Cody: Reduce scroll threshold for auto scroll #39

umpox · 2023-07-10T16:27:42Z

Description

To give a typing feel, we reduced streaming intervals between character chunks in this PR

We should also adjust this scroll threshold, as it is now more likely that a new chunk will come in while a user scrolls, and move them back to the bottom of the threshold.

Backports keep giving an error about not finding the right team, testing to see if it's a lint issue

) FIX https://linear.app/sourcegraph/issue/CODY-4529 # Fix Deep Cody Rate Limiter Reset Logic This PR fixes an issue where users were unable to use Deep Cody even after the 24-hour cooldown period had elapsed. ## Changes - Added explicit 24-hour reset check in `DeepCodyRateLimiter.isAtLimit()` - Simplified quota replenishment logic - Improved handling of initial quota state ## Technical Details Previously, the rate limiter used a continuous time-based quota replenishment system that calculated partial quotas based on elapsed time. This led to unexpected behavior where users wouldn't get a fresh quota even after not using Deep Cody for over 24 hours. The updated implementation: 1. First checks if a full day (24 hours) has passed since last usage 2. If yes, immediately resets to full quota 3. If no, calculates remaining quota using time-based replenishment ## Test plan Added new test case resets quota after 24 hours of non-use that specifically tests the 24-hour reset functionality Test verifies that: - When last use was 25 hours ago (>24h) - With an empty quota (0) - The limiter resets to full quota (50) - Confirms the storage is updated with new quota and timestamp Maintained all existing test cases to ensure no regressions Uses the same test infrastructure (mocks, beforeEach setup) ## Changelog Backport 34d5e74 from #6366 Co-authored-by: Beatrix <[email protected]>

FIX https://linear.app/sourcegraph/issue/CODY-4529 # Fix Deep Cody Rate Limiter Reset Logic This PR fixes an issue where users were unable to use Deep Cody even after the 24-hour cooldown period had elapsed. ## Changes - Added explicit 24-hour reset check in `DeepCodyRateLimiter.isAtLimit()` - Simplified quota replenishment logic - Improved handling of initial quota state ## Technical Details Previously, the rate limiter used a continuous time-based quota replenishment system that calculated partial quotas based on elapsed time. This led to unexpected behavior where users wouldn't get a fresh quota even after not using Deep Cody for over 24 hours. The updated implementation: 1. First checks if a full day (24 hours) has passed since last usage 2. If yes, immediately resets to full quota 3. If no, calculates remaining quota using time-based replenishment ## Test plan Added new test case resets quota after 24 hours of non-use that specifically tests the 24-hour reset functionality Test verifies that: - When last use was 25 hours ago (>24h) - With an empty quota (0) - The limiter resets to full quota (50) - Confirms the storage is updated with new quota and timestamp Maintained all existing test cases to ensure no regressions Uses the same test infrastructure (mocks, beforeEach setup) ## Changelog Backport 34d5e74 from #6366 Co-authored-by: Beatrix <[email protected]>

…n issue (#6378) Follow up on #6366 This PR fixes several issues with the Deep Cody rate limiter implementation: Root Cause Fix: Rate Limit State Synchronization - Fixed race condition in LocalStorageProvider.ts where concurrent Promise.all() usage led to inconsistent state between quota and timestamp storage - Changed to sequential storage operations to ensure timestamp is always saved after quota update - This resolves the core issue where users were locked out due to incorrect time calculations (CODY-4529) ## Test plan - Added precise tests for time-based quota calculations - Enhanced test coverage for various timing scenarios Before this change, the Deep Cody rate limit error would always return the "Retry time" with current time + 24 hours to indicate the current time is being used as the last used time. Submitted at 6:44 ![image](https://github.com/user-attachments/assets/d9b50ff8-bc09-473b-a387-497d1ccbe0c2) Submitted again at 6:45 ![image](https://github.com/user-attachments/assets/e1b0266b-cfbc-40f8-807b-a9d4c93be682) After this change, the "Retry Time" should stick with the last used time until it's been reset after 24 hours. ## Changelog Backport a7339ef from #6377 Co-authored-by: Beatrix <[email protected]>

…ue (#6379) Follow up on #6366 This PR fixes several issues with the Deep Cody rate limiter implementation: Root Cause Fix: Rate Limit State Synchronization - Fixed race condition in LocalStorageProvider.ts where concurrent Promise.all() usage led to inconsistent state between quota and timestamp storage - Changed to sequential storage operations to ensure timestamp is always saved after quota update - This resolves the core issue where users were locked out due to incorrect time calculations (CODY-4529) ## Test plan - Added precise tests for time-based quota calculations - Enhanced test coverage for various timing scenarios Before this change, the Deep Cody rate limit error would always return the "Retry time" with current time + 24 hours to indicate the current time is being used as the last used time. Submitted at 6:44 ![image](https://github.com/user-attachments/assets/d9b50ff8-bc09-473b-a387-497d1ccbe0c2) Submitted again at 6:45 ![image](https://github.com/user-attachments/assets/e1b0266b-cfbc-40f8-807b-a9d4c93be682) After this change, the "Retry Time" should stick with the last used time until it's been reset after 24 hours. ## Changelog Backport a7339ef from #6377 Co-authored-by: Beatrix <[email protected]>

