Particle developers should visit our particle developer website. This document is targeted at Arcs core system developers.
A hosted version of Arcs is available in both tagged and bleeding edge forms. Neither is stable -- the runtime, database and front-end are all iterating rapidly.
Tagged release URLs have the form
https://cdn.rawgit.com/PolymerLabs/arcs-live/<release_number>/shell/apps/web/index.html
(the list of releases is
here). A recent version
(latest as of this writing) is
v0.4.1.
Bleeding edge often works and is available via github pages: https://polymerlabs.github.io/arcs-live/shell/apps/web/.
Arcs is developed with a recent version of Node (v10.0.0 at the time of this writing), in particular as we use new ES6 features. You can check our Travis config to see what version is used for automated build status. More recent versions should work, but if for example you see test errors on a version that's a full release later (ex. v11+) you may want to try rolling back to an earlier version. We welcome patches that will allow more recent versions to operate, ideally without requiring an upgrade to our current version.
-
Install nvm.
As per the installation instructions, download and run the installation script directly in your terminal (yes, you read that correctly):
$ curl -o- https://raw.githubusercontent.com/creationix/nvm/v0.33.11/install.sh | bash
If you're using zsh you may need to
source ~/.nvm/nvm.sh
after this. -
Install node.
$ nvm install 10
-
If you need to update npm to a later version (our build checks for the minimum required version):
$ npm install -g npm # can use [email protected] to install a specific version
$ npm install
$ ./tools/sigh
npm install
is required on a fresh checkout. After that it only needs to be
re-run infrequently as new dependencies are included, and usually a build
failure will be the signal for that.
- Git for Windows is one of many Git options.
- Consider using nvm-windows to allow more easily switching between Node versions.
- As part of
npm install
you'll need to buildfibers
which usesnode-gyp
which in turn requireswindows-build-tools
. Follow the node-gyp Windows build instructions. If option 1 hangs or otherwise hits issues, you can try option 2. Note the Microsoft Build Tools 2015 can be downloaded separately from Visual Studio (and the links in thenode-gyp
documentation are stale), but you'll still need to do thenpm config set msvs_version 2015
bit, and similar for Python if you install that manually pernode-gyp
option 2 instructions.
After the full build (npm install && tools/sigh
) run: (note that npm start
will block, so you'll have to run the second command in a new shell):
$ npm start
Then open http://localhost:8080/shell/apps/web/index.html
in a web browser
(or, on MacOS, use open 'http://localhost:8080/shell/apps/web/index.html'
).
Subcomponents have more detailed descriptions. Particularly the extensions also have individual installation steps.
For more information on the shell, see shell.
See extension.
See devtools.
The simplest way to run tests is to let the targets do all the work. These commands will install all packages, run a build, start a background server, run all the tests, and kill the background server:
$ npm install
$ npm run test-with-start
There are additional targets provided to run subsets of those commands.
npm start
: spins up a server (and blocks), serving in port 8080.tools/sigh
: run a subset of tests and build packed artifacts.npm test
: run all tests (using currently built artifacts) against an already-running server (assumed to be port 8080).npm run test-no-web
: run all non-web tests.
To run a specific Selenium test using Mocha's 'grep' capability:
- In one terminal:
npm start
- In another:
npm run test-wdio-shells -- --mochaOpts.grep 'regex'
If you see errors like
ERROR: connect ECONNREFUSED 127.0.0.1:9515
chrome
at new RuntimeError (...\node_modules\webdriverio\build\lib\utils\ErrorHandler.js:144:12)
at Request._callback (...\node_modules\webdriverio\build\lib\utils\RequestHandler.js:327:43)
It may indicate that chromedriver hasn't been installed completely. Run the install script:
node node_modules\chromedriver\install.js
Selenium failures are often easy to cause due to seemingly unrelated changes, and difficult to diagnose.
There are 2 main avenues to debug them in this system. The first is to have the browser run in a graphical manner (as opposed to the default headless configuration). The second is to actually debug the running selenium instance.
There are some debugging hints (code and configuration you can uncomment to
make debugging easier) in test/specs/starter-test.js
and test/wdio.conf.js
marked with the phrase debug hint
.
To activate a sane set of helpful debugging flags, there's a wdio-debug
command line argument that you can pass in. This will run Chrome in a
non-headless fashion, and will increase timeouts.
Through npm: npm run test-wdio-shells --wdio-debug=true
(or npm test --wdio-debug=true
). Directly through wdio: node_modules/.bin/wdio --wdio-debug=true shell/test/wdio.conf.js
.
