The tilde is ambiguous and it can be helpful to know exactly what paths
code-server is trying to use, especially if it is running as a different
user than you expected.
* Avoid spawning code-server with --reuse-window and --new-window
These flags mean the user explicitly wants to open in an existing
instance so if the socket is down it should error rather than try to
spawn code-server normally.
* Set session socket into environment variable
While I was at it I added a CLI flag to override the default. I also
swapped the default to --user-data-dir.
The value is set on an environment variable so it can be used by the
extension host similar to VSCODE_IPC_HOOK_CLI.
* Add e2e test for opening files externally
The indented logs technically apply to the HTTP server so move the
session server log afterward to avoid making them look like they apply
to the session server.
Mostly just the usual shifting or changing of the surrounding context
but I did refactor the getting started block we insert because it keeps
getting mangled on each update. Instead of shifting things around the
columns I just prepend it to the right column.
Getting 404s on some vsda module but everything seems to work
without it and I do not see it referenced in the package.json nor a
nywhere on npmjs.com so it seems to be optional.
This is not a valid ID (to install a specific version you use @) and,
quite strangely, Code is now returning an "extension not found" error if
the ID is invalid, breaking the test since we expect it to error about
the marketplace not existing.
* Update to 1.78.1
No changes needed in the patches other than moving some lines around and
updating the CSP hash as usual.
The flake had to be updated as it was using Node 16.16 and 16.17 is
required at minimum now. Also python seems to install python2 which is
marked as deprecated so explicitly install python3.
* Update to 1.78.2
Patches applied without any conflicts.
* Update commit environment variable
This was causing the commit not to be set. It broke display languages
since that has a hard dependency on the commit for directory names.
Possibly broke other things.
Extracted host detection into a separate function to avoid multiple log
lines on each return and went with a thrown error to consolidate the
common log text.
* Update Code to 1.76.1
- worker-src already contains blob so we can avoid patching that.
- localeService moved.
- Remaining changes were just line changes.
* Make language extensions installable again
Still might want to look into making the native language support work
but for now it seems better not to break backwards compatibility since
the native implementation is quite different.
* Avoid "install in browser" for language packs
It will not work.
* Import correct locale service
I believe before the contributions imported this but now we have to do
it here.
Looks like the images got updated to v18 so they started failing. For
npm install v16 and for Docker just run the script directly, it seems
silly to waste time installing v16 just to run a script through yarn.
* Move splitOnFirstEquals to util
I will be making use of this to parse the forwarded header.
* Type splitOnFirstEquals with two items
Also add some test cases.
* Check origin header on web sockets
* Update changelog with origin check
* Fix web sockets not closing with error code
- Move differences to the Codespaces section since they apply to both
Codespaces and OpenVSCode-Server
- Add some important missing differences
- Exclude settings sync (not being worked on)
- Exclude the plugin API (deprecated)
- Exclude certificate support (browsers these days are starting to
require trusted certs so better not to recommend using this)
This does not seem to actually cause an issue (not resolving ends up
with the same behavior as resolving with false) but I am not sure if the
hanging promises would be a memory leak so seems best to fix.
* Update Code to 1.75.0
- getting-started.diff: The way to get an icon's class changed
- proxy-uri.diff: The product service is passed in so we can get the
proxy URI from that now instead of passing it in separately.
* Remove workspace trust test
Something in how/when Code displays the trust dialog appears to have
changed, failing the test. I am not sure it makes sense for us to be
testing upstream code anyway.
* Use regular Node for watch
Since we spawn the watch script with ts-node it was using ts-node for
the web server spawn as well. With latest Code there are for some
reason type errors (it cannot find @types/node) but this is already
compiled code which already passed type checks; any type errors here are
useless. To fix spawn with regular Node.
* Fix some workers not loading
This is failing CI on Dependabot PRs. Opted to just remove it since
most (all?) PRs will be from forks and this workflow will not run
anyway. If we figure out the secret situation we can add it back.
It is causing CI to fail for Dependabot (no access to the token) and it
does not work with forks and currently there is no one who pushes
straight to the repo so this will never be used.