CLOSE https://linear.app/sourcegraph/issue/CODY-4533/layout-bug-in-prompts-tab-and-new-user-menu-ui ![image](https://github.com/user-attachments/assets/a7d5fe5c-2635-47f1-8ffa-00c781287f62) Fix UserMenu dropdown appearing underneath PromptList search box The UserMenu dropdown and PromptList search box were competing for z-index positioning since both had z-index: 1. Because PromptList was rendered after UserMenu in the DOM order, it would appear on top of the UserMenu dropdown when opened. This fix increases UserMenu's z-index to 2, ensuring it properly stacks above the PromptList search box while maintaining both components' ability to overlay their respective underlying content. Changes - Updated UserMenu dropdown z-index from 1 to 2 to create proper stacking hierarchy - Maintains existing stacking contexts for both components ## Test plan After ![image](https://github.com/user-attachments/assets/9341c4cf-b90c-4e28-806e-5c6ce15d69ab) 1. Open Cody sidebar 2. Navigate to Prompts tab 3. Click on user avatar to open UserMenu dropdown 4. Verify UserMenu dropdown appears above the PromptList search box 5. Verify PromptList search box still properly overlays its content when scrolling 6. Verify UserMenu dropdown properly overlays all other UI elements ## Changelog Backport 03dbec6 from #6380 Co-authored-by: Beatrix <[email protected]>

…#6393) CLOSE https://linear.app/sourcegraph/issue/CODY-4533/layout-bug-in-prompts-tab-and-new-user-menu-ui ![image](https://github.com/user-attachments/assets/a7d5fe5c-2635-47f1-8ffa-00c781287f62) Fix UserMenu dropdown appearing underneath PromptList search box The UserMenu dropdown and PromptList search box were competing for z-index positioning since both had z-index: 1. Because PromptList was rendered after UserMenu in the DOM order, it would appear on top of the UserMenu dropdown when opened. This fix increases UserMenu's z-index to 2, ensuring it properly stacks above the PromptList search box while maintaining both components' ability to overlay their respective underlying content. Changes - Updated UserMenu dropdown z-index from 1 to 2 to create proper stacking hierarchy - Maintains existing stacking contexts for both components ## Test plan After ![image](https://github.com/user-attachments/assets/9341c4cf-b90c-4e28-806e-5c6ce15d69ab) 1. Open Cody sidebar 2. Navigate to Prompts tab 3. Click on user avatar to open UserMenu dropdown 4. Verify UserMenu dropdown appears above the PromptList search box 5. Verify PromptList search box still properly overlays its content when scrolling 6. Verify UserMenu dropdown properly overlays all other UI elements ## Changelog Backport 03dbec6 from #6380 Co-authored-by: Beatrix <[email protected]>

…query (#6440) When we remove the mentions, for some reason we're trimming spaces from the start, but not the end. This means our exact match heuristics won't work as well for these situations because there's a rogue space at the end of the query. This just fixes the logic to trim spaces from both the start and end ## Test plan This query used to give less relevant results because we kept the trailing space: ![CleanShot 2024-12-20 at 10 53 31@2x](https://github.com/user-attachments/assets/f49246ee-99cf-47b2-835a-34505ab7ac5d) Backport 78da6bb from #6432 Co-authored-by: Camden Cheek <[email protected]>

…#6446) When we remove the mentions, for some reason we're trimming spaces from the start, but not the end. This means our exact match heuristics won't work as well for these situations because there's a rogue space at the end of the query. This just fixes the logic to trim spaces from both the start and end ## Test plan This query used to give less relevant results because we kept the trailing space: ![CleanShot 2024-12-20 at 10 53 31@2x](https://github.com/user-attachments/assets/f49246ee-99cf-47b2-835a-34505ab7ac5d) Backport 78da6bb from #6432 Co-authored-by: Camden Cheek <[email protected]>

…text settings (#6619) CLOSE https://linear.app/sourcegraph/issue/CODY-4632 Updates the `ToolboxButton` component that provides a UI for managing agent context settings with a more detailed description of the feature. - Updates toggles for agentic context and terminal access, with appropriate warnings - Displays the button on all messages - Debounces the settings update to avoid excessive API calls with the removed save button - Display button on all messages - Display button without title text in non-first message - Hide terminal context on unsupported instance/clients ## Test plan UI update without feature change - storybook updated. ![image](https://github.com/user-attachments/assets/0459f2b1-e0b9-463f-83c8-0d20e6f789b4) ![image](https://github.com/user-attachments/assets/37ee046a-0b6c-4d1c-916f-fd593d636c5a) Hide shell command on unsupported client or instances ![image](https://github.com/user-attachments/assets/24cc5215-7118-4cfa-a5de-0b643cdf2a1f) ### Before The current one doesn't explain what agentic context and terminal context are: ![image](https://github.com/user-attachments/assets/27da9d5f-bf5e-4a34-8ed8-6e37a80dbd48) Backport d865d65 from #6596 Co-authored-by: Beatrix <[email protected]>

