The <app.graphite.dev> web app chronically display...
# report-bugs
z
The app.graphite.dev web app chronically displays the wrong number of unresolved comments in large stacks. That is, when I have a stack with upwards of 3 or 4 PRs, I often see "1/2 <Unresolved Comments>" or "2/3 <Unresolved Comments>" next to PRs in the stack but then when I click on the PR, I don't see any unresolved comments and then indeed the list view shows all resolved. I don't want to get too into the weeds, but if I leave that PR open for a few seconds longer, I notice the list view toggle back and forth every 30s or so between the correct and incorrect comment count. We use multiple AI code review bots (including Graphite's) and most of our PRs have 2+ human reviewers. It's very plausible that there could be new comments, and it's important that I follow up on all of them. These incorrect counts end up wasting a lot of my time. I'll comment with more diagnosis if interested but this is a definite bug and wastes a lot of my time.
p
(If applicable) To help us repro, please send the following if you haven't already: - PR links, screenshots, and/or debug output if applicable - Repro steps - Does this happen consistently? Just to you or others on your team also? - Is this a regression? Is the problem new? - If you're having an issue with the CLI, please send us debug logs using
gt feedback --with-debug-context <any message>
z
To me, it seems like you have two competing sync timers, one for the list of PRs, and one for the currently-open PR. If I had to guess, I'd say the sync operation for the currently-open PR is getting the correct information, and the sync operation for the PR list is getting the stale information.
Note that a hard refresh does not solve this - I guess some server-side caching is happening.
Okay from my browser network tab I can see which routes are problematic: When I load a given PR in the stack, this route loads the correct information:
<https://app.graphite.dev/api/v1/graphite/github-pr/><my org>/<my repo>/<pr number>
whose payload includes:
Copy code
"unresolvedThreadCount": 0,
"totalThreadCount": 2,
"branch": "MY BRANCH NAME",
I also see in that network tab a response from this endpoint with the stale/wrong information:
<https://app.graphite.dev/api/v1/graphite/pr-stack/><SOME LONG UID>
whose payload includes:
Copy code
"unresolvedThreadCount": 1,
"totalThreadCount": 2,
"branch": "MY BRANCH NAME",
So...yes I think my earlier diagnosis that one endpoint that populates the list of PRs in the stack and is wrong, and another endpoint loads a single PR in the stack and is correct.
There's a famous saying:
There are 2 hard problems in computer science: cache invalidation, naming things, and off-by-1 errors.
It must be some kind of cache invalidation error
p
hey Zach, apologies for the rough experience here and thank you providing all those details. I will append them in an internal tracking issue for this. To help with that could you please use the
Send feedback
option from in the Graphite app's Help menu (shortcut is
G
then
F
) . That'll capture all the info about your specific org and PRs for us to refer to the team.
z
Okay I submitted that as feedback in the graphite app @Gareth Brickman
👍 1
m
same here
this is game breaking for mke
p
hey Matthew, we're investigating this. Please also send in-app feedback for your impacted stack and we'll add to our internal tracking issue.
m
sendt feedback via CLI and sent you a loom below
z
This is mega bad @Gareth Brickman The "merge" buttons literally don't let you merge when there are unresolved comments including when graphite thinks there are unresolved comments. Forcing us to play whack-a-mole with this webapp bug to merge possibly time-sensitive changes
m
ye see my loom below
convinced my whole team to test this and now its a big issue
p
I understand, we are escalating it and will keep you updated.
hey Zach and Matthew, thanks for bearing with us and apologies for the inconvenience. The team has confirmed this is a caching issue on our end and they're investigating a long-term fix to it. In the near-term, selecting the individual PRs in the stack is expected to force the cache to update with the correct status, as Zach has indicated. Please let me know if any PR's status isn't updated or reverts to an incorrect one.
z
Hi @Gareth Brickman I'm experiencing this today 2025-11-12 FYI
p
hey Zach, affirmative. There hasn't been a permanent fix deployed for it yet so the workaround above is still necessary in these cases.
z
Bummer! Hopefully there's some progress on it.