Can always add it back if we figure out how to make the secrets work.
* Use fork instead of spawn
We no longer do in-place updating so no need for the spawn. The
advantage of a fork is that it preserves flags like --prof which you can
use to profile code-server.
Also I am not sure the comment about not being able to reload in place
with fork was even true to begin with.
* Refresh heartbeat patch
Seems to have gotten out of date a little.
* Propagate execArgv to extension host
This will let us profile the extension host.
* Include bin scripts for all platforms
These will get symlinked as part of the postinstall. These scripts
provide everything ours does inside the integrated terminal plus more.
* Improve OS detection
Specifically for Windows although we do not yet support Windows.
Also standardize the duplicate arch functions since they had drifted
from each other bit.
* Remove duplicate asar symlink
Since standalone releases run the postinstall they will get the asar
symlink there. That means the symlink will not exist for the npm
package and we will not need to ignore it.
The symlink portion is split out so it can be re-used for other
symlinks (for example linking bin scripts).
* Add symlinks to bin scripts
* Add test for opening a file from the terminal
* Add global Playwright timeout
Otherwise it will exceed the Actions timeout and get rudely killed
without any output.
* Make sed work on macOS
* Fix Node path in bin scripts
* Disable shellcheck expansion error
* Make scripts executable
* Remove .bak files created by sed
* Include Code build script in cache hash
Otherwise if we change the script it will not rebuild Code.
* Make sure the terminal opens
The selector was timing out even though it matched more than one element
but matching on the focused one appears to work.
In addition add a loop so it can keep trying to open the terminal
if something goes wrong with the focus.
Having NODE_OPTIONS set is unexpected and although the later flag should
override the previous flag it is not certain that will always be the
case.
Also some users are having issues with the 2 GB limit.
* chore: update Code to 1.67
Was able to remove our changes to common/webview.ts since they are
upstream now.
Other than that no serious changes, just context diffs.
* chore: update Code to 1.68
- Upstream moved the web socket endpoint so change the Express route
from / to *. That will let web sockets work at any endpoint.
- Everything in the workbench config is basically the same but
de-indented (upstream extracted it into a separate object which
resulted in a de-indent), the ordering is slightly different, and
instead of vscodeBase we now need vscodeBase + this._staticRoute since
everything is served from a sub-path now.
- Move manifest link back to the root since that is where we host our
manifest.
- Change RemoteAuthoritiesImpl to use the same path building method as
in other places (+ instead of using URI.parse/join).
- Use existing host/port in RemoteAuthoritiesImpl and
BrowserSocketFactory instead of patching them to use window.location
(these are set from window.location to begin with so it should be the
same result but with less patching).
- Since BrowserSocketFactory includes a sub-path now (endpoints were
changed upstream to serve from /quality/commit instead of from the
root) the patch there has changed to prepend the base to that
path (instead of using the base directly).
- The workbench HTML now natively supports a base URL in the form of
WORKBENCH_WEB_BASE_URL so no need for VS_BASE patches there anymore.
- Upstream added type="image/x-icon" so I did as well.
- Move the language patch to the end of the series so it is easier to
eventually remove.
- Remove the existing NLS config in favor of one that supports
extensions.
- Upstream deleted webview main.js and inlined it into the HTML so move
that code (the parent origin check) into both those HTML files
(index.html and index-no-csp.html).
- The remaining diff is from changes to the surrounding context or a
line was changed slightly by upstream (for example renamed files or
new arguments like to the remote authority resolver).
* fix: modify product.json before building
Code injects this into the client during the build process so it needs
to be updated before we build.
* fix: update inline script nonces
* Update HTML base path test
* fix: missing commit
Code overrides it with nothing.
The date is also already injected.
* fix: web extensions breaking when the commit changes
By just using the marketplace directly instead of going through the
backend. I am not sure what the point is when searching extensions
already goes directly to the marketplace anyway.
But also remove the prefix that breaks this as well because otherwise
existing installations will break.
* fix: install nfpm straight from GitHub
install.goreleaser.com appears to no longer be available.