…6618) CLOSE https://linear.app/sourcegraph/issue/CODY-4671 Fixes a regression where the agent would mark a context as needed but got filtered out during the validation step - Simplify the context review logic by directly extracting context names from the response and fetching the full content for the requested files. - Remove unnecessary callbacks and optimize the context fetching process. - Add a helper function to check if the response is ready to answer. ## Test plan Manual testing: 1. Ask in new chat: "how does agentic context work in @_cody_repo" 2. Add a follow up: "how many loop does deep cody run for the review step?" Before Deep Cody requested a file: ![image](https://github.com/user-attachments/assets/a15b9e8e-3d2f-4eaa-b143-db9384540f28) But it wasn't included in the final selected context: <img width="829" alt="image" src="https://github.com/user-attachments/assets/07f0d283-0e84-4879-81e8-f6f38c807ccb" /> After: <img width="839" alt="image" src="https://github.com/user-attachments/assets/bb846639-7a04-4ad8-a247-982dd9ede381" /> Backport 54982b5 from #6602 Co-authored-by: Beatrix <[email protected]>

… settings (#6626) CLOSE https://linear.app/sourcegraph/issue/CODY-4632 Updates the `ToolboxButton` component that provides a UI for managing agent context settings with a more detailed description of the feature. - Updates toggles for agentic context and terminal access, with appropriate warnings - Displays the button on all messages - Debounces the settings update to avoid excessive API calls with the removed save button - Display button on all messages - Display button without title text in non-first message - Hide terminal context on unsupported instance/clients ## Test plan UI update without feature change - storybook updated. ![image](https://github.com/user-attachments/assets/0459f2b1-e0b9-463f-83c8-0d20e6f789b4) ![image](https://github.com/user-attachments/assets/37ee046a-0b6c-4d1c-916f-fd593d636c5a) Hide shell command on unsupported client or instances ![image](https://github.com/user-attachments/assets/24cc5215-7118-4cfa-a5de-0b643cdf2a1f) ### Before The current one doesn't explain what agentic context and terminal context are: ![image](https://github.com/user-attachments/assets/27da9d5f-bf5e-4a34-8ed8-6e37a80dbd48) Backport d865d65 from #6596 Co-authored-by: Beatrix <[email protected]>

CLOSE https://linear.app/sourcegraph/issue/CODY-4671 Fixes a regression where the agent would mark a context as needed but got filtered out during the validation step - Simplify the context review logic by directly extracting context names from the response and fetching the full content for the requested files. - Remove unnecessary callbacks and optimize the context fetching process. - Add a helper function to check if the response is ready to answer. ## Test plan Manual testing: 1. Ask in new chat: "how does agentic context work in @_cody_repo" 2. Add a follow up: "how many loop does deep cody run for the review step?" Before Deep Cody requested a file: ![image](https://github.com/user-attachments/assets/a15b9e8e-3d2f-4eaa-b143-db9384540f28) But it wasn't included in the final selected context: <img width="829" alt="image" src="https://github.com/user-attachments/assets/07f0d283-0e84-4879-81e8-f6f38c807ccb" /> After: <img width="839" alt="image" src="https://github.com/user-attachments/assets/bb846639-7a04-4ad8-a247-982dd9ede381" /> Backport 54982b5 from #6602 Co-authored-by: Beatrix <[email protected]>

…d of changelog entries (#6712) This works by using the title as the changelog entry instead. We cannot use the compact mode version with this feature enabled, it fails to find a `section` `formatEntry` command and exits before the changelog is generated. Also updated the changelog marker since it's different now? ## Test plan I ran the [action](https://github.com/sourcegraph/cody/actions/runs/12799005063/job/35684227555) which produced this [PR](#6668) Backport 4ce3794 from #6669 Co-authored-by: Anish Lakhwara <[email protected]>

…ime error (#6729) - By testing locally, I reproduced the runtime error we see in our analytics data. It's coming from the diagnostics retriever where line numbers are not clamped. - Closes [CODY-4717: Fix the `illegal line` runtime error](https://linear.app/sourcegraph/issue/CODY-4717/fix-the-illegal-line-runtime-error) ## Test plan CI + updated unit tests Backport 962666c from #6727 Co-authored-by: Valery Bugakov <[email protected]>

…s to keep workflow referen… (#6739) running into errors in generate changelog action where it can't create the PRs due to conflicts, this removes the non essential git configuration steps to keep the reference branch the same ## Test plan N/A, couldn't test locally need the github UI Backport bfa18e6 from #6737 Co-authored-by: Kalan <[email protected]>

@mmanela

…aces vscode sign-in flow) (#6742) Part of https://linear.app/sourcegraph/issue/DES-235/[onboarding-polish]-extensions-logout-guidance-step This small change allows us to specify the instance URL for sigh up / auth flow, it's useful for workspace flow when we don't originally start our log-in flow from vscode and fill in the workspace URL but right from the workspace instance and we should notify cody to which URL it should try to log in with access token (code URL param) This PR is primarily needed for the workspace log-in into VSCode Cody PR in the Sourcegraph repository. https://github.com/user-attachments/assets/1dd1b90d-970e-4c2d-a71a-89cdaddf354a @mmanela I saw your contributions about instance param in this flow, do you think we should make users go through this flow manually? I still left your logic in there so we can switch it later if we want. @kalanchan, this PR is important for workspace setup flow. Can I ask you to make sure it's included in the upcoming Cody release? I think it's going to be 1.62, right? ## Test plan - Check that the standard auth flow has no regression - The workspace flow will be tested in the Sourcegraph repository PR Backport 7e4dbf9 from #6730 Co-authored-by: Vova Kulikov <[email protected]>

