We've simplified; HTTPS_PROXY or https_proxy is used for all requests.
We also require that only our self-signed certificates are used for
secure traffic. That rules out all SSL-terminating MITM proxies, since
we don't trust their root certificate.
Once we're sure that this system works for people, we'll improve config
on MacOS and Windows.
We pull proxy settings from environment variables:
- HTTPS_PROXY for sending, profile pulls, and attachment download/upload
- WSS_PROXY for connecting to the websocket for receiving messages
- ALL_PROXY to provide one server for both
More details on our proxy handling:
- https://github.com/Rob--W/proxy-from-env#environment-variables
- https://github.com/TooTallNate/node-proxy-agent
This is the natural way of things for Linux. My understanding is that
most proxies on MacOS are system-wide and transparent, so it's not so
urgent. But Windows will likely require further UI for configuration.
Will need to do some testing with Windows users.
* Emoji: Ensure that all clicks work by using emoji data directly
* Eliminate a number of unused emoji images from final build
* Re-add the generic sheets directory, which we partially filter
* Re-enable libtextsecure unit tests, get passing, run in CI
* Save prekeys optimistically, track confirmed, new clean behavior
* Eliminate potential conflicts when rotating on startup
* Remove last symlink: get libtextsecure tests running on windows
* Log the files discovered in logPath
I've encountered some logs which include very old entries; and my
suspicion is that we're not cleaning up old log files properly.
* Log prekey fetches (success and failure), just like signed keys
* Force log file information into the final web-ready log
* Log when we get a blocked numbers sync message
* Save three old signed keys in addition to the current active
* Remove the mystery from all the error-related log messages
* Log successful load of signed key - to help debug prekey errors
* removeSignedPreKey: Don't hang or crash in error cases
* Log on top-level unhandled promise rejection
* Remove trailing comma in param list, Electron 1.6 does not like
* Harden top-level error handler for strange object shapes
* Retry failed signed key rotation; start rotation when registered (#1772)
* rotateSignedPrekeys: Fix 'res is not defined' error
* If the server rejects key rotation, don't retry immediately
* Force a signed key rotation on launch of any new version
* web requests: Don't fail if JSON not well-formed in error cases
Turns out that before our move to node-fetch, we ignored JSON.parse()
failures: https://github.com/WhisperSystems/Signal-Desktop/pull/1552/files#diff-1103a6aff0f28b6066715c6994278767L37518
* Update to libsignal v1.2.0, delete sessions on reset session
Previously we only archived sessions when sending or receiving a
'reset secure session, which didn't match up with the mobile apps.
* Script for beta config; unique data dir, in-app env/type display
To release a beta build, increment the version and add -beta-N to the
end, then go through all the standard release activities.
The prepare-build npm script then updates key bits of the package.json
to ensure that the beta build can be installed alongside a production
build. This includes a new name ('Signal Beta') and a different location
for application data.
Note: Beta builds can be installed alongside production builds.
As part of this, a couple new bits of data are shown across the app:
- Environment (development or test, not shown if production)
- App Instance (disabled in production; used for multiple accounts)
These are shown in:
- The window title - both environment and app instance. You can tell
beta builds because the app name, preceding these data bits, is
different.
- The about window - both environment and app instance. You can tell
beta builds from the version number.
- The header added to the debug log - just environment. The version
number will tell us if it's a beta build, and app instance isn't
helpful.
* Turn on single-window mode in non-production modes
Because it's really frightening when you see 'unable to read from db'
errors in the console.
* aply.sh: More instructions for initial setup and testing
* Gruntfile: Get consistent with use of package.json datas
* Linux: manually update desktop keys, since macros not available
This fix is related about issue #1587
window.drawAttention() repositioned to allow complete control over notifications by settings. It now does not draw attention when the notifications are off.
* Add emoji button and popup panel
This integrates a simple third party emoji panel with a few css overrides to
correct some relative paths and colors.
The trickiest thing about this is ensuring we don't break the layout, which is
acheived through precise control over the panel's height, and prodigious calls
to updateMessageFieldSize.
// FREEBIE
* Don't close emoji panel on click, do close on send
To better facilitate multiple emoji entry.
// FREEBIE
* Make panel emojis bigger and higher resolution
// FREEBIE
* Move paperclip button to the right of the microphone
This makes our bottom-bar button arrangement more comfortable and consistent
with Android.
// FREEBIE
* Move emoji picker padding to inner container
* Insert emojis at cursor position
Don't just append to the end like a n00b! Also handle selected text correctly.
https://stackoverflow.com/questions/11076975
// FREEBIE
* A few visual tweaks to reduce visual complexity of emoji panel
- No gray buffer on the right side of the emoji panel
- No gray buffer between message compose text box and emoji window
- The scroll bar for the emojis is the same as our normal scrollbars
I got a 413 (Rate limit exceeded) error from the server while fetching prekeys.
The client tried to parse the response as json since we expect json from the
prekey endpoint, which threw an exception because the response was not json.
This change prevents us from treating the response as json unless it has the
Content-Type header set accordingly.
If for some reason, the client and server disagree on whether the response
should be or is json, we'll default to treating it as text.
// FREEBIE
* Online/offline: Always stop timer, don't connect if closed
* Sockets: Send our own close event faster, shutdown only on close
Seems that we were too-aggressively disconnecting from all socket
events. Also, we should be able to send our own close event a lot
faster with no ill effects.
* Catch-up libtextsecure changes
* Use node-fetch instead of xhr
* Remove XMLHttpRequest.js
// FREEBIE
* Avoid calling json() on non json responses
Previously we would catch and swallow JSON parsing errors resulting from an
empty response, though empty responses are normal from a few endpoints, like
requesting sms or voice registration codes.
Since the JSON parsing call is now handled internally by node-fetch, we have to
keep closer track of our expected response type to avoid throwing an exception.
// FREEBIE
* Wait to manually disconnect after receiving 'offline' event
We've received Linux logs indicating that we can get offline/online
blips - only 30ms between the two events. The app was never realy
offline. So this change slows the whole disconnect process down.
* Handle offline startup: register for online event, don't connect
* Remove reload options, new file/help menus, tools/log at bottom
* Further menus refactor: install handlers at template creation
* WIP: Further tune menus, add custom about window
* New About window, new help menu items, menu labels now i18n
* Default device name on registration is now computer hostname
The OS of the device makes sense for those of us testing across a lot of
different OSes. And maybe for a user with just one desktop device. But
most users with multiple desktop devices are using the same OS for both.
* About window: Only show window when content is ready
* Fix typo in app/menu.js