* Add -f to curl commands
This might have made it so we got the right error rather than erroring
on the envsubst step.
This is because Node uses SIGUSR1 to enable the debug listener so even
if you just want to restart code-server you end up enabling the debug
listener as well.
Opted to leave the SIGUSR1 handler in to avoid breaking existing
workflows even though it does mean even if you only want to enable the
debug listener you will end up restarting code-server as well. We could
consider removing it after a transition phase.
* Link upstream sync section in MAINTAINING to CONTRIBUTING
* Update MAINTAINING.md table of contents
Co-authored-by: Joe Previte <jjprevite@gmail.com>
* Fix code-server version not appearing in other languages
It needs to be separate from the localize call since the language
version of that string is used and it will not include a spot for the
code-server version.
I also moved the "v" so we do not get "vUnknown".
* Add code-server version to product configuration
Before 1.64 the entire product configuration was sent to the client but
that was removed so we have to add anything we want to use on the
client, like the code-server version (used in the about dialog).
Fixes#5027.
* Refresh patches
* Change version test to look for specific version
This will catch if we are not sending the actual version to the client.
* Use --exclude to skip node_modules
Instead of copying and then deleting them. This will also catch some
node_modules directories that were missed.
* Remove per-extension dependency install
Code packages all the dependencies using webpack for each extension so
there are no dependencies to install.
* Include source maps
I also moved this to its own patch because it feels sufficiently
standalone.
Fixes#5026.
* Refresh language patch
The base is slightly different so it needed to be refreshed.
* Add missing package.json
This was caused by switching to Code's package step which does not
include the package.json.
Fixes#5019.
* Include keytar
It seems this actually is used now.
* Regenerate last opened patch
The lines were a bit off.
* Remove packaged .gitignore files
Fixes#4964.
* Remove extra Node binary
This gets overidden in the standalone but it was getting uselessly
included in the npm package.
* Move integration types into code-server
This will be easier to maintain than to have it as a patch.
* Disable connection token
Using a flag means we will not need to patch it out. I think this is
new from 1.64?
* Add product.json to build process
This way we do not have to patch it.
* Ship with remote agent package.json
Instead of the root one. This contains fewer dependencies.
* Let Code handle errors
This way we will not have to patch Code to make this work and I think it
makes sense to let Code handle the request.
If we do want to handle errors we can do it cleanly by patching their
error handler to throw instead.
* Move manifest override into code-server
This way we will not have to patch it.
* Move to patches
- Switch submodule to track upstream
- Add quilt to the process
- Add patches
The node-* ignore was ignoring one of the diffs so I removed it. This
was added when we were curling Node as node-v{version}-darwin-x64 for
the macOS build but this no longer happens (we use the Node action to
install a specific version now so we just use the system-wide Node).
* Use pre-packaged Code
* Move Code to a submodule
Closes#4901.
* Base Code cache on hash and re-enable node_modules cache
The current setup appears to only rebuild VS Code if the dependencies
change but we need to rebuild it if anything changes.
I also re-enabled the commented out node_modules caches. They look like
they should work to me with the submodule method. I think the problem
occurred because Code itself was being installed in the yarn step.
* Merge setup and navigate functions
Whenever we navigate we probably want to make sure the editor is ready
so might as well just have one function.
* Add customizable entry and workspace directory
* Add test for state db migration
* Update Code
This contains the state migrations.
* feat: github-auth flag
This will allow injecting credentials into code-server if you already
have them.
* Update Code
Contains the GitHub auth changes.
* Add e2e test for GitHub token
* Add helper for navigating the quick picker
This has problems similar to the menu except instead of closing it gets
re-created which interrupts the hover call and causes the test to fail.
Now it will keep trying just like the menu.
* Add a test for opening a file
* Add test for colliding state
* Update VS Code
This contains the colliding state fix.
* Disable BROWSER env var
Right now the browser helper script does not actually work. It seems
safer to skip this until we can fix it.
* Bump to 4.0.2
* Update changelog for 4.0.2
* Add test extension
This will let us test extension-related features (like the proxy URI).