… when enabled (#6758) FIX https://linear.app/sourcegraph/issue/CODY-4737 The issue was that the Tool Cody model was being added to the list of primary models regardless of whether the feature was enabled or not. The root cause was that the check for the existence of the Tool Cody model was not properly scoped to the `isToolCodyEnabled` flag but was checking the observable instead, which would always returns true as the observable is defined. This change fixes the issue by only adding the Tool Cody model to the list of primary models if the `isToolCodyEnabled` flag is true and the Tool Cody model is not already present in the list of primary models. Included some minor clean up. ## Test plan Verify Tool Cody is not showing up in your model dropdown if you don't have the configuration in your settings: <img width="603" alt="image" src="https://github.com/user-attachments/assets/6ebdbb6e-650d-4bc4-ad62-40a285a13f5f" /> Backport ae351c2 from #6753 Co-authored-by: Beatrix <[email protected]>

…n enabled (#6759) FIX https://linear.app/sourcegraph/issue/CODY-4737 The issue was that the Tool Cody model was being added to the list of primary models regardless of whether the feature was enabled or not. The root cause was that the check for the existence of the Tool Cody model was not properly scoped to the `isToolCodyEnabled` flag but was checking the observable instead, which would always returns true as the observable is defined. This change fixes the issue by only adding the Tool Cody model to the list of primary models if the `isToolCodyEnabled` flag is true and the Tool Cody model is not already present in the list of primary models. Included some minor clean up. ## Test plan Verify Tool Cody is not showing up in your model dropdown if you don't have the configuration in your settings: <img width="603" alt="image" src="https://github.com/user-attachments/assets/6ebdbb6e-650d-4bc4-ad62-40a285a13f5f" /> Backport ae351c2 from #6753 Co-authored-by: Beatrix <[email protected]>

@mmanela

… for quick sign-in (#6653) (#6818) This is cherry-picking @mmanela commit from #6653 It was recently reverted here #6794 due to problems with reverting other PRs and merge conflicts. This PR contains a required change for a successful Workspace Auth flow so we should include this in the upcoming 1.64 release ## Original PR description VS Code side part of https://linear.app/sourcegraph/issue/SRC-908/automate-cody-authflow-vs-code After a user creates a workspace we want to make it easy for them to sign-in to the instance in their editor. For enterprise instances today, they need to copy the instance url, go into VS Code and then choose to signin with enterprise instance and then paste the url. This change will help stream-line that flow by allowing the UI in sourcegraph to link to VS Code with a url like `vscode://sourcegraph.cody-ai?instance=someinstance.sourcegraphdev.app` VS Code will handle and either switch to that account if you already are auth'd OR pop-open the signin menu directly ![image](https://github.com/user-attachments/assets/cdab4135-2dae-48d3-945c-2e70187c37f0) Note: For initial version I limited just to workspace domains. But we can generalize that to allow other instance domains. ## Test plan 1. Validate when client doesn't have instance already 2. Validate when client already has instance connected (cherry-picked from commit 9719dc4) Backport ef65034 from #6812 Co-authored-by: Vova Kulikov <[email protected]>

@mmanela

… URL for quick sign-in (#6653) (#6817) This is cherry-picking @mmanela commit from #6653 It was recently reverted here #6794 due to problems with reverting other PRs and merge conflicts. This PR contains a required change for a successful Workspace Auth flow so we should include this in the upcoming 1.64 release ## Original PR description VS Code side part of https://linear.app/sourcegraph/issue/SRC-908/automate-cody-authflow-vs-code After a user creates a workspace we want to make it easy for them to sign-in to the instance in their editor. For enterprise instances today, they need to copy the instance url, go into VS Code and then choose to signin with enterprise instance and then paste the url. This change will help stream-line that flow by allowing the UI in sourcegraph to link to VS Code with a url like `vscode://sourcegraph.cody-ai?instance=someinstance.sourcegraphdev.app` VS Code will handle and either switch to that account if you already are auth'd OR pop-open the signin menu directly ![image](https://github.com/user-attachments/assets/cdab4135-2dae-48d3-945c-2e70187c37f0) Note: For initial version I limited just to workspace domains. But we can generalize that to allow other instance domains. ## Test plan 1. Validate when client doesn't have instance already 2. Validate when client already has instance connected (cherry-picked from commit 9719dc4) Backport ef65034 from #6812 Co-authored-by: Vova Kulikov <[email protected]>

…tic model (#6828) The display name for the deep-cody model has been updated from 'Agentic Chat' to 'Agentic chat'. ## Test plan Verify the name update: <img width="496" alt="image" src="https://github.com/user-attachments/assets/8e1cc5a3-d8bc-44ea-94c9-1a1488e39541" /> Backport ca3fcd1 from #6827 Co-authored-by: Beatrix <[email protected]>

…model (#6829) The display name for the deep-cody model has been updated from 'Agentic Chat' to 'Agentic chat'. ## Test plan Verify the name update: <img width="496" alt="image" src="https://github.com/user-attachments/assets/8e1cc5a3-d8bc-44ea-94c9-1a1488e39541" /> Backport ca3fcd1 from #6827 Co-authored-by: Beatrix <[email protected]>