Webdriver takes screenshots of failures, which are saved to the
./shell/test/errorShots/
directory. When running on Travis, the screenshots
are uploaded to the Arcs Webdriver Screenshots
team drive.
It may be easiest to see the problem in a browser window to diagnose it. Edit
wdio.conf.js
in the branch with failures, comment out the '--headless'
option and increase the mocha timeout. In combination, these two changes will
allow you to see what's happening on the screen, and will give you enough time
to debug the situation.
arcs/shell> vi test/wdio.conf.js
arcs/shell> git diff test/wdio.conf.js
diff --git a/test/wdio.conf.js b/test/wdio.conf.js
index 0e36452..8ecf3d6 100644
--- a/test/wdio.conf.js
+++ b/test/wdio.conf.js
@@ -50,7 +50,7 @@ exports.config = {
chromeOptions: {
args: [
// arcs note: comment this out to see the system running
- '--headless'
+ // '--headless'
]
}
}
@@ -139,7 +139,7 @@ exports.config = {
mochaOpts: {
ui: 'bdd',
// arcs note: increase this timeout for debugging
- timeout: 20003
+ timeout: 2000003
}
//
// =====
Then, in your test, you can add a breakpoint (via browser.debug();
) to pause
execution so you can debug in the browser. It may be worthwhile to add several
browser.debug()
invocations through your flow to trace execution (.exit
will exit the debugger and continue execution of the test).
At that point you can open up DevTools in the browser to debug the current
state, or inspect it visually. Some utilities (those in selenium-utils.js
,
including pierceShadows
) have already been loaded.
There are also some commands available natively at that point, including
.help
and the browser
variable (including methods such as
browser.execute()
).
To attach a debugger, uncomment the execArgv
--inspect
configuration option.
It's likely that you'll still want to have increased the mochaTimeout
and to
be running graphically, so those are in the example as well:
arcs/shell> git diff test/wdio.conf.js
diff --git a/test/wdio.conf.js b/test/wdio.conf.js
index 0e36452..4240c0a 100644
--- a/test/wdio.conf.js
+++ b/test/wdio.conf.js
@@ -50,11 +50,12 @@ exports.config = {
chromeOptions: {
args: [
// arcs note: comment this out to see the system running
- '--headless'
+ // '--headless'
]
}
}
],
+ execArgv: ['--inspect'],
//
// ===================
// Test Configurations
@@ -139,7 +140,7 @@ exports.config = {
mochaOpts: {
ui: 'bdd',
// arcs note: increase this timeout for debugging
- timeout: 20003
+ timeout: 2000003
}
//
// =====
When starting, you should see log item like debugger listening on ws://127.0.0.1:9229/..
as normally appears for node
debugging. Passing the --inspect
argument will also enable the V8 Inspector
Integration which may be easier to use
(to activate this, look for a node icon in a Chrome DevTools process).
Adding debugger;
statements may be the easiest way to activate the debugger.
Using browser.debug();
statements to pause execution to give you time to
attach a debugger may be helpful as well.
Our release process is pretty minimal, but requires a few steps across the arcs and arcs-live repositories.
Our standard is to have the stable versions start with clean (empty) databases, but to continue a single mainline/unstable database.
-
Decide what your new mainline and stable versions will be. For an example here, I'll use
0.3.5-alpha
as the old mainline,0.3.6-alpha
as the new mainline, and0.3.5
as the new stable version. -
In order to keep the mainline data roughly consistent, clone the data at the current firebase key to the new mainline release number. To do this, I used the firebase web interface to "Export JSON" for the current tree, and "Import JSON" to the new tree.
For example, clone from <snip>/database/arcs-storage/data/0_3_5-alpha
to
<snip>/database/arcs-storage/data/0_3_6-alpha
.
If the web interface is read-only due to too many nodes, you can visit the new version's URL directly to Import JSON.
-
Update the version in
shell/apps/common/firebase-config.js
to a reasonable stable version (in our example,0.3.5
). See #1114 for an example. Update the links README.md (this file) to reflect this new version. -
Once the deploy is done to arcs-live, create a new release. Note that we remap the versions slightly between the two systems for legibility in different systems - a version of
0_3_5
(infirebase-config.js
) becomesv0.3.5
(in the arcs-live repo). -
Update the version in
shell/apps/common/firebase-config.js
to the new mainline development version (perhaps using the-alpha
suffix; in our example,0.3.6-alpha
). See #1155 for an example.