I removed the environment variables in the script because they override
the ones you set yourself. We still set defaults in constants.ts.
* Add changelog entry for VSCODE_PROXY_URI
* Add terminal test for VSCODE_PROXY_URI
* Update VS Code
This adds the VSCODE_PROXY_URI environment variable.
I think the problem is that when a proxy is not in use proxy-agent
returns the global agent...which is itself since we set it globally,
causing the loop.
VS Code already covers proxies meaning we only need to do it in our own
requests so to fix this pass in the agent in the version fetch request
instead of overidding globally.
Also avoid proxy-from-env and pass in the proxy URI instead as both
http_proxy and https_proxy can be used for either http or https requests
but it does not allow that.
* Implement last opened functionality
Fixes https://github.com/cdr/code-server/issues/4619
* Fix test temp dirs not being cleaned up
* Mock logger everywhere
This suppresses all the error and debug output we generate which makes
it hard to actually find which test has failed. It also gives us a
standard way to test logging for the few places we do that.
* Use separate data directories for unit test instances
Exactly as we do for the e2e tests.
* Add integration tests for vscode route
* Make settings use --user-data-dir
Without this test instances step on each other feet and they also
clobber your own non-test settings.
* Make redirects consistent
They will preserve the trailing slash if there is one.
* Remove compilation check
If you do a regular non-watch build there are no compilation stats so
this bricks VS Code in CI when running the unit tests.
I am not sure how best to fix this for the case where you have a build
that has not been packaged yet so I just removed it for now and added a
message to check if VS Code is compiling when in dev mode.
* Update code-server update endpoint name
I have been wondering why my editor is not defaulting to two spaces and
it is because indent_size was removed from the * block.
I am not sure why we would need a separate block for package.json and
yaml anyway since they use the same settings as the rest of our code.
I would revert the commit but this change was made in a larger commit
with a bunch of unrelated changes.
The problem before was that the pop() caused the open in existing
instance functionality to break because the arguments no longer
contained the file.
We could simply remove the pop() but since `workspace` and `folder` are
not CLI arguments I think it makes sense to handle them in a separate
function which can be called at the point where they are needed. This
also lets us de-duplicate some logic since we create these arguments in
two spots and lets us skip this logic when we do not need it.
The pop() is still avoided because manipulating a passed-in object
in-place seems like a risky move. If we really need to do this we
should copy the positional argument array instead.
* Add tests for relativeRoot
* Remove path.posix.join
Since this is for file system paths it feels incorrect to use it on
URL paths as they are different in many ways.
* Rewrite cookie path logic
Before we relied on the client to resolve the base given to it by the
backend against the path.
Instead have the client pass that information along so we can resolve it
on the backend. This means the client has to do less work.
* Do not remove out directory before watch
This is re-used for incremental compilation.
Also remove del since that was the only use (and we can use fs.rmdir in
the future if we need something like this).
* Remove unused function resolveBase
* Remove extra VS Code CLI spawn
We already spawn VS Code's CLI when necessary in the lines below.
Having the CLI spawn unconditionally when in a VS Code environment makes
it impossible to run code-server within code-server (for example to
develop it).
* Update VS Code
This sanitizes our environment variables so code-server does not always
think it is a child spawn.
Fixes https://github.com/cdr/code-server/issues/4519.
The watch script was reloading the web server after every extension
compilation which is not necessary plus VS Code will not even be ready
at that point anyway.
Instead restart when the main compilation is finished. The string to
match with includes a "with" because otherwise it would match "Finished
compilation extensions" which is not the main compilation task where we
actually need to restart the web server.
I also replaced this.log with console.log because the former does not
include a newline and it appears we want newlines with all
these (otherwise the next log starts on the same line which looks odd).
I removed the cache clean as well because the cache is meant to stay
there to speed up builds.
The "Starting watch-client" string no longer appears in the latest build
output. We could look for "Finished compilation with" to avoid
restarting when other tasks restart (since they also include the name
i.e. "Finished compilation extensions with 0 errors") but I figure we
might as well restart code-server when any compilation task completes in
case other tasks include changes that need to be reloaded.
Asher