… some clean up (#6883) - Name all of the release workflows consistently so they will appear together in GitHub's actions list. - Remove the unused experimental changelog generator. - Remove stale release instructions from CONTRIBUTING.md, instead point to Notion. ## Test plan I am release captain, I will dogfood this week. Backport 495962e from #6879 Co-authored-by: Dominic Cooney <[email protected]>

…ease.sh about release branches (#6884) These scripts predate release branches and used to always look at global tags. This change makes it work on release branches. It uses git to list the commits on the release branch, then does the same version bump calculation using the branch's tags. It takes care to handle newly budding branches which are just a commit on `main` that has no backports yet. In that case it looks at all of the available tags. It will work carrying a branch from prerelease through to release: ``` # At the start of the branch $ ./scripts/push-git-tag-for-next-release.sh --minor --nightly # During prerelease, assume we have accumulated some commits on the branch now: $ ./scripts/push-git-tag-for-next-release.sh --patch --nightly # Transitioning to release: $ ./scripts/push-git-tag-for-next-release.sh --minor # Hopefully never needed, but making emergency patches on stable: $ ./scripts/push-git-tag-for-next-release.sh --patch ``` There are some caveats with this tool: - When working on a release branch, it only has *local* view of *that branch's tags.* Misusing it should be harmless because creating a tag which conflicts with an existing tag will fail when the tag is pushed. - It does not enforce the new practice of stable builds having even version numbers and prerelease builds having odd version numbers. If you ever need to work around this tool, just author and push git tags yourself. To this point release captains have been doing that anyway. Adds a `--dry-run` option which makes testing easier. Removes outdated docs and points to Notion instead. ## Test plan ``` # Pretend you're just cooking up a release branch $ git remote update origin $ git checkout -b my-release-branch origin/main $ cd jetbrains $ ./scripts/push-git-tag-for-next-release.sh --minor --nightly --dry-run # It should suggest a minor version bump based on all the version tags in the repo ... # Pretend you're on a release branch, we're doing a patch release $ git fetch origin refs/heads/jb-v7.8.x $ git checkout FETCH_HEAD $ git checkout <this PR> -- scripts/next-release.sh scripts/push-git-tag-for-next-release.sh $ git commit -m 'test test test do not push' # script insists the repo is clean $ ./scripts/push-git-tag-for-next-release.sh --minor --dry-run # It should suggest v7.8.1 ``` I am the release captain and will dogfood this script in the current release. Backport b912145 from #6881 Co-authored-by: Dominic Cooney <[email protected]>

…s. (#6977) We will create releases in sourcegraph/cody now, because that's where the commits are! And link to this page to display them: https://github.com/sourcegraph/cody/releases?q=JetBrains&expanded=true ## Test plan ``` cd jetbrains ./gradlew verifyPlugin ``` Land this, backport it, then release a v7.67 nightly and check that [the JetBrains Marketplace page](https://plugins.jetbrains.com/plugin/9682-cody-ai-code-assistant/edit/versions/nightly) looks good. Backport 15147f6 from #6976 Co-authored-by: Dominic Cooney <[email protected]>

… follow-up message (#7009) The user knows who they are, why show it next to every message? It clutters the UI. So I removed the avatar from the human messages (we still have the avatar in the top right). I also removed the name from the follow-up message. It doesn't add anything. Ideally we'd remove the "user name" every where but when I did that, the layout looked kinda bad because there was still the 'split into chat' button. ## Before ![before](https://github.com/user-attachments/assets/2ce844e1-275a-42e4-876c-5d18bb011a47) ## After ![after](https://github.com/user-attachments/assets/364a23a7-242b-4ed7-a49f-08de8df6b47f) ## Test plan - N/A Backport 127d52c from #6981 Co-authored-by: Thorsten Ball <[email protected]>

…S Code (#7006) Previously, pressing Enter in a chat message in VS Code's Cody chat would cause the iframe to be shifted up by ~5px. This was an annoying visual jitter. Now, this no longer happens. ## Test plan CI Also: Run in VS Code, type a chat message, press Enter, and ensure there is no jitter. Also ask a long chat question and ensure that the scroll-to-end works. ## Changelog - Fixed an issue where pressing Enter in chat would cause brief visual jitter in the UI. Backport bc8f1fb from #6991 Co-authored-by: Quinn Slack <[email protected]>

…constant (#7019) This bugged me while reading and to warm up to the codebase, I decided to replace the hardcoded strings with the reference. Can't do it in `lib/shared`, but in `vscode`. ## Test plan - Run Cody and see that agentic chat works. Backport 386c0b6 from #6958 Co-authored-by: Thorsten Ball <[email protected]>

…ing images (#7055) ## Description closes https://linear.app/sourcegraph/issue/CODY-4869/auto-edit-support-custom-font-size-and-line-height Use the users' default font size and determine a suitable line height from this when we are generating image suggestions ## Test plan Manually testing when creating auto edits Tested on MacOS and Linux Backport 5d7d4e5 from #7036 Co-authored-by: Tom Ross <[email protected]>

