For an easier implementation, we change our original definition of
`initializeAttachmentMetadata`. This means we have to re-run it marked as
version 6 and mark schema version 5 as deprecated as its definition has changed.
When indexing message attachment metadata using numeric indexes such as:
```javascript
{
conversationId: '+12223334455',
received_at: 123,
attachments: […],
numAttachments: 2,
},
{
conversationId: '+12223334455',
received_at: 456,
attachments: [],
numAttachments: 0,
}
{
conversationId: '+12223334455',
received_at: 789,
attachments: [],
numAttachments: 1,
}
```
It creates an index as follows:
```
[conversationId, received_at, numAttachments]
['+12223334455', 123, 2]
['+12223334455', 456, 0]
['+12223334455', 789, 1]
```
This means a query such as…
```
lowerBound: ['+12223334455', 0, 1 ]
upperBound: ['+12223334455', Number.MAX_VALUE, Number.MAX_VALUE]
```
…will return all three original entries because they span the `received_at`
from `0` through `Number.MAX_VALUE`. One workaround is to index booleans using
`1 | undefined` where `1` is included in the index and `undefined` is not, but
that way we lose the ability to query for the `false` value. Instead, we flip
adjust the index to `[conversationId, hasAttachments, received_at]` and can
then query messages with attachments using
```
[conversationId, 1 /* hasAttachments */, 0 /* received_at */]
[conversationId, 1 /* hasAttachments */, Number.MAX_VALUE /* received_at */]
```
- messages.getQuoteObjectUrl: early return
- backup.js: explaining variables for long if statement
- types/messages.js: Log if thumbnail has neither data nor path
- sendmessage.js:
- remove extraneous logging
- fix indentation
- upload attachments and thumbnails in parallel
- preload: don't load fs for tests, just fse
- _conversation.scss: split two selectors into two lines, 0px -> 0
- backup_test.js: use fse.existsSync and comment twoSlashes regex
- network_tests_view_test.js: Comment duplicate assignment to window.getSocketStatus
- Remove extra padding at top of Android bubbles, via sibling selector
- Don't include .attachments, .quote-wrapper, .content in bubble unless
we actually need them. This allows for sibling selectors.
- This is a different technique for adding the ReactWrapperView for
quotes - it is now appended to the DOM instead of attaching to
something already in the DOM. This allows us to use .remove(), so it's
a bit cleaner.
- Users of ReactWrapperView can now specify tagName and className
But only if it doesn't have an error.
Also: reformatted message template in legacy_templates.js to match what
is in background.html for easier diffing.
1. MessageReceiver always pulls down thumbnails included in quotes
2. Message.upgradeSchema has a new schema that puts all thumbnails on
disk just like happens with full attachments.
3. handleDataMessage pipes quote from dataMessage into the final message
destined for the database
Separate linting from testing as follows:
- `yarn jscs`: Run JSCS.
- `yarn jshint`: Run JSHint.
- `yarn lint`: Run all linters, i.e. ESLint, TSLint, JSHint, and JSHint.
- `yarn test-node`: Run Mocha tests in Node.js environment.
- `yarn test-electron`: Run tests in Electron environment via Grunt.
- `yarn test`: Run all tests.
CI
- Align Travis and AppVeyor scripts as much as possible.
- Run linting before tests to fail fast.
- Run Node.js (headless and fast) tests first.
- Run Electron tests last (Travis seems to require custom setup in `travis.sh`).
Quite a bit of change here.
First, the basics:
- New dependencies were added: react, typescript, tslint, and react-styleguidist
- A new npm script: transpile. It uses typescript to process .tsx files in js/react, putting .js files next to the original file. It's part of the watch functionality of grunt dev as well as the default task run with just grunt (used to build the app prior to release). A lighter-weight to get watch behavior when just working on React components is to run yarn transpile --watch.
- yarn run clean-transpile will remove generated .js files
Style guide via react-styleguidist. Example site: https://react-styleguidist.js.org/examples/basic/
- Start with yarn styleguide
- Component.md files right next to the .tsx file
- jsdoc-style comments are picked up and added to the generated part of the styleguide - the overall summary and a table listing methods and properties of the component
- It has hot-reloading!
- It uses webpack, which means that our app now pulls in webpack though we don't use it to generate anything for the production app.
- I did a bunch of work to enable the use of Backbone views in this context, which will allow us to move smoothly from the old world to the new. First, add all the permutations in the old way, and then slowly start to re-render those same views with React.
A bit of dependency cleanup to enable use in React components:
- moment was moved from our Bower dependencies to our npm dependencies, so it can be used in React components not running in a browser window.
- i18n was moved into the new commonjs format, so it can be used in React components even if window is not available.
Lastly, a bit of Gruntfile cleanup:
- Removal of Chrome App-era modifications of background.js
- Make jshint/jscs watch more targeted, since more and more we'll be using other tools
Split out test-specific and general utility react components too.
And moved our test/legacy* files for the Style Guide into a styleguide/
subdirectory of test/.
I think we'll be able to live in this directory structure for a while.
Due to a number of hacks, the style guide can be used to show Backbone
views. This will allow a smooth path from the old way of doing things to
the new.
Class: `ConversationController`.
This function should not be used in application code as it creates potentially
invalid `Conversation` instances in our global conversation collection. We keep
making it available for testing purposes.
Using 2 hex characters [0-9a-f] will give us 16 * 16 = 256 root folders which
seems more manageable than 4096 (16^3). Assuming a user has 10,000 attachments,
they should roughly distribute at ~40 per folder with prefix length 2 rather
than ~2.5 per folder with a prefix of 3.
Using hashes, we get the benefit of deduplication but if a user receives two
messages with the same attachment, deleting one would delete it for both since
they are only stored once. To avoid the complexity of tracking number of
references, we simply generate random file names similar to iMessage on MacOS
(?) and Signal Android.
Backup creates, in a target directory:
- An attachments folder, with all attachments, each named for their
parent message's id - a GUID. If there is more than one attachment
in a given message, each attachment beyond the first will end with
'-N', zero-indexed.
- A file named messages.zip. It contains exactly what went to disk in
the original export code, but zipped up.
Export is now only 'light,' and in this new messages.zip format.
Import supports both the new format and the old format. If the target
directory has a messages.zip file, we'll treat it as the new format.
Next up: Encrypting attachments and the messages.zip!
