIO and the default routing table (for a star topology) is not suitable to your needs, prepare and load a different routing table. See :ref:`Using DRTIO <using-drtio>`.