Upgrades changelog generator from my old script to dev infra's. I added a new `betweenCommit` strategy to their tooling, which let's us parse total git history between 2 commits. see sourcegraph/devx-service#299 for full details. Using the dev infra changelog binary will now let us group PRs by their titles and types ($type/$domain) and will make the changelog process much easier. We should start following the conventions laid out on the [Cody PR conventions page](https://www.notion.so/sourcegraph/PR-and-Changelog-conventions-ecfd817301614c399d4c8b9aa5832370) Release captains can now run this workflow from the milestone branch to generate the changelog. See #7077 as an example of what this outputs. // TODO for after Force push the changelog PR branch so that subsequent workflow runs won't fail trying to create the same branch ## Test plan tested the new workflow on my branch kalan/changelog-M68, worked well for upcoming milestone 68 release Backport 5c5b448 from #7081 Co-authored-by: Kalan <[email protected]>

…logging payload (#7198) Currently `insert` events are missing in the smart apply logging because they return early, the PR adds them in the current logging pipeline ## Test plan - Open a empty file and ask the chat to make some changes - Click on the smart apply event - Observe the telemetry in the `Cody by Sourcegraph` output channel ``` █ telemetry-v2 recordEvent: cody.smart-apply.context/applied: { // ... metadata .... "privateMetadata": { "smartApplyContext": { "smartApplyModel": "anthropic::2024-10-22::claude-3-5-haiku-latest", "userQuery": "Similar to fireworks.ts add the anthropic provide in the file anthropic.ts ", "replacementCodeBlock": "import from '../ [...other chars] }\n}", "filePath": "src/autoedits/adapters/anthropic.ts", "fileContent": "", "selectionType": "insert", "selectionRange": [ 0, 0 ], "applyTaskId": "57b470f50a63a8af761abb9c5ed73c4806c78d473e5526067ea92f4d1fc22a4e" } }, }, } ``` Backport 96bc4a4 from #7197 Co-authored-by: Hitesh Sagtani <[email protected]>

…ssage (#7210) RE: https://linear.app/sourcegraph/issue/CODY-4785 This PR adds a simple hack to support rendering <think> tags in the ChatMessageContent component. The <think> content is displayed in a collapsible details element, allowing users to view the AI's internal thought process. The MarkdownFromCody component is also updated to allow the <think> element. ## Test plan Ask Cody "how many files are there in the codebase? Enclose your thoughts inside <think> tags before answering" - Verify that <think> tags are properly extracted and displayed in the ChatMessageContent component - Ensure that the collapsible details element functions as expected, allowing users to view the think content - Confirm that the MarkdownFromCody component correctly renders the <think> element Example: https://github.com/user-attachments/assets/0a5cff8f-1b08-48e4-9cf9-4fd6d13ef05a Backport 710ac73 from #6845 Co-authored-by: Beatrix <[email protected]>

…ssage (#7209) RE: https://linear.app/sourcegraph/issue/CODY-4785 This PR adds a simple hack to support rendering <think> tags in the ChatMessageContent component. The <think> content is displayed in a collapsible details element, allowing users to view the AI's internal thought process. The MarkdownFromCody component is also updated to allow the <think> element. ## Test plan Ask Cody "how many files are there in the codebase? Enclose your thoughts inside <think> tags before answering" - Verify that <think> tags are properly extracted and displayed in the ChatMessageContent component - Ensure that the collapsible details element functions as expected, allowing users to view the think content - Confirm that the MarkdownFromCody component correctly renders the <think> element Example: https://github.com/user-attachments/assets/0a5cff8f-1b08-48e4-9cf9-4fd6d13ef05a Backport 710ac73 from #6845 Co-authored-by: Beatrix <[email protected]>

…or (#7239) This PR addresses the issue where models with reasoning capabilities were causing edits to fail when selected in the Inline Edit model selector. Changes Made - Added logic to filter out the ModelUsage.Edit capability from any model that has the reasoning capability Introduced new model capabilities reasoning and tools to properly identify models with these features Added corresponding model tags Reasoning and Tools to the tag enum Updated code to maintain the model capabilities but selectively remove edit usage when reasoning is detected Added entry to the changelog to document this fix - Implementation Details - The core fix is implemented by checking if a model's capabilities include 'reasoning' and, if so, removing the Edit usage from that model's available usages. This prevents reasoning-enabled models from appearing in the Inline Edit model selector while maintaining their availability for chat and other supported features. ## Test Plan The change has been tested to ensure that: - Reasoning models no longer appear in the Inline Edit model selector - Reasoning models remain available for chat and other supported features - Non-reasoning models with edit capability still work as expected Before - connect to dotcom and open the inline edit model selector, 3.7 sonnet is listed in the dropdown ![image](https://github.com/user-attachments/assets/585dc922-5eb1-4009-8e03-28826013c4f4) After - 3.7 removed ![image](https://github.com/user-attachments/assets/df30e861-08a1-4b1d-a9b3-545635f280a4) Backport a7ec5fb from #7238 Co-authored-by: Beatrix <[email protected]>

