Developer: Hashir Sarwar
Organization: Zulip
Mentor: Hemanth V. Alluri
My GSoC project was focused on extending features, fixing high-priority bugs, and improving the overall experience of Zulip. This post serves as my final submission to the project.
Under this project there are three important areas covered:
- Push notifications and mobile features that require server-side changes
- Zulip frontend on mobile and desktop browsers
- Inline previews and markdown support
Following is the chronological detail of the work done during my GSoC period.
Previously, the Zulip server didn't support sending notification counts to iOS devices. I have worked on adding this support along with automated testing.
Related PR
Add support for setting counts in iOS push notifications #15179
This is one of the most important features that I have worked on during my GSoC period. End-to-end encryption is a vital security enhancement, especially for team chat software like Zulip.
I have used the AES-GCM algorithm to encrypt the data before sending it from the server to the mobile devices.
Related PRs
Store tokens locally even when bouncer is used #15261 (Preparatory work)
Add end-to-end encryption for push notifications #15229 (Actual work)
The size of the payload sent from the Zulip server to the browsers and mobile apps has been significantly reduced. To be more specific, the network bandwidth is reduced from 1.3MB compressed and 10.4MB total to 0.92MB compressed and 8.8MB total.
This has been made possible by not sending avatar URLs of long-term idle users from the server. This field is not compressible and thus eats up a lot of bandwidth. If we ever need the avatar URL of a long-term idle user, we'll compute it from the frontend/mobile app.
Related PRs
Avoid sending avatar URLs of long-term idle users to clients #15359
Add user_avatar_url_field_optional support to the web client #15546
Zulip on Chrome mobile browser is seriously slow; particularly the scrolling is not smooth. I have investigated, posted details, and proposed some fixes for laggy scrolling on Chrome mobile.
It turned out that Chrome is creating multiple composite layers when Zulip is being run on mobile which is causing layer invalidation and long "update layer tree" operations every time we scroll.
I have not opened any PR to fix this since it mostly involved refactoring HTML/CSS code and I have mostly worked on Python/JavaScript code in Zulip but I am sure my findings will be helpful to fix this in the future.
Many of the cases for deactivated users were left unhandled causing unexpected behaviors when a user interacts with deactivated users. Based on the discussions with other members of the community, the following cases have been worked on:
- Avoid the PM list from collapsing mysteriously when a deactivated user is clicked on.
- Allow deactivated users' pills in the compose box but distinguish them visually from other pills.
- Display a message that the user you are trying to interact with is deactivated.
The image below demonstrates the front-end changes made when a user interacts with a deactivated user.
Related PR
Fix the behaviour of PM list with deactivated user #13795
Typeahead in the navbar didn't close unless an option was selected from the typeahead itself. The code causing this bug was investigated addressed in my PR.
Related PR
Fix binding of event handler to blur event #15945
Browsers display a "mixed content" warning when initial HTML is loaded over a secure HTTPS connection, but other resources (such as images, videos, stylesheets, scripts) are loaded over an insecure HTTP connection. In this case, the preview images from unencrypted sites caused this warning. I resolved this by using the "thumbor" library to serve preview images. The use of thumbor has two main purposes:
- This is important to avoid mixed-content warnings from browsers and does have some real security benefits in protecting users from malicious content.
- Potentially unnecessary bandwidth that might be used in communication between the Zulip server and clients would be minimized. Uploading large photos could result in a bad experience for users with a slow network connection.
Related PR
Use thumbor for inline URL preview images #16037
Zulip didn't support the  markdown syntax for images. I have adjusted the Zulip's existing model to make use of this markdown syntax and tweaked inline images' CSS properties accordingly. The syntax has also been made default for sending images as attachments or pasting images from the system's clipboard ([title](url) syntax was used previously for this purpose). This would allow the users to send images without any text link. The added advantage is that this would give more control to the user for positioning of previews within a message.
Related PR
markdown: Add support for new syntax . #16117
The idea was to provide an option to the user to remove any undesired URL preview from a sent message. One of Zulip's maintainers has worked on this previously but his PR was stale and outdated. I have reworked this with the required updates.
Related PRs
Remove preview for a specific URL in a message #16231
The purpose of feature_level is to provide a way for (non-webapp) clients, like the mobile and terminal apps, to tell whether the server it's talking to is new enough to support a given API feature -- in particular a way that
- is finer-grained than release numbers, so that for features developed after e.g. 2.1.0 we can use them immediately on servers deployed from master (like chat.zulip.org and zulipchat.com) without waiting the months until a 2.2 release;
- is reliable, unlike e.g. looking at the number of commits since a release;
- doesn't lead to a growing bag of named feature flags which the server has to go on sending forever.
I had started this project before my official GSoC period and continued working on this parallel to other projects.
Related PR
Store server feature level in Redux #4079
It has been an honour to be a part of one of the leading open-source communities. My next plan is to continue working on my open PRs as per the directions of the reviewers and get them merged. Furthermore, I will keep contributing to the project to stay an active part of this amazing open-source community.
Lastly, huge thanks to Google and Zulip for providing the opportunity and to my mentor Hemanth for guiding me throughout the program.