This PR improves VS Code release notes generation by introducing changelog summarization. The new `release-notes.ts` script creates easy to understand, categorized release notes with emojis and clear formatting. This refactoring also makes it more simple to plugin future changelogs. The following output is an example of what will be used in the release script. Example output for `v1.70.1`: ``` ✨ For the full technical changelog, see [What’s new in v1.70.0](https://github.com/sourcegraph/cody/blob/main/vscode/CHANGELOG.md) since v1.68.0 ✨ This release brings significant improvements to the chat interface, enhanced auto-edit functionality, and various bug fixes to improve the overall user experience. ✨ New Features - Support for dynamic mentions in prompt templates, making @ mentions more flexible and context-aware [#6793](#6793) - Enhanced image handling in auto-edit with better alignment and user-specific font size support [#7036](#7036) 🔄 Improvements - Streamlined chat interface with simplified UI and improved user experience [#6992](#6992) - Enhanced prompt caching for code context messages, improving response times [#6953](#6953) - Improved code border styles for better visual clarity [#7024](#7024) 🐛 Bug Fixes - Fixed 500ms delay when submitting chat messages [#6991](#6991) - Resolved issues with @ mention menu appearing outside the editor window [#7045](#7045) - Fixed low resolution images on low DPI screens [#7112](#7112) - Corrected language filters in rules not applying properly [#7025](#7025) 🎨 UI Enhancements - Improved repository filter dropdown functionality [#6994](#6994) - Enhanced history search interface with fixed border styling [#6983](#6983) - Simplified chat UI by removing user avatars and streamlining message presentation [#6981](#6981) **Full Comparison**: M68...M70 ``` Example output for `v1.70.1`: ``` ✨ For the full technical changelog, see [What’s new in v1.70.1](https://github.com/sourcegraph/cody/blob/main/vscode/CHANGELOG.md) since v1.70.0 ✨ This minor update addresses a specific issue with Cody's agentic chat functionality. 🐛 Bug Fixes - Fixed an issue where terminal and openCtx tools were not properly registered in agentic chat, ensuring better functionality and tool availability [#7132](#7132) **Full Comparison**: M68...M70 ``` ## Test plan tested locally using `npx ts-node ./scripts/release-notes.ts` Backport 2e3b785 from #7150 Co-authored-by: Kalan <[email protected]>

…mpletion menu (#7406) https://github.com/user-attachments/assets/5d105794-516b-44dd-b81f-b0c991c835a1 ## Description Fixes an issue where we attempt to suggest an auto-edit that could conflict with the completion menu in a bad way. E.g. it may delete or add code that doesn't reflect the selected item in the menu. If the user was to accept this item with Tab, the edit would be accepted including the unwanted code. ## Test plan - Unit tests - Manual testing with completion items Backport 2c430e9 from #7389 Co-authored-by: Tom Ross <[email protected]>

Fixes #7446 (comment) Potential fix for the issue with multiple agents spawned in parallel. In our current restart code we do not sufficiently guard us against parallel cody restarts. In what situations it can happen? For example if agent is starting doing multiple independent actions (e.g. syncing some files, updating chat state, etc) and it will crash, each of those action can attempt to restart a server. That should be limited to actions explicitly triggered by user, but I also fixed one place where it was not (and which could have been source of the problem). Regardless, we should also safeguard it on the level of the `restart` function. ## Test plan 1. Run 'Cody: Restart Cody' a dozen of times as quickly as you can (you can assign it shortcut to make it easier) 2. Using process monitor make sure number of the agent processes is not increasing 3. Make sure cody is functioning properly after restart is done Backport 137fe67 from #7466 Co-authored-by: Piotr Kukiełka <[email protected]>

#7257)" (#7475) This reverts commit a31acd0. I'm reverting this as we get few reports of multiple agent processes being spawned, and this is only recent change in this area of code. There is a chance that my fix is solving the underlaying issue: #7466 But if my fix is proper, then we do no need `WebUIService.getInstance(project).reset()` in both `start` and `stop` methods. I believe before my PR it was sometimes beneficial because there was race between `stop` and `start` code. But there is also a chance that it was breaking something is timing was slightly different. ## Test plan N/A Backport 27990ca from #7469 Co-authored-by: Piotr Kukiełka <[email protected]>

…s default model (#7473) This PR fixes a critical issue where reasoning models could be incorrectly set as the default model for Edit operations, causing the application to break for users. This commit addresses an issue where reasoning models could be incorrectly set as the default model for Edit operations. The problem occurred because when Edit model is not available, Edit would fallback to the `firstModelUserCanUse` logic which doesn't filter out reasoning models for Edit usage. This could cause Inline Edit to use reasoning model even when they are not assigned for Edit usage. ### The Problem The issue occurred in the `getDefaultModel` and `getDefaultEditModel` methods in `modelsService.ts`. When a user didn't have a selected model preference for Edit and the server didn't provide a default Edit model, the code would use the first available model that the user could access. However, this logic didn't properly filter out reasoning models, which are not suitable for Edit operations. This resulted in scenarios where reasoning models were being used for Edit operations, causing unexpected behavior or application failures. ### The Solution The fix includes: 1. Enhanced filtering in `getDefaultModel` to explicitly exclude reasoning models when finding the first available model for Edit usage 2. Improved fallback logic in `getDefaultEditModel` to: - First try to use a valid Edit model (non-reasoning) - If that's not available, fall back to a valid Chat model (non-reasoning) - If neither is available, return undefined 3. Added comprehensive test coverage to ensure: - Reasoning models are never used for Edit operations - Models with problematic tags (Waitlist, OnWaitlist, Deprecated) are not used for Edit - The fallback mechanism to Chat models works correctly - User preferences are respected when valid - Invalid user preferences (like reasoning models) are ignored ## Changes - Added filtering to exclude reasoning models from being used for Edit operations - Improved fallback logic to use valid Chat models when no suitable Edit models are available - Added comprehensive test coverage to ensure reasoning models and other problematic model types (waitlist, deprecated) are never used for Edit - Updated existing tests to reflect the new behavior - Filtering out deprecated models from the available models list. - Preventing the selection of reasoning models for Edit usage. - Prioritizing chat models as a fallback when no valid edit model is available. - Ensuring that user preferences for edit models are respected unless the selected model is a reasoning model. - Handling the scenario where only reasoning models are available. ## Test plan Added a new test suite `Default model selection for Edit and Chat` with 8 test cases covering all the scenarios mentioned above. Updated existing tests to reflect the new behavior. Backport e927763 from #7461 Co-authored-by: Beatrix <[email protected]>

…sCell (#7516) Close https://linear.app/sourcegraph/issue/CODY-5444/bug-search-results-and-file-tool-results-are-not-clickable The Issue: In the original implementation, if the vscodeAPI prop was undefined during initial component render (because the VSCode API wasn't fully initialized yet), the onFileLinkClicked function would be created with a reference to that undefined value. Even if the API became available later, the callback would still reference the initial undefined value due to how closures work. The Fix: By calling getVSCodeAPI() at the time the link is clicked (rather than capturing the value during render), the code now gets the most current API reference at the exact moment it's needed. This guarantees that even if the API wasn't available during the component's first render, the link functionality will still work when clicked as long as the API is available by then. ## Test plan Clicking on agent search result should open the file link in editor for you <img width="593" alt="image" src="https://github.com/user-attachments/assets/468e39d3-0a9a-4050-bd2b-a8c9bf7285f8" /> Backport 5b8c338 from #7496 Co-authored-by: Beatrix <[email protected]>

…ents (#7586) This commit addresses an issue where completion requests might not have associated events that is causing unexpected error that stops chat requests. These changes ensure that the completion service gracefully handles scenarios where no usage data is available for a completion request, preventing potential errors and improving the robustness of the system. ### Root Cause The error occurs in the sendEvents method of the SourcegraphCompletionsClient class when it tries to process events but there are no events to process. ### Summary Implemented Fixes: 1. In nodeClient.ts: - Added a check to ensure we only call sendEvents when there are actually events to process - Added better error logging when no events are detected - Improved the error handling in the request.on('close') handler to provide more detailed diagnostics 2. In client.ts: - Added a guard clause at the beginning of sendEvents to handle empty event arrays gracefully - Added a warning log instead of throwing an error when no events are detected 3. Improved Error Handling: - Enhanced the error messages to be more descriptive - Added better logging to help diagnose the issue in the future - Made the code more robust by handling edge cases where no events are received Implementation details: - In `vscode/src/completions/nodeClient.ts`, a check is added to ensure that `parseResult.events` has a length greater than 0 before sending events. If no events are detected, a warning is logged, but the request is not failed. - In `lib/shared/src/sourcegraph-api/completions/client.ts`, a check is added in `sendEvents` to handle cases where no events are provided. A warning is logged to the console, and the function returns early to prevent errors. These changes will prevent the "no usage data detected for completion request" error from occurring and provide better diagnostics when there are issues with the completion requests. The code now gracefully handles cases where no events are received from the server, which could happen due to network issues or upstream LLM provider outages. ## Test plan Green CI Reproduce steps ### Before 1. Ask Cody in Agent mode: `create a unified component for all the components in vscode/webviews/chat/cells/toolCell directory, or a base component that each component could build on. Use the ContextItemToolState interface for reference @message.ts` 2. Agent would try to gather all the info it needs but would exit the task with `Request Failed: no usage data detected for completion request` | Take 1 | Take 2 | |:----------------------:|:----------------------:| | ![](https://github.com/user-attachments/assets/8d868f67-6085-4cc2-8d32-dac8e51d63c0) | ![](https://github.com/user-attachments/assets/135d9169-cdae-4008-b95a-35587e2da19a) | ### After Task completed with no error <img width="740" alt="image" src="https://github.com/user-attachments/assets/f3818e84-51f7-4629-a722-da76c17254cb" /> Backport 0f3aa23 from #7585 Co-authored-by: Beatrix <[email protected]>

Cody: Reduce scroll threshold for auto scroll

5aa0eec

umpox requested review from a team July 10, 2023 16:27

abeatrix approved these changes Jul 10, 2023

View reviewed changes

umpox merged commit 7fcdb53 into main Jul 10, 2023

umpox deleted the tr/halve-scroll-threshold branch July 10, 2023 16:54

valerybugakov pushed a commit that referenced this pull request Nov 28, 2024

[Backport vscode-v1.48.x] remove last line in backport GHA (#6205)

bb81853

Backports keep giving an error about not finding the right team, testing to see if it's a lint issue

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Cody: Reduce scroll threshold for auto scroll #39

Cody: Reduce scroll threshold for auto scroll #39

umpox commented Jul 10, 2023

Cody: Reduce scroll threshold for auto scroll #39

Cody: Reduce scroll threshold for auto scroll #39

Conversation

umpox commented Jul 10, 2023

Description